Toggle navigation

.rst .pdf

Contents

Changelog

Note

Changelog entries for Distributed are now included in the Dask changelog.

2023.9.3

Released on September 29, 2023

Highlights

Reduce memory consumption during merge and shuffle graph optimizations

Previously there would be a large memory spike when optimizing task graphs for shuffling and merge operations (see GH#8196 for an example). This release removes that memory spike.

See GH#8197 from Patrick Hoefler for more details.

Quiet JupyterLab shutdown

Previously when running Jupyter on a scheduler (e.g. --jupyter CLI flag), an error would be raised when the notebook server was shutdown from the web application. This release ensures an error isn’t raised and the shutdown process is clean.

See GH#8220 from Thomas Grainger for details.

Additional changes

• Decompress pickled messages (GH#8216) Mads R. B. Kristensen

• Fix regression in pytest-xdist (GH#8221) crusaderky

• Hide pytest from code snippets (GH#8198) crusaderky

• Python 3.9-style multi-line with statements (GH#8211) crusaderky

• Bump actions/checkout from 4.0.0 to 4.1.0 (GH#8209)

• Update gpuCI RAPIDS_VER to 23.12 (GH#8206)

• Do not reset CUDA context after UCX tests (GH#8201) Peter Andreas Entschev

• Centralize and type no_default (GH#8171) crusaderky

• Off-by-one in the retries count in KilledWorker (GH#8203) crusaderky

• Remove deprecated aliases in distributed.utils (GH#8193) crusaderky

• Remove unspecified n_workers deprecation in wait_for_workers (GH#8192) crusaderky

• Review log-length configuration (GH#8173) crusaderky

2023.9.2

Released on September 15, 2023

Highlights

Reduce memory footprint of P2P shuffling

Significantly reduced the peak and average memory used by P2P shuffling (up to a factor of 2x reduction). This change also increases the P2P minimum supported verions of pyarrow to pyarrow=12.

See GH#8157 from Hendrik Makait for details.

Improved plugin API

Two plugin changes have been introduced to provide a more consistent and convienent plugin UX:

1. Plugins must now inherit from WorkerPlugin, SchedulerPlugin, or NannyPlugin base classes. Old-style plugins that don’t inherit from a base class will still work, but with a deprecation warning.

2. A new Client.register_plugin() method has been introduced in favor of the previous Client.register_worker_plugin() and Client.register_scheduler_plugin() methods. All plugins should now be registered using the centralized Client.register_plugin() method.

from dask.distributed import WorkerPlugin, SchedulerPlugin

class MySchedulerPlugin(SchedulerPlugin):      # Inherits from SchedulerPlugin
    def start(self, scheduler):
        print("Hello from the scheduler!")

class MyWorkerPlugin(WorkerPlugin):            # Inherits from WorkerPlugin
    def setup(self, worker):
        print(f"Hello from Worker {worker}!")

client.register_plugin(MySchedulerPlugin())    # Single method to register both types of plugins
client.register_plugin(MyWorkerPlugin())

See GH#8169 and GH#8150 from Hendrik Makait for details.

Emit deprecation warnings for configuration option renames

When a Dask configuration option that has been renamed is used, users will now get a deprecation warning pointing them to the new name.

See GH#8179 from crusaderky for details.

Additional changes

• Skip rechunker in code samples (GH#8178) Matthew Rocklin

• Ensure an error during ShuffleRun.close cannot block worker shutdown (GH#8184) Florian Jetter

• Fix race condition between MemorySampler and scheduler shutdown (GH#8172) crusaderky

• Fix post-stringification info pages (GH#8161) Florian Jetter

• Fix validation in unpack phase of P2P shuffle (GH#8160) Hendrik Makait

• Use config_for_cluster_tests in sync fixture (GH#8180) crusaderky

• Simplify boilerplate for P2P shuffles (GH#8174) Hendrik Makait

2023.9.1

Released on September 6, 2023

Enhancements

• Raise in P2P if column dtype is wrong (GH#8167) Hendrik Makait

• Auto-fail tasks with deps larger than the worker memory (GH#8135) crusaderky

• Make workers table sortable (GH#8153) Jacob Tomlinson

• Support for unsetting environment variables (GH#8144) crusaderky

Deprecations

• Deprecate asynchronous Listener.stop() (GH#8151) Hendrik Makait

Maintenance

• Initial tweaks after Dask key type changes (GH#8162) crusaderky

• Bump actions/checkout from 3.6.0 to 4.0.0 (GH#8159)

• Fix flaky test_worker_metrics (GH#8154) crusaderky

2023.9.0

Released on September 1, 2023

Enhancements

• Don’t capture functools frames in code (GH#8146) crusaderky

• Ignore asyncio when scraping code (GH#8122) Matthew Rocklin

• Remove stringification (GH#8083) Florian Jetter

Bug Fixes

• Ensure NannyPlugin are always installed (GH#8107) Florian Jetter

• Don’t use exception hooks to shorten tracebacks (GH#8127) crusaderky

• Fix P2P shuffle with LocalCluster(..., processes=False) (GH#8125) Hendrik Makait

• Merge with P2P shuffle fails if left_index or right_index is True (GH#8121) Patrick Hoefler

Documentation

• Do not advertise @span decorator (GH#8120) crusaderky

Maintenance

• Assert DF equality in P2P tests (GH#8117) Hendrik Makait

• Fix warnings from pandas=2.1.0 (GH#8145) Patrick Hoefler

• Enforce dtypes early in P2P shuffling (GH#8131) Hendrik Makait

• Bump actions/checkout from 3.5.3 to 3.6.0 (GH#8139)

• Fix flakiness in test_spans (GH#8132) crusaderky

• Remove duplicated teardown logic in Server.close, Nanny.close, and Server.stop (GH#8129) David Gold

• Fix flaky test_task_counter (GH#8134) crusaderky

• Use asyncio.timeout() in ConnectionPool where available (GH#8109) Thomas Grainger

2023.8.1

Released on August 18, 2023

New Features

• Add memray integration (GH#8044) Florian Jetter

Enhancements

• Await async listener.stop in Worker.close (GH#8118) Hendrik Makait

• Minor fixes in memray (GH#8113) Florian Jetter

• Enable basic p2p shuffle for dask-cudf (GH#7743) Richard (Rick) Zamora

• Don’t shut down unresponsive workers on gather() (GH#8101) crusaderky

• Propagate CancelledError in gather_from_workers (GH#8089) crusaderky

• Better logging for anomalous task termination (GH#8082) crusaderky

Bug Fixes

• Handle null partitions in P2P shuffling (GH#8116) Hendrik Makait

• Handle CancelledError properly in ConnectionPool (GH#8110) Florian Jetter

• Fix additional race condition that can cause P2P restart to deadlock (GH#8094) Hendrik Makait

• Ensure x-axis is uniform when plotting (GH#8093) Florian Jetter

• Fix deadlock in P2P restarts (GH#8091) Hendrik Makait

Documentation

• Add memray integration to API docs (GH#8115) James Bourbeau

• Fix default in description of LocalCluster s scheduler_port (GH#8073) Danferno

Maintenance

• Remove types_mapper arg now that it is captured in from_pyarrow_table_dispatch (GH#8114) Richard (Rick) Zamora

• Make P2P shuffle extensible (GH#8096) Hendrik Makait

• Make PreloadManager a Sequence (GH#8112) Hendrik Makait

• Introduce PreloadManager to handle failures in preload setup/teardown (GH#8078) Hendrik Makait

• Restructure P2P code (GH#8098) Hendrik Makait

• Make ToPickle a Generic (GH#8097) Hendrik Makait

• Dedicated job for memray tests (GH#8104) Florian Jetter

• Fix test_task_groups_update_start_stop, again (GH#8102) crusaderky

• Remove dumps_task (GH#8067) Florian Jetter

• Simplify usage of queues in nanny (GH#6655) Florian Jetter

• Fix flakiness in tests caused by WindowsTime (GH#8087) crusaderky

• Overhaul gather() (GH#7997) crusaderky

• Fix flaky test_asyncprocess.py::test_simple (GH#8085) crusaderky

• Skip test_client.py::test_file_descriptors_dont_leak on Mac OS (GH#8080) Hendrik Makait

• Reorder operations in Worker.close (GH#8076) Hendrik Makait

2023.8.0

Released on August 4, 2023

Enhancements

• Offload CPU intensive sections of update graph to unblock event loop (GH#8049) Florian Jetter

• Log worker close reason in events (GH#8042) Florian Jetter

• Exclude comm handshake from connect timeout (GH#7698) Florian Jetter

• Automatically restart P2P shuffles when output worker leaves (GH#7970) Hendrik Makait

• Add Client.unregister_scheduler_plugin method (GH#7968) Brian Phillips

• Fix log message (GH#8029) Hendrik Makait

• Send shards grouped by input chunk in P2P rechunking (GH#8010) Hendrik Makait

Bug Fixes

• Close state machine and add-ins first in Worker.close (GH#8066) Hendrik Makait

• Fix decide_worker picking a closing worker (GH#8032) crusaderky

• Raise CommClosedError in get_stream_address (GH#8020) jochenott

• Respect average nthreads in adaptive (GH#8041) Matthew Rocklin

• Use queued tasks in adaptive target (GH#8037) Matthew Rocklin

• Restore support for yield unsafe Client context managers and deprecate that support (GH#7987) Thomas Grainger

Documentation

• Change worker_saturation default value to 1.1 in the documention (GH#8040) minhnguyenxuan60

• Clarified concurrent.futures section in client.rst (GH#8048) mercyo12

Maintenance

• Fix flaky test_worker_metrics (GH#8069) crusaderky

• Use SPDX in license metadata (GH#8065) jakirkham

• Rebalance ci1 markers (GH#8061) Florian Jetter

• Ensure stream messages are always ordered (GH#8059) Florian Jetter

• Simplify update graph (GH#8047) Florian Jetter

• Provide close reason when signal is caught (GH#8045) Florian Jetter

• Allow unclosed comms in tests (GH#8057) Florian Jetter

• Cosmetic tweak to adaptive_target (GH#8052) crusaderky

• Fix linting (GH#8046) Florian Jetter

• Update gpuCI RAPIDS_VER to 23.10 (GH#8033)

• Test against more recent pyarrow versions (GH#8021) James Bourbeau

• Add a test for GraphLayout with scatter (GH#8025) Irina Truong

• Fix compatibility variable naming (GH#8030) Hendrik Makait

2023.7.1

Released on July 20, 2023

Enhancements

• gather_dep should handle CancelledError (GH#8013) crusaderky

• Pass stimulus_id to SchedulerPlugin.remove_worker and SchedulerPlugin.transition (GH#7974) Hendrik Makait

• Log stimulus_id in retire_worker (GH#8003) crusaderky

• Use BufferOutputStream in P2P (GH#7991) Florian Jetter

• Add Coiled to ignored modules for code sniffing (GH#7986) Matthew Rocklin

• Progress bar can group tasks by span (GH#7952) Irina Truong

• Improved error messages for P2P shuffling (GH#7979) Hendrik Makait

• Reduce removing comms log to debug level (GH#7972) Florian Jetter

Bug Fixes

• Fix for TypeError: '<' not supported in graph dashboard (GH#8017) Irina Truong

• Fix shuffle code to work with pyarrow 13 (GH#8009) Joris Van den Bossche

Documentation

• Add some top-level exposition to the p2p rechunking code (GH#7978) Lawrence Mitchell

Maintenance

• Add test when not repartitioning for p2p in set_index (GH#8016) Patrick Hoefler

• Bump JamesIves/github-pages-deploy-action from 4.4.2 to 4.4.3 (GH#8008)

• Configure asyncio loop using loop_factory kwarg rather than using the set_event_loop_policy (GH#7969) Thomas Grainger

• Fix P2P worker cleanup (GH#7981) Hendrik Makait

• Skip click v8.1.4 in mypy pre-commit hook (GH#7989) Thomas Grainger

• Remove accidental duplicated conversion of pyarrow Table to pandas (GH#7983) Joris Van den Bossche

2023.7.0

Released on July 7, 2023

Enhancements

• Propagate spans to tasks (GH#7898) crusaderky

• Make Fine Performance Metrics bar graph horizontal (GH#7966) crusaderky

• Don’t pile up context_meter callbacks (GH#7961) crusaderky

• Polish Fine Performance Metrics plot (GH#7963) crusaderky

• Sign task-erred with run_id and reject outdated responses (GH#7933) Hendrik Makait

• Set Client.as_current when entering ctx (GH#6527) Florian Jetter

• Re-run erred task on ComputeTaskEvent (GH#7967) Hendrik Makait

Bug Fixes

• Fix crash in spans when time() is not monotonic (GH#7960) crusaderky

Documentation

• Documentation for Fine Performance Metrics and Spans (GH#7945) crusaderky

• Update client.py to be consistent with the docstring (GH#7705) Sultan Orazbayev

Maintenance

• Use distributed.wait_for in test_close_async_task_handles_cancellation (GH#7955) Thomas Grainger

• Fix flaky UCX tests (GH#7950) Peter Andreas Entschev

2023.6.1

Released on June 26, 2023

Enhancements

• Add idle time to fine performance metrics (GH#7938) crusaderky

• Spans: capture code snippets (GH#7930) crusaderky

• Improve memory footprint of P2P rechunking (GH#7897) Hendrik Makait

• Improve error message on invalid state in _handle_remove_replicas (GH#7920) Hendrik Makait

• Make ShuffleSchedulerExtension.remove_worker more robust (GH#7921) Hendrik Makait

• Provide more information if occupancy drops below zero (GH#7924) Hendrik Makait

• Improved conversion between pyarrow and pandas in P2P shuffling (GH#7896) Hendrik Makait

Bug Fixes

• Add Cluster.called_from_running_loop and fix Cluster.asynchronous (GH#7941) Jacob Tomlinson

• Fix annotations and spans leaking between threads (GH#7935) Irina Truong

• Handle null partitions in P2P shuffling (GH#7922) Jonathan De Troye

• Fix race condition in Fine Performance Metrics sync (GH#7927) crusaderky

• Avoid (GH#7923) by starting run_id at 1 (GH#7925) Hendrik Makait

• Fix glitches in Fine Performance Metrics stacked graph (GH#7919) crusaderky

Maintenance

• Wipe the cache after (GH#7935) (GH#7946) crusaderky

• Remove grace period for unclosed comms in gen_cluster (GH#7937) Thomas Grainger

• raise pytest.skip is redundant (GH#7939) crusaderky

• Fix test_rechunk_with_{fully|partially}_unknown_dimension on CI (GH#7934) Hendrik Makait

• Fix compatibility with numpy 1.25 (GH#7932) crusaderky

• Spans: refactor sums of mappings (GH#7918) crusaderky

• Fix flaky test_send_metrics_to_scheduler (GH#7931) crusaderky

• Avoid calls to make_current() and make_clear() by using asyncio.run in LoopRunner (GH#7467) Thomas Grainger

• Add needs triage label to re/opened PRs and issues (GH#7916) Miles

• Remove span_id from global metrics on scheduler (GH#7917) crusaderky

• Add spans to Fine Performance Metrics bokeh dashboard (GH#7911) crusaderky

• FinePerformanceMetrics dashboard overhaul (GH#7910) crusaderky

• Check for skip-caching label (GH#7907) Miles

• Fix CI changes from (GH#7902) (GH#7905) Hendrik Makait

• Rename get_default_shuffle_algorithm to get_default_shuffle_method (GH#7902) Hendrik Makait

• Bump actions/checkout from 3.5.2 to 3.5.3 (GH#7904)

• Refactor P2P rechunk validation (GH#7890) Hendrik Makait

2023.6.0

Released on June 9, 2023

Enhancements

• Post fine performance metrics to spans (GH#7885) crusaderky

• Unique Spans (GH#7882) crusaderky

• Add a timeout to client.as_completed that mirrors concurrent.futures.as_completed timeout (GH#7811) Thomas Grainger

• Enforce dtypes in P2P shuffle (GH#7879) Hendrik Makait

• Support load= keyword for Client.upload_file (GH#7873) James Bourbeau

• Support get_worker() and worker_client() in async tasks (GH#7844) Thomas Grainger

• Capture line number for code frames (GH#7786) Miles

Bug Fixes

• Avoid meta roundtrip in P2P shuffle (GH#7895) Hendrik Makait

• Fix Fine Performance Metrics mis-aligned ColumnData lengths (GH#7893) Miles

• Fix Fine Performance Metrics spilling crash (GH#7878) Miles

• Fix spans bug when scatter or client_desires_new_key creates a task (GH#7886) crusaderky

• Fix Fine Performance Metrics w/ Bokeh 3 (GH#7874) Miles

• TaskGroup.start can move backwards (GH#7867) crusaderky

• Use properly imported MatDescriptor for cupy dispatch registration (GH#7868) Charles Blackmon-Luca

• Ensure retire_workers works if AMM extension hasn’t been loaded (GH#7863) crusaderky

Maintenance

• Review user-defined fine performance metrics (GH#7894) crusaderky

• Fix tests that disable the shuffle extension (GH#7883) crusaderky

• Refactor Scheduler.is_idle (GH#7881) crusaderky

• Link TaskGroups to Spans (GH#7869) crusaderky

• Spans skeleton (GH#7862) crusaderky

• Update gpuCI RAPIDS_VER to 23.08 (GH#7855)

• Bump JamesIves/github-pages-deploy-action from 4.4.1 to 4.4.2 (GH#7865)

2023.5.1

Released on May 26, 2023

Note

This release drops support for Python 3.8. As of this release Dask supports Python 3.9, 3.10, and 3.11. See this community issue for more details.

Enhancements

• Exclude IPython code from computations (GH#7788) Miles

• Drop Python 3.8 support (GH#7840) Thomas Grainger

• Add storage_options to performance_report (GH#7636) ypogorelova

• Don’t warn about mismatched msgpack (GH#7839) Irina Truong

• Clean up sys.path on Server shutdown (GH#7838) James Bourbeau

• Dashboard: Fine Performance Metrics (GH#7725) Miles

Bug Fixes

• Properly handle unknown chunk sizes in P2P rechunking (GH#7856) Hendrik Makait

• Minimal change to work around (GH#7726) / support for UCX (GH#7851) Benjamin Zaitlen

• Don’t end computations until cluster is truly idle (GH#7790) crusaderky

Maintenance

• Explicitly install anaconda-client from conda-forge when uploading conda nightlies (GH#7861) Charles Blackmon-Luca

• Fix is_idle docs build (GH#7854) James Bourbeau

• Add tests for P2P barrier fusion (GH#7845) Hendrik Makait

• Avoid DeprecationWarning in cupy dispatch registration (GH#7836) Lawrence Mitchell

2023.5.0

Released on May 12, 2023

Enhancements

• Client.upload_file send to both Workers and Scheduler and rename scratch directory (GH#7802) Miles

• Allow dashboard to be used with bokeh prereleases (GH#7814) James Bourbeau

Bug Fixes

• Ensure log_event of non-msgpack serializable object do not kill servers (GH#7472) Florian Jetter

• Fix test_nanny.py duplicated pytestmark definitions (GH#7819) Thomas Grainger

• Fix flaky test_dask_worker.py::test_single_executable_deprecated (GH#7817) Thomas Grainger

Maintenance

• Annotation-less P2P shuffling (GH#7801) Hendrik Makait

• Fix docstring for batch_size in client.map (GH#7833) David Chudzicki

• Refactor test_protocol.py (GH#7829) crusaderky

• Lint #6496 (GH#7828) crusaderky

• Remove hardcoded 60s timeout (GH#6496) Florian Jetter

• Add __init__.py files to template and static directories (GH#7809) Thomas Grainger

• Disable compression for fast comms (GH#7768) crusaderky

• Avoid deprecated pd.api.types.is_sparse (GH#7813) James Bourbeau

• Bump gpuCI PYTHON_VER from 3.8 to 3.9 (GH#7812) Charles Blackmon-Luca

2023.4.1

Released on April 28, 2023

Enhancements

• Enable GIL monitoring when gilknocker installed (GH#7730) Miles

• By default only set logging handler if no other handler has been set to avoid double logging (GH#7750) Thomas Grainger

• Cluster wait (GH#6700) Iain Dorrington

• Add Prometheus counter for SystemMonitor.last_time (GH#7785) Miles

Bug Fixes

• Partial revert defaultclient config setting (GH#7803) Florian Jetter

• Delay awaiting async SchedulerPlugin.{add|remove}_worker hooks in order to immediately execute all sync ones (GH#7799) Hendrik Makait

• Fix check_idle not returning the correct value if no change to idleness (GH#7781) Jacob Tomlinson

Maintenance

• Avoid warning when gilknocker not installed (GH#7808) James Bourbeau

• Only set worker/nanny to Status.running if it is in Status.init (GH#7773) Thomas Grainger

• Add --cov-config=pyproject.toml so config is always correctly loaded by pytest-cov (GH#7793) Thomas Grainger

• gilknocker from conda-forge (GH#7791) James Bourbeau

• Minor zict cleanup (GH#7783) crusaderky

• Bump actions/checkout from 3.5.0 to 3.5.2 (GH#7784)

• Fix typing now that code is tuple of frame(s) (GH#7778) Nat Tabris

2023.4.0

Released on April 14, 2023

Note

With this release we are making a change which will require the Dask scheduler to have consistent software and hardware capabilities as the client and workers.

It’s always been recommended that your client and workers have a consistent software and hardware environment so that data structures and dependencies can be pickled and passed between them. However recent changes to the Dask scheduler mean that we now also require your scheduler to have the same consistent environment as everything else.

Enhancements

• Meter queue time to the offload executor (GH#7758) crusaderky

• Add GIL contention metric to Prometheus (GH#7651) Miles

• Add methods Client.forward_logging() and Client.unforward_logging(). (GH#7276) Max Bane

• Optionally capture more frames in computations (GH#7656) Gabe Joseph

• Consider Jupyter activity in idle timeout (GH#7687) Gabe Joseph

• Add a dashboard component that displays RMM memory (GH#7718) Peter Andreas Entschev

• Improve error message if shuffle/rechunk lost annotations (GH#7707) Hendrik Makait

• Exception chaining in P2P shuffling (GH#7706) Hendrik Makait

• Use pickle for graph submissions from client to scheduler (GH#7564) Florian Jetter

Bug Fixes

• Fix crash on missing env var in dashboard link formatting (GH#7729) Miles

• Fix randbytes() on Python 3.8 (GH#7771) crusaderky

• Run scheduler of SubprocessCluster in subprocess (GH#7727) Hendrik Makait

• Drop id from RMM dashboard component (GH#7739) James Bourbeau

Maintenance

• Bump peter-evans/create-pull-request from 4 to 5 (GH#7766)

• Fix flaky test_malloc_trim_threshold in CI (GH#7764) crusaderky

• Minor polish in spill and worker_memory_manager (GH#7752) crusaderky

• Merge identical tool.mypy.overrides sections (GH#7749) Thomas Grainger

• Add changelog section for 2023.3.2.1 (GH#7755) Charles Blackmon-Luca

• Specify ts resolution explicitly in test_processing_chain (GH#7744) Patrick Hoefler

• Unignore Sphinx ref.python (GH#7713) Thomas Grainger

• Temporary fix for test_merge_by_multiple_columns with pandas 2.0 (GH#7747) James Bourbeau

• Remove dask/gpu from gpuCI update reviewers (GH#7741) Charles Blackmon-Luca

• Update gpuCI RAPIDS_VER to 23.06 (GH#7728)

• Remove test for DataFrame.to_hdf (GH#7735) Hendrik Makait

• Test P2P shuffling with DataFrame.to_hdf (GH#7720) Hendrik Makait

• scheduler.py typing - remove allow_incomplete_defs (GH#7721) Florian Jetter

• Remove bokeh upper bound (GH#7413) James Bourbeau

• Use declarative setuptools (GH#7629) Thomas Grainger

• Store performance metrics on scheduler (GH#7701) Miles

• Upgrade readthedocs config to ubuntu 22.04 and Python 3.11 (GH#7722) Thomas Grainger

• Clean up legacy cruft from worker reconnection (GH#7712) crusaderky

• Bump actions/checkout from 3.4.0 to 3.5.0 (GH#7711)

• Drop support for zict 2.1.0 (GH#7709) crusaderky

• Fix mypy warning in test_client.py (GH#7710) crusaderky

• Test P2P shuffling with DataFrame.categorize (GH#7708) Hendrik Makait

2023.3.2.1

Released on April 5, 2023

Bug Fixes

• Register atexit handler before Distributed handlers to unblock hanging UCX clusters Lawrence Mitchell Ben Zaitlen

2023.3.2

Released on March 24, 2023

Enhancements

• Enhanced thread-safety in zict.File (GH#7691) crusaderky

• Future deserialization without available client (GH#7580) Florian Jetter

• Support adjusting GIL monitoring interval (GH#7650) Miles

• Gracefully stop GIL monitoring if running (GH#7652) Miles

• Fine performance metrics for execute, gather_dep, etc. (GH#7586) crusaderky

• Add GIL metric to dashboard (GH#7646) Miles

• Expose scheduler idle via RPC and HTTP API (GH#7642) Jacob Tomlinson

• Add full dashboard link in scheduler logs (GH#7631) Miles

Bug Fixes

• Tell workers when their peers have left (so they don’t hang fetching data from them) (GH#7574) Thomas Grainger

• Fix regression in dashboard after (GH#7586) (GH#7683) crusaderky

• Fix OverflowError in Cluster._sync_cluster_info() (GH#7648) Hendrik Makait

• Ensure that serialized data is measured correctly (GH#7593) Florian Jetter

Documentation

• Fix unexpected indentation in Client.cancel docstring (GH#7694) Thomas Grainger

• Improve plugin API documentation (GH#7653) Florian Jetter

Maintenance

• Configure sphinx warnings as errors (GH#7697) Thomas Grainger

• Fix naming comparison in test-report workflow script (GH#7695) Miles

• Temporarily restrict ipywidgets<8.0.5 (GH#7693) crusaderky

• Bump actions/checkout from 3.3.0 to 3.4.0 (GH#7685)

• Temporarily restrict ipykernel<6.22.0 (GH#7689) James Bourbeau

• Fix typo in CODEOWNERS (GH#7670) Hendrik Makait

• Avoid bool object has no attribute close in @gen_cluster (GH#7657) Thomas Grainger

• Fix failing test_server_close_stops_gil_monitoring (GH#7659) James Bourbeau

• Add CODEOWNERS file (GH#7645) Jacob Tomlinson

• Remove weakref finalizer for Offload Executor (GH#7644) Florian Jetter

2023.3.1

Released on March 10, 2023

Enhancements

• Add Jupyter link to dashboard menu if --jupyter flag is set (GH#7638) Jacob Tomlinson

• Bump minimum click version from 7.0 to 8.0 (GH#7637) Miles

• Extend dask metapackage dependencies (GH#7630) James Bourbeau

• Further improvements to Client.restart_workers (GH#7620) Miles

• P2P offload get_output_partition (GH#7587) Florian Jetter

• Initial integration of GIL contention metric (GH#7624) Miles

• Add dashboard documentation links (GH#7610) Miles

• Rename shuffle/rechunk config option/kwarg to method (GH#7623) Hendrik Makait

• Return results in restart_workers (GH#7606) Miles

• Ensure client key cancellation uses ordered messages (GH#7583) Florian Jetter

Bug Fixes

• Fix undefined async_wait_for -> async_poll_for (GH#7627) Miles

• Don’t send client heartbeat without a scheduler_comm (GH#7612) James Bourbeau

• Do not unspill on free-keys (GH#7607) crusaderky

Documentation

• Add notes to Client.submit, Client.map, and Client.scatter with the description of the current task graph resolution algorithm limitations (GH#7588) Eugene Druzhynin

Maintenance

• Use range with pickle protocol versions (GH#7635) jakirkham

• Share thread pool among P2P shuffle runs (GH#7621) Hendrik Makait

• Replace psutil suspend with BlockedGatherDep in test_failing_worker_with_additional_replicas_on_cluster (GH#7633) Thomas Grainger

• Ignore pkg_resources DeprecationWarning for mindeps (GH#7626) Miles

• Implement wait_for using asyncio.timeout() on 3.11 (GH#7571) Thomas Grainger

• Use tmp_path fixture instead of outdated tmpdir fixture (GH#7582) ypogorelova

• Only one crick callback (GH#7614) crusaderky

• Add mindeps + numpy job to tests CI (GH#7609) Miles

• Do not xfail whole tests due to (GH#6705) (GH#7611) crusaderky

2023.3.0

Released on March 1, 2023

Bug Fixes

• Remove pyarrow dependency for rechunking (GH#7604) Florian Jetter

• Update rechunk_transfer and rechunk_unpack errors (GH#7600) James Bourbeau

Maintenance

• Remove dead code and document arguments to ShardBuffer constructors (GH#7590) Lawrence Mitchell

• Fix tests for p2p by default (GH#7595) Florian Jetter

• Remove obsolete cast (GH#7596) Florian Jetter

2023.2.1

Released on February 24, 2023

Enhancements

• P2P for array rechunking (GH#7534) Hendrik Makait

• P2P HashJoin (GH#7514) Florian Jetter

• Unpickle Events, Variables, Queues and Semaphore safely without Client context (GH#7579) Florian Jetter

• Allow pickle to fall back to dask_serialize (GH#7567) Florian Jetter

• make ConnectionPool.remove cancel connection attempts (GH#7547) Thomas Grainger

• Meter how long each task prefix stays in each state (GH#7560) crusaderky

Bug Fixes

• Avoid parsing sys.argv when starting jupyter server (GH#7573) Brett Naul

• str/bytes compatibility for PyNVML device name (GH#7563) James Bourbeau

• metrics.monotonic() is not monotonic on Windows (GH#7558) crusaderky

• Fix for bytes/str discrepancy after PyNVML update (GH#7544) Peter Andreas Entschev

Maintenance

• Raise when attempting P2P with active fuse optimization (GH#7585) Hendrik Makait

• Fix test_shuffling (GH#7581) Hendrik Makait

• P2P: raise RuntimeError if pyarrow version is not sufficient (GH#7578) Florian Jetter

• Check for dtype support in p2p (GH#7425) Hendrik Makait

• Update parsing of FULL_RAPIDS_VER/FULL_UCX_PY_VER (GH#7568) Charles Blackmon-Luca

• move retry from get_data_from_worker to gather_from_workers (GH#7546) Thomas Grainger

• Increase numpy and pandas version pins for nightlies (GH#7562) James Bourbeau

• Set validate=True in all tests (GH#7557) crusaderky

• Remove dead code from _get_task_finished_msg (GH#7561) crusaderky

• Mark tests that take >2s as slow (GH#7556) crusaderky

• Fix test_scatter_no_workers on slow CI (GH#7559) crusaderky

• Unskip test_delete_some_results (GH#7508) Hendrik Makait

• scatter() should not sidestep the worker transition machinery (GH#7545) crusaderky

• pre-commit bump (GH#7541) crusaderky

• Better assertions in Worker.validate_state() (GH#7549) crusaderky

• Bump jacobtomlinson/gha-find-replace from 2 to 3 (GH#7540) James Bourbeau

• Bump black to 23.1.0 (GH#7542) crusaderky

• Run GPU tests on python 3.8 & 3.10 (GH#7537) Charles Blackmon-Luca

2023.2.0

Released on February 10, 2023

Enhancements

• Rate limit the worker memory logs (GH#7529) Florian Jetter

• Move P2P barrier logic to scheduler extension (GH#7519) Hendrik Makait

Maintenance

• Use PEP 673 Self type (GH#7530) Thomas Grainger

• Tentatively fix test_pause_while_spilling (GH#7517) crusaderky

• Annotate asyncio_tcp.py (GH#7522) crusaderky

• Use dask git tip for mypy (GH#7516) crusaderky

• Upgrade to mypy v1 (GH#7525) Thomas Grainger

• Clean up calls to captured_logger (GH#7521) crusaderky

• Update isort version to 5.12.0 (GH#7513) Lawrence Mitchell

2023.1.1

Released on January 27, 2023

Enhancements

• P2P shuffle deduplicates data and can be run several times (GH#7486) Hendrik Makait

• Reverse order of get_logs() and get_worker_logs() (GH#7475) Nicholas R. Knezek

• Add prometheus metric for time and memory used per task prefix (GH#7406) Thomas Grainger

• Additive worker counts in Prometheus (GH#7468) crusaderky

• Add help tool for taskstream (GH#7478) Florian Jetter

Bug Fixes

• Do not allow for a worker to reject a drop replica request (GH#7490) Hendrik Makait

• Fix un/packing for namedtuples with custom constructors (GH#7465) antonymayi

Documentation

• Remove timeout= from docstring example for worker_client (GH#7497) Florian Jetter

Maintenance

• Ignore get_default_shuffle_algorithm linting issue (GH#7506) Florian Jetter

• Remove set_config when using default client (GH#7482) Florian Jetter

• Update gpuCI RAPIDS_VER to 23.04 (GH#7501)

• Fix test_balance_expensive_tasks and improve helper functions in test_steal.py (GH#7253) Hendrik Makait

• Sign every compute task with run ID to correlate response (GH#7463) Hendrik Makait

2023.1.0

Released on January 13, 2023

New Features

• Add local SubprocessCluster that runs workers in separate processes (GH#7431) Hendrik Makait

Enhancements

• Ensure client session is quiet after cluster.close() or client.shutdown() (GH#7429) James Bourbeau

• Set lifetime-stagger default value to None (GH#7445) bstadlbauer

• Memory thresholds should never be exactly 0.0 (GH#7458) Stuart Berg

• Remove the Incorrect-Sizeof-Warning (GH#7450) Mads R. B. Kristensen

• Log exceptions in P2P shuffle tasks (GH#7442) Hendrik Makait

• Add support for packing namedtuple and add test for future resolution in submit (GH#7292) Andrew

• Avoid deep copy on lz4 decompression (GH#7437) crusaderky

• Avoid deep copy of numpy buffers on unspill (GH#7435) crusaderky

• Don’t error when clicking on empty task stream plot (GH#7432) James Bourbeau

• Do not count spilled memory when comparing vs. process memory (GH#7430) crusaderky

• Stop Client periodic callbacks during shutdown() (GH#7428) James Bourbeau

• Add dask spec CLI (GH#7427) Matthew Rocklin

• Create new zstd (de)compressor for each compression call (GH#7404) Dylan Wragge

• Rename managed_in_memory etc. to match GUI (GH#7418) crusaderky

• Warn users when sizeof() returns inflated output (GH#7419) crusaderky

Bug Fixes

• Ensure dicts are properly recognized as msgpack serializable (GH#7473) Florian Jetter

• Reset state of ShuffleSchedulerExtension on restart (GH#7446) Hendrik Makait

• Reject non-string column names in P2P shuffle (GH#7447) Hendrik Makait

• Avoid int32 in dashboard (GH#7443) Matthew Rocklin

• Fix P2PShuffle serialization for categorical data (GH#7410) Hendrik Makait

• WorkerPorcess blocks on kill if still starting (GH#7424) Matthew Rocklin

Documentation

• Move Prometheus docs from dask/dask (GH#7405) crusaderky

Maintenance

• Various cleanups in semaphore (GH#5885) Florian Jetter

• test_rlimit fails on MacOSX (GH#7457) crusaderky

• Bump actions/checkout from 3.2.0 to 3.3.0 (GH#7464)

• Remove conditional imports of psutil (GH#7462) crusaderky

• Drop support for zict < 2.1.0 (GH#7456) crusaderky

• Fix flaky test_digests (GH#7454) crusaderky

• Add minimum dependency testing to CI (GH#7285) Charles Blackmon-Luca

• Avoid overflow in statitics.mean (GH#7426) Matthew Rocklin

• Ignore numpy bool8 deprecation (GH#7423) Matthew Rocklin

• Add missing skips for pyarrow (GH#7416) Elliott Sales de Andrade

• Be more permissive about expected ciphers in tests (GH#7417) Elliott Sales de Andrade

• Revert “TST: Fetch executables from build root (GH#2551)” (GH#7415) Elliott Sales de Andrade

2022.12.1

Released on December 16, 2022

Enhancements

• SpillBuffer metrics (GH#7368) crusaderky

• Prometheus: measure how much spilling blocks the event loop (GH#7370) crusaderky

• Add transfer_outgoing_bytes_total metric (GH#7388) Gabe Joseph

• Fail P2PShuffle gracefully upon worker failure (GH#7326) Hendrik Makait

Bug Fixes

• Select queued tasks in stimuli, not transitions (GH#7402) Gabe Joseph

• Check ContextVar in default_client (GH#7369) Matthew Rocklin

• Fix sending event messages to non-subscribers (GH#7014) Laurence Watts

• Set sizing mode on Tabs to avoid layout collapse (GH#7365) Mateusz Paprocki

Maintenance

• Restructure P2PShuffle extensions (GH#7390) Hendrik Makait

• Add Python 3.11 classifier (GH#7408) James Bourbeau

• Add support for Python 3.11 (GH#7249) Thomas Grainger

• Add test for using annotations with client.submit and client.map (GH#7399) James Bourbeau

• Bump actions/checkout from 3.1.0 to 3.2.0 (GH#7393)

• Remove superfluous ShuffleSchedulerExtension.barriers (GH#7389) Hendrik Makait

• Remove ignore annotation-unchecked (GH#7379) crusaderky

• Remove tornado max version from nightly recipe (GH#7376) Charles Blackmon-Luca

• Remove the experimental feature warning for Semaphore (GH#7373) Florian Jetter

2022.12.0

Released on December 2, 2022

Enhancements

• Expose event loop health metrics in Prometheus (GH#7360) Hendrik Makait

• Allow log propagation by default (GH#5669) Florian Jetter

• Clean up of unpack_remotedata() (GH#7322) Mads R. B. Kristensen

• Upgrade to tornado 6.2 (GH#7286) Thomas Grainger

• Introduce Server level comm counters (GH#7332) Florian Jetter

• Prometheus debug log (GH#7302) Florian Jetter

Bug Fixes

• Catch BaseException s from user tasks (GH#5997) Gabe Joseph

• Impossible use case of erred deps in transition to waiting (GH#7354) crusaderky

• Fix a deadlock when queued tasks are resubmitted quickly in succession (GH#7348) Florian Jetter

Documentation

• Editorial changes to Prometheus documentation (GH#7350) Hendrik Makait

Maintenance

• Fetch all artifacts (GH#7355) Enrico Minack

• Delay fsspec and urllib3 import time (GH#6659) Florian Jetter

• Bump mypy (GH#7349) crusaderky

• Bump mypy and remove win specific run (GH#7344) Florian Jetter

• Finish overhaul of SchedulerState annotations (GH#7333) crusaderky

• Fix flaky test_pause_while_spilling (GH#7334) Gabe Joseph

• Update gpuCI RAPIDS_VER to 23.02 (GH#7337)

2022.11.1

Released on November 18, 2022

Enhancements

• Restrict bokeh=3 support (GH#7329) Gabe Joseph

• Respect death timeout when waiting for scheduler file (GH#7296) Florian Jetter

• Always raise exception if P2PShuffle s send fails (GH#7317) Hendrik Makait

Documentation

• Fix typo in client.run() docstring (GH#7315) Richard Pelgrim

• Note queuing default change in changelog (GH#7314) Gabe Joseph

Maintenance

• Update ga-yaml-parser step in gpuCI updating workflow (GH#7335) Charles Blackmon-Luca

• Remove exception handling from transitions (GH#7316) crusaderky

• Turn private functions into private SchedulerState methods (GH#7260) Hendrik Makait

• Bump toolz minimum version to 0.10.0 (GH#7309) Sam Grayson

2022.11.0

Released on November 15, 2022

Note

This release changes the default scheduling mode to use queuing. This will significantly reduce cluster memory use in most cases, and generally improve stability and performance. Learn more here and please provide any feedback on this discussion.

In rare cases, this could make some workloads slower. See the documentation for more information, and how to switch back to the old mode.

New Features

• Add ForwardOutput worker plugin to forward stdout and stderr to client. (GH#7297) Hendrik Makait

• Duration counters on prefix level (GH#7288) Florian Jetter

• Include button for launching JupyterLab layout in repr (GH#7218) Ian Rose

Enhancements

• Support MIG parsing during CUDA context creation in UCX initialization (GH#6720) Peter Andreas Entschev

• Handle /metrics endpoint without prometheus-client installed (GH#7234) Hendrik Makait

• Add support for unpacking namedtuples in remote data (GH#7282) Andrew

• Enable queuing by default (GH#7279) Florian Jetter

• Fix exists->``exist`` typo in scheduler error messages (GH#7281) Matthew Plough

• If there’s an exception in the Client async context manager body then close fast (GH#6920) Thomas Grainger

Documentation

• Copyediting typos + codespell pre-commit hook for docs (GH#7294) Matthew Plough

• Queuing docs (GH#7203) Gabe Joseph

Bug Fixes

• Ensure category is optional when logging "warn" events (GH#7169) James Bourbeau

• Edge and impossible transitions to memory (GH#7205) crusaderky

Maintenance

• Use conda-incubator/setup-miniconda@v2.2.0 (GH#7310) jakirkham

• Allow bokeh=3 (GH#5648) James Bourbeau

• Fix typos in P2P shuffle code (GH#7304) Hendrik Makait

• Reenable test_bad_disk (GH#7300) Florian Jetter

• Reduce max-runs in test reports (GH#7299) Florian Jetter

• Revert idle classification when worker-saturation is set (GH#7278) Florian Jetter

• Fix flaky deadline_expiration (GH#7287) Florian Jetter

• Rewrite of P2P control flow (GH#7268) Florian Jetter

• Add codecov token (GH#7277) Florian Jetter

• Bump minimum bokeh version to 2.4.2 (GH#7271) James Bourbeau

• Remove deprecated code calls to IOLoop.make_current() (GH#7240) Thomas Grainger

• Improved test for balancing expensive tasks (GH#7272) Hendrik Makait

• Refactor semaphore._Watch into general-purpose Deadline utility (GH#7238) Hendrik Makait

• Coverage report fixing (GH#7270) Tom Hu

• Require Click 7.0+ (GH#7226) jakirkham

• Drop tests (GH#7269) Hendrik Makait

• Replace test_(do_not_)steal_communication_heavy_tasks tests with more robust versions (GH#7243) Hendrik Makait

• xfail test_bad_disk (GH#7265) crusaderky

• Move transition_log from Scheduler to SchedulerState (GH#7254) crusaderky

• Remove Scheduler.log (GH#7258) crusaderky

• Use latest pickle (GH#5826) jakirkham

• Polish parsing of worker-saturation from config (GH#7255) crusaderky

• Avoid expensive occupancy calculation when unused (GH#7257) Gabe Joseph

• Un-skip test_nested_compute (GH#7247) Gabe Joseph

• Review test_do_not_steal_communication_heavy_tasks (GH#7250) crusaderky

• Fix test_stress_creation_and_deletion (GH#7215) crusaderky

• Raise exceptions in Server.handle_stream instead of swallowing/logging (GH#7162) Hendrik Makait

• Upgrade to mypy v0.982 (GH#7241) Thomas Grainger

• Fix _update_scheduler_info hanging failed tests (GH#7225) Gabe Joseph

• Bump xarray-contrib/ci-trigger from 1.1 to 1.2 (GH#7232)

2022.10.2

Released on October 31, 2022

• Reverted a bug where Bokeh was accidentally made non-optional (GH#7230) Oliver Holworthy

• Schedule a queued task when a task secedes (GH#7224) Gabe Joseph

This was a hotfix release

2022.10.1

Released on October 28, 2022

New Features

• Add Client.restart_workers method (GH#7154) James Bourbeau

• Implement PackageInstall plugin for pip and conda (GH#7126) Hendrik Makait

Enhancements

• Add prometheus collector for work-stealing (GH#7206) Hendrik Makait

• Track reason of workers closing and restarting (GH#7166) Hendrik Makait

• Show no-worker on task progress bar (GH#7171) Florian Jetter

• Set OPENBLAS_NUM_THREADS by default (GH#7177) James Bourbeau

• Optionally provide local directory to data constructor (GH#7153) Lawrence Mitchell

• Introduce distributed.comm.ucx.environment config slot (GH#7164) Lawrence Mitchell

• Log information about memory limit (GH#7160) Florian Jetter

• Improve log messages on scheduler for restart (GH#7150) Florian Jetter

• More comprehensive WorkerState task counters (GH#7167) crusaderky

Documentation

• Add note to changelog about new CLI (GH#7178) James Bourbeau

• Update AMM docs (GH#7158) Benjamin Zaitlen

• Add CondaInstall to plugins doc (GH#7149) James Bourbeau

Bug Fixes

• Update minimum bokeh version message (GH#7172) James Bourbeau

• Revamped implementations of remote print() and warn(), fixing #7095 (GH#7129) Max Bane

Maintenance

• Temporarily restrict bokeh<3 (GH#7219) James Bourbeau

• Make Scheduler.reschedule private (GH#7216) crusaderky

• Fix decide_worker_rootish_queuing_disabled assert (GH#7065) Gabe Joseph

• Fix flaky test_include_communication_in_occupancy (GH#7212) Gabe Joseph

• Do not raise on leaked websockets (GH#7199) Florian Jetter

• Update nightly recipes with CLI tests, dependency changes (GH#7201) Charles Blackmon-Luca

• Make p2p shuffle submodules private (GH#7186) Florian Jetter

• Backport tornado PeriodicCallback (GH#7165) Florian Jetter

• Fix mypy failure on CI (GH#7198) Florian Jetter

• User a layer for p2p shuffle (GH#7180) Florian Jetter

• Type annotations for shuffle (GH#7185) Florian Jetter

• Do not close worker on comm error in heartbeat (GH#7163) Hendrik Makait

• Errors when setting TCP timeouts log as error (GH#7161) Florian Jetter

• Remove incorrect advice from pre-commit config (GH#7159) crusaderky

• Bump the-coding-turtle/ga-yaml-parser from 0.1.1 to 0.1.2 (GH#7146)

• Bump JamesIves/github-pages-deploy-action from 4.1.7 to 4.4.1 (GH#7145)

• Use functionalities network for codecov uploader (GH#7148) Florian Jetter

• Use counter metric type where appropriate, incoming_count was reporting bytes (GH#7125) Nat Tabris

2022.10.0

Released on October 14, 2022

Note

This release deprecates dask-scheduler, dask-worker, and dask-ssh CLIs in favor of dask scheduler, dask worker, and dask ssh, respectively. The old-style CLIs will continue to work for a time, but will be removed in a future release.

As part of this migration the --reconnect, --nprocs, --bokeh, --bokeh-port CLI options have also been removed for both the old- and new-style CLIs. These options had already previously been deprecated.

Enhancements

• Use of new dask CLI (GH#6735) Doug Davis

• Refactor occupancy (GH#7075) Hendrik Makait

• Expose managed/unmanaged/spilled memory to Prometheus (GH#7112) crusaderky

• Round up saturation-factor (GH#7116) Gabe Joseph

• Return default on KeyError at any level in get_metadata (GH#7109) Hendrik Makait

• Count task states per task prefix and expose to Prometheus (GH#7088) Nat Tabris

• Add scheduler-sni option for dask workers (GH#6290) Burt Holzman

Bug Fixes

• Improve exception catching in UCX communication (GH#7132) Peter Andreas Entschev

• Improve robustness of PipInstall plugin (GH#7111) Hendrik Makait

Maintenance

• Fix dependencies that should point to dask/dask (GH#7138) James Bourbeau

• Hold on to z.sum() until test completes (GH#7136) Lawrence Mitchell

• Bump peter-evans/create-pull-request from 3 to 4 (GH#7120)

• Update typing for system_monitor after python/typeshed#8829 (GH#7131) Lawrence Mitchell

• Fix two potentially flaky queuing tests (GH#7124) Gabe Joseph

• Bump EnricoMi/publish-unit-test-result-action from 1 to 2 (GH#7121)

• Bump actions/checkout from 2 to 3.1.0 (GH#7119)

• Revamp SystemMonitor (GH#7097) crusaderky

• Bump actions/cache from 2 to 3 (GH#7118)

• Bump actions/upload-artifact from 2 to 3 (GH#7117)

• Move dependabot configuration file (GH#7115) James Bourbeau

• Enable dependabot for GitHub Actions (GH#7101) Florian Jetter

• Update coverage upload action (GH#7100) Florian Jetter

• Adjust hardware benchmarks bokeh test (GH#7096) Florian Jetter

• Multi-platform mypy checks (GH#7094) crusaderky

• Update gpuCI RAPIDS_VER to 22.12 (GH#7084)

2022.9.2

Released on September 30, 2022

Enhancements

• Smarter stealing with dependencies (GH#7024) Hendrik Makait

• Enable Active Memory Manager by default (GH#7042) crusaderky

• Allow timeout strings in distributed.wait (GH#7081) James Bourbeau

• Make AMM memory measure configurable (GH#7062) crusaderky

• AMM support for actors (GH#7072) crusaderky

• Expose message-bytes-limit in config (GH#7074) Hendrik Makait

• Detect mismatching Python version in scheduler (GH#7018) Hendrik Makait

• Improve KilledWorker message users see (GH#7043) James Bourbeau

• Support for cgroups v2 and respect soft limits (GH#7051) Samantha Hughes

Bug Fixes

• Catch BaseException on UCX read error (GH#6996) Peter Andreas Entschev

• Fix transfer limiting in _select_keys_for_gather (GH#7071) Hendrik Makait

• Parse worker-saturation if a string (GH#7064) Gabe Joseph

• Nanny(config=...) parameter overlays global dask config (GH#7069) crusaderky

• Ensure default clients don’t propagate to subprocesses (GH#7028) Florian Jetter

Documentation

• Improve documentation of message-bytes-limit (GH#7077) Hendrik Makait

• Minor tweaks to Sphinx documentation (GH#7041) crusaderky

• Improve upload_file API documentation (GH#7040) Florian Jetter

Maintenance

• test_serialize_numba: Workaround issue with np.empty_like in NP 1.23 (GH#7089) Graham Markall

• Type platform constants for mypy (GH#7091) jakirkham

• dask-worker-space (GH#7054) crusaderky

• Remove failing test case (GH#7087) Hendrik Makait

• test_default_client (GH#7058) crusaderky

• Fix pre-commit fails with recent versions of mypy and pandas (GH#7068) crusaderky

• Add factorization utility (GH#7048) James Bourbeau

2022.9.1

Released on September 16, 2022

Enhancements

• Add dashboard component for size of open data transfers (GH#6982) Hendrik Makait

• Allow very fast keys and very expensive transfers as stealing candidates (GH#7022) Florian Jetter

Bug Fixes

• No longer double count transfer cost in stealing (GH#7036) Hendrik Makait

Maintenance

• Make test_wait_first_completed robust (GH#7039) Florian Jetter

• Partial annotations for SchedulerState (GH#7023) crusaderky

• Add more type annotations to stealing.py (GH#7009) Florian Jetter

• Update codecov settings (GH#7015) Florian Jetter

• Speed up test_balance (GH#7008) Florian Jetter

• Fix test report after queuing job added (GH#7012) Gabe Joseph

• Clean up env variables in Gihub Actions (GH#7001) crusaderky

• Make test_steal_reschedule_reset_in_flight_occupancy non timing dependent (GH#7010) Florian Jetter

• Replaced distributed.utils.key_split with dask.utils.key_split (GH#7005) Luke Conibear

• Revert “Revert “Limit incoming data transfers by amount of data” (GH#6994)” (:pr:`7007) Florian Jetter

• CI job running tests with queuing on (GH#6989) Gabe Joseph

• Fix distributed/tests/test_client_executor.py::test_wait (GH#6990) Florian Jetter

2022.9.0

Released on September 2, 2022

Enhancements

• Limit incoming data transfers by amount of data (GH#6975) Hendrik Makait

• Expose transfer-related metrics in Worker.get_metrics and WorkerMetricCollector (GH#6936) Hendrik Makait

• Withhold root tasks (no co assignment) (GH#6614) Gabe Joseph

• Improve differentiation between incoming/outgoing connections and transfers (GH#6933) Hendrik Makait

• Change memory bars color on spilling/paused status (GH#6959) crusaderky

• Ensure restart clears taskgroups et al (GH#6944) Florian Jetter

• Optimise scheduler.get_comm_cost set difference (GH#6931) Lawrence Mitchell

• Expose setting multiple protocols and ports via the dask-scheduler CLI (GH#6898) Jacob Tomlinson

• Make TextProgressBar clear the line when finished (GH#5968) Vincenzo Eduardo Padulano

Bug Fixes

• Revert getaddrinfo fast path for Python 3.8 (GH#6978) Florian Jetter

• cancelled/resumed->long-running transitions (GH#6916) crusaderky

Deprecations

• Deprecate default value for Client.wait_for_workers (GH#6942) Florian Jetter

Documentation

• Document Scheduler and Worker state machine (GH#6948) crusaderky

• Insert memory_limit parameter into LocalCluster docstring (GH#6839) Crislana Rafael

Maintenance

• Revert “Limit incoming data transfers by amount of data” (GH#6994) Florian Jetter

• Cache conda environment between CI test runs (GH#6855) Charles Blackmon-Luca

• Revert “Fix co-assignment for binary operations” (GH#6985) Gabe Joseph

• Cache test_report shelves in CI (GH#6937) Florian Jetter

• Cleanup ipywidgets mocking (GH#6918) Thomas Grainger

• Improve testing of {Scheduler|Worker}MetricCollector (GH#6945) Hendrik Makait

• Clean up nanny WorkerProcess.kill (GH#6972) Gabe Joseph

• Rewrite test_reconnect to use subprocess to kill scheduler reliably (GH#6967) Florian Jetter

• Task state domain on the scheduler side (GH#6929) crusaderky

• Remove @avoid_ci from test_steal (GH#6872) crusaderky

• Use async with Worker in tests (GH#6958) crusaderky

• Ignore spurious warnings in test_quiet_close_process (GH#6955) crusaderky

• Fix tests on Windows (GH#6954) Hendrik Makait

• Prevent duplicates in HeapSet.sorted() (GH#6952) crusaderky

• Propagate worker address and improve _remove_from_processing behavior (GH#6946) Hendrik Makait

• Add HeapSet._sorted internal flag (GH#6949) Gabe Joseph

• Add HeapSet.peekn (GH#6947) Gabe Joseph

• Fix pyright error when importing from distributed (GH#6904) Ian Liu Rodrigues

• Always return ws.address from _remove_from_processing (GH#6884) Hendrik Makait

• Use async with Client: in tests (GH#6921) Thomas Grainger

• Ensure relative memory limits work as percentage of system memory (GH#6923) Florian Jetter

2022.8.1

Released on August 19, 2022

New Features

Enhancements

• Drop comparison of versions against all clients (GH#6861) Hendrik Makait

• Log the worker name if set (GH#6866) Johannes Lange

• Skip getaddrinfo thread if host is already resolved, using socket.AI_NUMERIC* (GH#6847) Thomas Grainger

• Display unexpected state in Worker.execute validation (GH#6856) James Bourbeau

• pre-spawn-environ (GH#6841) crusaderky

• Dump has_what, missing_dep_flight (GH#6830) Gabe Joseph

Bug Fixes

• cancelled / resumed -> rescheduled transition (GH#6913) crusaderky

• Fix resource deallocation for resumed tasks (GH#6914) crusaderky

• Only close scheduler in SpecCluster if it exists (GH#6888) Matthew Rocklin

• Fix issue if exc.reason is None (GH#6881) Hendrik Makait

• Always close BatchedSend write coroutines (GH#6865) Gabe Joseph

• Harden preamble of Worker.execute against race conditions (GH#6878) crusaderky

Documentation

• Fix typo (GH#6870) Pieter Gijsbers

Maintenance

• Use retries for the test report (GH#6926) Florian Jetter

• Duplicated code: in_flight_tasks validation (GH#6917) crusaderky

• ipywidgets 8 compatibility (GH#6912) James Bourbeau

• Overhaul transitions for the resumed state (GH#6699) crusaderky

• Don’t upgrade to ipywidgets 8 (GH#6910) crusaderky

• Clean up cluster process reaping (GH#6840) Gabe Joseph

• Don’t use bokeh Figure in tests (GH#6721) Bryan Van de Ven

• Work around incompatibility of crick with setuptools 65 (GH#6887) crusaderky

• Add max version constraint for dask-core in nightlies (GH#6862) Charles Blackmon-Luca

• Replace port = random.randint( with port = d.utils.open_port (GH#6883) Thomas Grainger

• Fix flaky test_wall_clock (GH#6879) crusaderky

• Add descriptive error message to assert (GH#6871) Hendrik Makait

• Increase timeout in test_quiet_process (GH#6857) Florian Jetter

• Descriptive title for test report (GH#6849) Hendrik Makait

• Add flake8-bugbear as plugin to pre-commit (GH#6809) Hendrik Makait

• Remove redundant use of with clean(): (GH#6852) Thomas Grainger

• Show actual Job URL on test report (GH#6837) Florian Jetter

• Update pre-commit dependencies (GH#6851) Hendrik Makait

• Call exit callback even if AsyncProcess is reaped elsewhere (GH#6684) Thomas Grainger

• Avoid function calls in argument defaults (GH#6812) Hendrik Makait

• Ignore warning for unclose SSHCluster in tests (GH#6827) Florian Jetter

2022.8.0

Released on August 5, 2022

New Features

• Add Jupyter Server to Dask Scheduler (GH#6737) Matthew Rocklin

Enhancements

• Human-readable formatting for disk I/O and renaming to diff net and disk (GH#6835) Hendrik Makait

• Add Cluster.get_client() method (GH#6745) Julia Signell

• Start bokeh app to activate bokeh’s clean session callbacks (GH#6728) Martí Zamora

Bug Fixes

• Ensure Nanny doesn’t restart workers that fail to start, and joins subprocess (GH#6427) Gabe Joseph

• Don’t connect to cluster subprocesses at shutdown (GH#6829) Gabe Joseph

• Fix restart wait for workers edge case (GH#6823) Gabe Joseph

• Fix spilled size calculation in Slow (GH#6789) Hendrik Makait

Deprecations

• Deprecate passing stopped loops to LoopRunner (and therefore Client/Cluster) (GH#6680) Thomas Grainger

Documentation

• Add text to top of API docs to make sure that users are exposed to LocalCluster early (GH#6793) Julia Signell

• Change title for plugins documentation (GH#6733) Sarah Charlotte Johnson

Maintenance

• Only set 5s connect timeout in gen_cluster tests (GH#6822) Gabe Joseph

• Fix flaky test_worker_who_has_clears_after_failed_connection (GH#6832) Gabe Joseph

• Add missing skips for pyarrow (GH#6787) Elliott Sales de Andrade

• Miscellaneous flake8-bugbear issues (GH#6814) Hendrik Makait

• Assert otherwise pointless comparisons (B015) (GH#6811) Hendrik Makait

• Remove unused functions from utils_test.py (GH#6807) Hendrik Makait

• Fix Jupyter security note (GH#6818) Jacob Tomlinson

• Improve check_thread_leak output (GH#6797) Gabe Joseph

• Use contextmanager to ensure clients are closed and do not leak (GH#6817) Hendrik Makait

• Robust thread termination in test_watch and test_watch_requires_lock_to_run (GH#6788) Hendrik Makait

• Avoid unused loop control variable or name them _ (GH#6813) Hendrik Makait

• Replace assert False where an exception should always be thrown (GH#6815) Hendrik Makait

• Avoid mutable argument defaults in tests (GH#6810) Hendrik Makait

• Avoid mutable argument defaults outside of tests (GH#6665) Hendrik Makait

• Update gpuCI RAPIDS_VER to 22.10 (GH#6798)

• Use same Python for dask worker tests (GH#6786) Elliott Sales de Andrade

2022.7.1

Released on July 22, 2022

New Features

• Dashboard for failed tasks (GH#6595) Ian Rose

Enhancements

• Wait for workers to return in Client.restart (GH#6714) Gabe Joseph

• Remove global mutable Cluster._cluster_info (GH#6487) Thomas Grainger

Bug Fixes

• Fix: nvml no early init (GH#6678) Lawrence Mitchell

• Fix bug when restarting client (GH#6654) Iain Dorrington

• Failure to spill breaks available resources (GH#6703) crusaderky

• Fix resource allocation for tasks with dependencies (GH#6676) Hendrik Makait

Maintenance

• Revert “Set MALLOC_TRIM_THRESHOLD_ before interpreter start” (GH#6777) Gabe Joseph

• Fix mypy lint in CI (GH#6779) jakirkham

• Remove test_restart_fast_sync, test_fast_kill (GH#6750) Gabe Joseph

• Fix flaky test_async_task_group_call_later_executes_delayed_task_in_background (GH#6744) Hendrik Makait

• Drop redundant geninc (GH#6740) Hendrik Makait

• Remove unused worker_coroutines (GH#6739) Gabe Joseph

• Store ready and constrained tasks in heapsets (GH#6711) crusaderky

• Improve tests for cancelled state (GH#6717) crusaderky

• Future-proof Bokeh value import (GH#6707) Bryan Van de Ven

• Revert temporary stress test (GH#6712) crusaderky

• Validate constrained tasks (GH#6698) crusaderky

• Minor quality-of-life tweaks to cancelled state (GH#6701) crusaderky

• Pickle worker state machine exceptions (GH#6702) crusaderky

• Partial matches for worker state machine instructions (GH#6704) crusaderky

• Automatically mark all WorkerState tests (GH#6706) crusaderky

2022.7.0

Released on July 8, 2022

Enhancements

• Use a tempdir path by default instead of cwd for the worker scratch dir (GH#6658) Florian Jetter

• Add WorkerState.all_running_tasks (GH#6690) Hendrik Makait

• Scheduler.reschedule() works only by accident (GH#6339) crusaderky

• Remove spurious UnpauseEvent at worker start (GH#6652) crusaderky

• Log if closing an executor is not possible in thread (GH#6644) Florian Jetter

• Cloudpickle register by value (GH#6466) Ian Rose

• Adding replicas to a task in fetch now sends it to flight immediately (GH#6594) crusaderky

Bug Fixes

• Fix dump output of parameter-less events (GH#6695) crusaderky

• Set MALLOC_TRIM_THRESHOLD_ before interpreter start (GH#6681) crusaderky

• Fix deadlocks around rescheduled and resumed states (GH#6673) crusaderky

• has_arg returns True for keyword-only arguments (GH#6648) Lawrence Mitchell

• Transitions caused by worker death use old ‘worker-connect’ stimulus_id (GH#6657) crusaderky

• A key is forgotten while acquire-replicas is running (GH#6638) crusaderky

Maintenance

• Revisit WorkerState.long_running set (GH#6697) crusaderky

• WorkerState unit tests for resumed state (GH#6688) crusaderky

• Bump version of pandas-stubs (GH#6691) crusaderky

• Add dummy factory methods for ExecuteSuccessEvent and ExecuteFailureEvent (GH#6687) Hendrik Makait

• Pin tornado<6.2 in nightly conda recipes (GH#6675) Peter Andreas Entschev

• Refactor resource restriction handling in WorkerState (GH#6672) Hendrik Makait

• test_signal and test_terminate occasionally return SIGKILL on MacOS (GH#6671) crusaderky

• Use the loop fixture in even more tests (GH#6674) Thomas Grainger

• Unconditionally import ssl (GH#6670) Thomas Grainger

• Use the loop fixture in more tests (GH#6642) Thomas Grainger

• Pin tornado to <6.2 (GH#6668) Florian Jetter

• Handle AsyncTaskGroupClosedError (GH#6664) Hendrik Makait

• Replace occurrences of large delay slowinc with locks (GH#6656) Florian Jetter

• Merge extend-ignore and ignore values for flake8 (GH#6660) Hendrik Makait

• Remove server close background task grace period (GH#6633) Thomas Grainger

• Do not use tempfile in utils_test (GH#6651) Florian Jetter

• close_worker cleanup (GH#6650) crusaderky

• Rewrite test_cancelled_resumed_after_flight_with_dependencies using WorkerState (GH#6645) crusaderky

• Log the actual duration to create a directory (GH#6647) Florian Jetter

• pandas type stubs (GH#6635) crusaderky

• Remove unused __started Event in Server (GH#6615) Florian Jetter

• Use safe temp directory in gen_cluster (GH#6628) Florian Jetter

• Print CI host info (GH#6629) crusaderky

• Deduplicate data_needed (GH#6587) crusaderky

• Remove EnsureCommunicatingAfterTransitions (GH#6462) crusaderky

• Pickle WorkerState (GH#6623) crusaderky

• Harden vs. TaskState collisions (GH#6593) crusaderky

• Do not interact with the event loop when the cluster is garbage collected (GH#6627) Thomas Grainger

2022.6.1

Released on June 24, 2022

Highlights

This release includes the Worker State Machine refactor. The expectation should be that the worker state is its own synchronous subclass. Pulling all the state out into its own class allows us to write targeted unit tests without invoking any concurrent or asynchronous code.

See GH#5736 for more information.

Enhancements

• Make worker state machine methods private (GH#6564) crusaderky

• Yank state machine out of Worker class (GH#6566) crusaderky

• Track worker_state_machine.TaskState instances (GH#6525) Hendrik Makait

• Trivial tweaks to the Worker State Machine (GH#6586) crusaderky

• Replace loop.call_later and loop.add_callback with background tasks added to Server. (GH#6603) Thomas Grainger

• Support for neater WorkerState tests (GH#6609) crusaderky

• Limit TCP writes with Tornado to 2GB (GH#6557) hhuuggoo

• Enable no_implicit_optional for scheduler (GH#6622) Thomas Grainger

Bug Fixes

• Partial revert of compute-task message format (GH#6626) Florian Jetter

• Restore log message about received signals in CLI (GH#6618) Florian Jetter

• Handle empty memoryviews of bytearrays when (de)serializing (GH#6576) Benjamin Zaitlen

• Ensure steal requests from same-IP but distinct workers are rejected (GH#6585) Florian Jetter

• Fix tls_(min|max)_ version having no effect on openssl 1.1.0g or lower (GH#6562) Thomas Grainger

• Fix idle_timeout and unxfail test (GH#6563) Matthew Rocklin

• Fix crashing debug statement in _purge_state (GH#6589) crusaderky

• Abort connections on CancelledError (GH#6574) Thomas Grainger

• Fix Active Memory Manager ignores nbytes thresholds (GH#6583) crusaderky

Deprecations

• Deprecate WorkerState accessors (GH#6579) crusaderky

Maintenance

• Remove ipython hack (GH#6599) crusaderky

• Mypy enforce --no-implicit-optional (GH#6606) crusaderky

• Update versioneer: change from using SafeConfigParser to ConfigParser (GH#6605) Thomas A Caswell

• Warn unreachable for scheduler.py (GH#6611) Florian Jetter

• Refactor wait_for_state() (GH#6581) crusaderky

• Hardcode wait_for_signals signal list (GH#6619) Thomas Grainger

• Always pick an open port when running tests (GH#6591) Florian Jetter

• Log popen stdout/err when subprocess times out (GH#6567) Gabe Joseph

• Fix test_error_during_startup (GH#6608) Florian Jetter

• Make test_idle_timeout_no_workers more robust (GH#6602) Florian Jetter

• Mypy enforce --disallow-incomplete-defs (GH#6601) crusaderky

• Do not log during signal handler (GH#6590) Florian Jetter

• Don’t initialize mp_context on import (GH#6580) Lawrence Mitchell

• Test retire workers deadlock (GH#6240) Gabe Joseph

• Rework some tests related to gather_dep (GH#6472) crusaderky

• Minor cosmetic review of scheduler_story and worker_story (GH#6442) crusaderky

• Force __future__.annotations with isort (GH#6621) Thomas Grainger

2022.6.0

Released on June 10, 2022

Enhancements

• Make disk access in system monitor configurable (GH#6537) Matthew Rocklin

• Log and except errors on preload start (GH#6553) Matthew Rocklin

• Fix Scheduler.restart logic (GH#6504) Gabe Joseph

• Don’t heartbeat while Worker is closing (GH#6543) Gabe Joseph

• No longer retry LocalCluster on errno.EADDRINUSE (GH#6369) Thomas Grainger

• Don’t invoke log_event from state machine (GH#6512) crusaderky

• Add config option to disable profiling and disable it in many tests per default (GH#6490) Hendrik Makait

• Encapsulate Worker.batched_stream.send() (GH#6475) crusaderky

Bug Fixes

• refresh-who-has can break the worker state machine (GH#6529) crusaderky

• Restart worker if it’s unrecognized by scheduler (GH#6505) Gabe Joseph

• Fix import error when distributed.rmm.pool-size is set (GH#6482) KoyamaSohei

Deprecations

• Restore signature compatibility for dask-gateway (GH#6561) Tom Augspurger

• Deprecate the io_loop and loop kwarg to Server, Worker, and Nanny (GH#6473) Thomas Grainger

• Deprecate the loop kwarg to Scheduler (GH#6443) Thomas Grainger

Documentation

• Fix typo in .nthreads() docstring example (GH#6545) Pavithra Eswaramoorthy

• Update docs theme for rebranding (GH#6495) Sarah Charlotte Johnson

Maintenance

• Refactor gather_dep (GH#6388) crusaderky

• Fix flaky test_gather_dep_one_worker_always_busy (GH#6554) crusaderky

• Remove missing-data message (GH#6546) crusaderky

• Port test_local.LocalTest to pytest tests to allow use of loop fixture (GH#6523) Thomas Grainger

• Fix test_quiet_client_close (GH#6541) Gabe Joseph

• Use default_initializer in WorkerProcess (GH#6534) jakirkham

• Avoid deadlocks in tests that use popen (GH#6483) Gabe Joseph

• Revert “Fix CLI Scheduler Tests (GH#6502)” (GH#6547) Gabe Joseph

• Update test report URL in summary message (GH#6532) Gabe Joseph

• Update test report url (GH#6531) Ian Rose

• Assert AsyncProcess.set_exit_callback is not called with a coroutine function (GH#6526) Thomas Grainger

• Typing and docstring for Worker.close (GH#6518) Hendrik Makait

• Fix CLI Scheduler Tests (GH#6502) Benjamin Zaitlen

• Collect assertions in test_as_current_is_thread_local (GH#6520) Thomas Grainger

• Link test report from test results comment (GH#6524) Hendrik Makait

• Ignore the return value of signal.signal (GH#6519) Thomas Grainger

• Refactor all event handlers (GH#6410) crusaderky

• Fix dashboard favicon background (GH#6514) Jacob Tomlinson

• Update dashboard logo (GH#6513) Jacob Tomlinson

• Fix test_stress_scatter_death (GH#6404) Florian Jetter

• Remove CrossFilter widget (GH#6484) crusaderky

• data_needed exclusively contains tasks in fetch state (GH#6481) crusaderky

• Assert possible previous states (GH#6488) Florian Jetter

• @fail_hard can kill the whole test suite; hide errors (GH#6474) crusaderky

• Assert that a fetch->cancelled->resumed->fetch cycle is impossible (GH#6460) crusaderky

• Refactor busy workers reinsertion (GH#6379) crusaderky

• Refactor find_missing and refresh_who_has (GH#6348) crusaderky

• Rename test_collections.py to test_dask_collections.py (GH#6486) crusaderky

• update_who_has can remove workers (GH#6342) crusaderky

• Restructure test_watch_requires_lock_to_run to avoid flakes (GH#6469) Hendrik Makait

• Fix intermittent test_profile_plot failure (GH#6456) Matthew Rocklin

• Use asyncio.run to run gen_cluster, gen_test and cluster (GH#6231) Thomas Grainger

• Improve tests that watch for subprocess logs (GH#6461) Gabe Joseph

2022.05.2

Released on May 26, 2022

Enhancements

• Add a lock to distributed.profile for better concurrency control (GH#6421) Hendrik Makait

• Send SIGKILL after SIGTERM when passing 95% memory (GH#6419) crusaderky

Bug Fixes

• Log rather than raise exceptions in preload.teardown() (GH#6458) Matthew Rocklin

• Handle failing plugin.close() calls during scheduler shutdown (GH#6450) Matthew Rocklin

• Fix slicing bug in ensure_memoryview (GH#6449) jakirkham

• Generalize UCX errors on connect() and correct pytest fixtures (GH#6434) Peter Andreas Entschev

• Run cluster widget periodic callbacks on the correct event loop (GH#6444) Thomas Grainger

Maintenance

• Disable pytest-asyncio if installed (GH#6436) Jacob Tomlinson

• Close client in sync test_actor tests (GH#6459) Thomas Grainger

• Ignore ServerSession.with_document_locked unawaited (GH#6447) Thomas Grainger

• Remove coverage pin from Python 3.10 environment (GH#6439) Thomas Grainger

• Annotate remove_worker (GH#6441) crusaderky

• Update gpuCI RAPIDS_VER to 22.08 (GH#6428)

2022.05.1

Released on May 24, 2022

New Features

• Add HTTP API to scheduler (GH#6270) Matthew Murray

• Shuffle Service with Scheduler Logic (GH#6007) Matthew Rocklin

Enhancements

• Follow-up on removing report and safe from Worker.close (GH#6423) Gabe Joseph

• Server close faster (GH#6415) Florian Jetter

• Disable HTTP API by default (GH#6420) Jacob Tomlinson

• Remove report and safe from Worker.close (GH#6363) Florian Jetter

• Allow deserialized plugins in register_scheduler_plugin (GH#6401) Matthew Rocklin

• WorkerState are different for different addresses (GH#6398) Florian Jetter

• Do not filter tasks before gathering data (GH#6371) crusaderky

• Remove worker reconnect (GH#6361) Gabe Joseph

• Add SchedulerPlugin.log_event handler (GH#6381) Matthew Rocklin

• Ensure occupancy tracking works as expected for long running tasks (GH#6351) Florian Jetter

• stimulus_id for all Instructions (GH#6347) crusaderky

• Refactor missing-data command (GH#6332) crusaderky

• Add idempotent to register_scheduler_plugin client (GH#6328) Alex Ford

• Add option to specify a scheduler address for workers to use (GH#5944) Enric Tejedor

Bug Fixes

• Remove stray breakpoint (GH#6417) Thomas Grainger

• Fix API JSON MIME type (GH#6397) Jacob Tomlinson

• Remove wrong assert in handle compute (GH#6370) Florian Jetter

• Ensure multiple clients can cancel their key without interference (GH#6016) Florian Jetter

• Fix Nanny shutdown assertion (GH#6357) Gabe Joseph

• Fix fail_hard for sync functions (GH#6269) Gabe Joseph

• Prevent infinite transition loops; more aggressive validate_state() (GH#6318) crusaderky

• Ensure cleanup of many GBs of spilled data on terminate (GH#6280) crusaderky

• Fix WORKER_ANY_RUNNING regression (GH#6297) Florian Jetter

• Race conditions from fetch to compute while AMM requests replica (GH#6248) Florian Jetter

• Ensure resumed tasks are not accidentally forgotten (GH#6217) Florian Jetter

• Do not allow closing workers to be awaited again (GH#5910) Florian Jetter

Deprecations

• Move wait_for_signals to private module and deprecate distributed.cli.utils (GH#6367) Hendrik Makait

Documentation

• Fix typos and whitespace in worker.py (GH#6326) Hendrik Makait

• Fix link to memory trimming documentation (GH#6317) Marco Wolsza

Maintenance

• Make gen_test show up in VSCode test discovery (GH#6424) Gabe Joseph

• WSMR / deserialize_task (GH#6411) crusaderky

• Restore signal handlers after wait for signals is done (GH#6400) Thomas Grainger

• fail_hard should reraise (GH#6399) crusaderky

• Revisit tests mocking gather_dep (GH#6385) crusaderky

• Fix flaky test_in_flight_lost_after_resumed (GH#6372) Florian Jetter

• Restore install_signal_handlers due to downstream dependencies (GH#6366) Hendrik Makait

• Improve catch_unhandled_exceptions (GH#6358) Gabe Joseph

• Remove all invocations of IOLoop.run_sync from CLI (GH#6205) Hendrik Makait

• Remove transition-counter-max from config (GH#6349) crusaderky

• Use list comprehension in pickle_loads (GH#6343) jakirkham

• Improve ensure_memoryview test coverage & make minor fixes (GH#6333) jakirkham

• Remove leaking reference to workers from gen_cluster (GH#6337) Hendrik Makait

• Partial annotations for stealing.py (GH#6338) crusaderky

• Validate and debug state machine on handle_compute_task (GH#6327) crusaderky

• Bump pyupgrade and clean up # type: ignore (GH#6293) crusaderky

• gen_cluster to write to /tmp (GH#6335) crusaderky

• Transition table as a ClassVar (GH#6331) crusaderky

• Simplify ensure_memoryview test with array (GH#6322) jakirkham

• Refactor ensure_communicating (GH#6165) crusaderky

• Review scheduler annotations, part 2 (GH#6253) crusaderky

• Use w for writeable branch in pickle_loads (GH#6314) jakirkham

• Simplify frame handling in ws (GH#6294) jakirkham

• Use ensure_bytes from dask.utils (GH#6295) jakirkham

• Use ensure_memoryview in array deserialization (GH#6300) jakirkham

• Escape < > when generating Junit report (GH#6306) crusaderky

• Use codecs.decode to deserialize errors (GH#6274) jakirkham

• Minimize copying in maybe_compress & byte_sample (GH#6273) jakirkham

• Skip test_release_evloop_while_spilling on OSX (GH#6291) Florian Jetter

• Simplify logic in get_default_compression (GH#6260) jakirkham

• Cleanup old compression workarounds (GH#6259) jakirkham

• Re-enable NVML monitoring for WSL (GH#6119) Charles Blackmon-Luca

2022.05.0

Released on May 2, 2022

Highlights

This is a bugfix release for this issue.

Enhancements

• Handle writeable in buffer_callback (GH#6238) jakirkham

• Use .data with NumPy array allocation (GH#6242) jakirkham

Bug Fixes

• Close executor in event loop if interpreter is closing (GH#6256) Matthew Rocklin

2022.04.2

Released on April 29, 2022

Enhancements

• Unblock event loop while waiting for ThreadpoolExecutor to shut down (GH#6091) Florian Jetter

• RetireWorker policy is done if removed (GH#6234) Gabe Joseph

• Pause to disable dependency gathering (GH#6195) crusaderky

• Add EOFError to nanny multiprocessing.queue except list (GH#6213) Matthew Rocklin

• Re-interpret error in lost worker scenario (GH#6193) Matthew Rocklin

• Add Stimulus IDs to Scheduler (GH#6161) Florian Jetter

• Set a five minute TTL for Dask workers (GH#6200) Matthew Rocklin

• Add distributed.metrics.monotonic (GH#6181) crusaderky

• Send worker validation errors to scheduler and err on test completion (GH#6192) Matthew Rocklin

• Redesign worker exponential backoff on busy-gather (GH#6173) crusaderky

• Log all invalid worker transitions to scheduler (GH#6134) Matthew Rocklin

• Make Graph dashboard plot have invisible axes (GH#6149) Matthew Rocklin

• Remove Nanny auto_restart state (GH#6138) Matthew Rocklin

Bug Fixes

• Ensure scheduler events do not hold on to TaskState objects (GH#6226) Florian Jetter

• Allow pausing and choke event loop while spilling (GH#6189) crusaderky

• Do not use UUID in stealing (GH#6179) Florian Jetter

• Handle int worker names in info page (GH#6158) Brett Naul

• Fix psutil dio counters none (GH#6093) ungarj

• Join Nanny watch thread (GH#6146) Matthew Rocklin

• Improve logging when closing workers (GH#6129) Matthew Rocklin

• Avoid stack overflow in profiling (GH#6141) Matthew Rocklin

• Clean up SSHCluster if failure to start (GH#6130) Matthew Rocklin

Deprecations

• Deprecate rpc synchronous context manager (GH#6171) Thomas Grainger

Documentation

• Update actors.rst (GH#6167) Scott Sievert

Maintenance

• Add fail_hard decorator for worker methods (GH#6210) Matthew Rocklin

• Do not require pytest_timeout (GH#6224) Florian Jetter

• Remove remaining run_sync calls from tests (GH#6196) Thomas Grainger

• Increase test timeout if debugger is running (GH#6218) Florian Jetter

• Do not list closes keyword in list of bullet points (GH#6219) Florian Jetter

• Harmonize (GH#6161) and (GH#6173) (GH#6207) crusaderky

• Xfail test_worker_death_timeout (GH#6186) Matthew Rocklin

• Use random port in test_dask_spec.py::test_text (GH#6187) Matthew Rocklin

• Mark all websocket tests as flaky (GH#6188) Matthew Rocklin

• Fix flaky test_dont_steal_long_running_tasks (GH#6197) crusaderky

• Cleanup names in stealing (GH#6185) Matthew Rocklin

• log_errors decorator (GH#6184) crusaderky

• Pass mypy validation on Windows (GH#6180) crusaderky

• Add locket as a dependency instead of vendoring (GH#6166) Michael Adkins

• Remove unittestmock for gather_dep and get_data_from_worker (GH#6172) Florian Jetter

• mypy tweaks (GH#6175) crusaderky

• Avoid easy deprecated calls to asyncio.get_event_loop() (GH#6170) Thomas Grainger

• Fix flaky test_cancel_fire_and_forget (GH#6099) crusaderky

• Remove deprecated code (GH#6144) Matthew Rocklin

• Update link of test badge (GH#6154) Florian Jetter

• Remove legacy state mappings (GH#6145) Matthew Rocklin

• Fix test_worker_waits_for_scheduler (GH#6155) Matthew Rocklin

• Disallow leaked threads on windows (GH#6152) Thomas Grainger

• Review annotations and docstrings in scheduler.py, part 1 (GH#6132) crusaderky

• Relax test_asyncprocess.py::test_simple (GH#6150) Matthew Rocklin

• Drop cast ing which is effectively a no-op (GH#6101) jakirkham

• Mark tests that use a specific port as flaky (GH#6139) Matthew Rocklin

• AMM Suggestion namedtuples (GH#6108) crusaderky

2022.04.1

Released on April 15, 2022

New Features

• Add KillWorker Plugin (GH#6126) Matthew Rocklin

Enhancements

• Sort worker list in info pages (GH#6135) Matthew Rocklin

• Add back Worker.transition_fetch_missing (GH#6112) Matthew Rocklin

• Log state machine events (GH#6092) crusaderky

• Migrate ensure_executing transitions to new WorkerState event mechanism - part 1 (GH#6003) crusaderky

• Migrate ensure_executing transitions to new WorkerState event mechanism - part 2 (GH#6062) crusaderky

• Annotate worker transitions to error (GH#6012) crusaderky

Bug Fixes

• Avoid transitioning from memory/released to missing in worker (GH#6123) Matthew Rocklin

• Don’t try to reconnect client on interpreter shutdown (GH#6120) Matthew Rocklin

• Wrap UCX init warnings in importable functions (GH#6121) Charles Blackmon-Luca

• Cancel asyncio tasks on worker close (GH#6098) crusaderky

• Avoid port collisions when defining port ranges (GH#6054) crusaderky

Maintenance

• Avoid intermittent failure in test_cancel_fire_and_forget (GH#6131) Matthew Rocklin

• Ignore bokeh warning in pytest (GH#6127) Matthew Rocklin

• Start uncythonization (GH#6104) Martin Durant

• Avoid redundant cleanup fixture in gen_test tests (GH#6118) Thomas Grainger

• Move comm.close to finally in test_comms (GH#6109) Florian Jetter

• Use async with Server in test_core.py (GH#6100) Matthew Rocklin

• Elevate warnings to errors in the test suite (GH#6094) Thomas Grainger

• Add urllib3 to nightly conda builds (GH#6102) James Bourbeau

• Drop Blosc (GH#6027) Matthew Rocklin

• Robust test_get_returns_early (GH#6090) Florian Jetter

• Overhaul test_priorities.py (GH#6077) crusaderky

• Remove pytest-asyncio (GH#6063) Thomas Grainger

• Clean up usage around plain rpc (GH#6082) Florian Jetter

• Drop OSX builds for Python 3.9 (GH#6073) Florian Jetter

• Bump periods in utils_test.wait_for (GH#6081) Florian Jetter

• Check for ucx-py nightlies when updating gpuCI (GH#6006) Charles Blackmon-Luca

• Type annotations for profile.py (GH#6067) crusaderky

• Fix flaky test_worker_time_to_live (GH#6061) crusaderky

• Fix flaky test_as_completed_async_for_cancel (GH#6072) crusaderky

• Fix regression in test_weakref_cache (GH#6033) crusaderky

• Trivial fix to test_nanny_worker_port_range (GH#6070) crusaderky

• Drop deprecated tornado.netutil.ExecutorResolver (GH#6031) Thomas Grainger

• Delete asyncio.py (GH#6066) Thomas Grainger

• Tweak conda environment files (GH#6037) crusaderky

• Harden test_abort_execution_to_fetch and more (GH#6026) crusaderky

• Fix test_as_completed_with_results_no_raise and name comm (GH#6042) Matthew Rocklin

• Use more robust limits in test_worker_memory (GH#6055) Florian Jetter

2022.04.0

Released on April 1, 2022

Note

This is the first release with support for Python 3.10

New Features

• Add Python 3.10 support (GH#5952) Thomas Grainger

• New cluster dump utilities (GH#5920) Simon Perkins

• New ClusterDump SchedulerPlugin for dumping cluster state on close (GH#5983) Simon Perkins

• Track Event Loop intervals in dashboard plot (GH#5964) Matthew Rocklin

• ToPickle - Unpickle on the Scheduler (GH#5728) Mads R. B. Kristensen

Enhancements

• Retry on transient error codes in preload (GH#5982) Matthew Rocklin

• More idiomatic mypy configuration (GH#6022) crusaderky

• Name extensions and enable extension heartbeats (GH#5957) Matthew Rocklin

• Better error message on misspelled executor annotation (GH#6009) crusaderky

• Clarify that SchedulerPlugin must be subclassed (GH#6008) crusaderky

• Remove duplication from stealing (GH#5787) Duncan McGregor

• Remove cache in iscoroutinefunction to avoid holding on to refs (GH#5985) Florian Jetter

• Add title to individual plots (GH#5967) Matthew Rocklin

• Specify average in timeseries titles (GH#5974) Matthew Rocklin

Bug Fixes

• Do not catch CancelledError in CommPool (GH#6005) Florian Jetter

Deprecations

• Remove distributed._ipython_utils and dependents (GH#6036) Thomas Grainger

• Remove support for PyPy (GH#6029) James Bourbeau

• Drop runtime dependency to setuptools (GH#6017) crusaderky

• Remove heartbeats from events (GH#5989) Matthew Rocklin

Documentation

• Mention default value of Client’s timeout (GH#5933) Eric Engestrom

• Update celery and other outdated 3rd party URLs (GH#5988) Thomas Grainger

Maintenance

• Improve test_hardware test (GH#6039) Matthew Rocklin

• Short variant of test_report.html (GH#6034) crusaderky

• Make test_reconnect async (GH#6000) Matthew Rocklin

• Update gpuCI RAPIDS_VER to 22.06 (GH#5962)

• Add tiny test for ToPickle (GH#6021) Matthew Rocklin

• Remove check_python_3 (broken with click>=8.1.0) (GH#6018) Thomas Grainger

• Fix black in CI (GH#6019) crusaderky

• Add a hardware benchmark to test memory, disk, and network bandwidths (GH#5966) Matthew Rocklin

• Relax variable test_race (GH#5993) Matthew Rocklin

• Skip dask-ssh tests without paramiko (GH#5907) Elliott Sales de Andrade

• Remove test_restart_sync_no_center (GH#5994) Matthew Rocklin

• Set lower tick frequency in tests (GH#5977) Matthew Rocklin

• Catch NotADirectoryError in SafeTemporaryDirectory (GH#5984) Florian Jetter

• Fix flaky test_weakref_cache (GH#5978) crusaderky

• Fixup test_worker_doesnt_await_task_completion (GH#5979) Matthew Rocklin

• Use broader range in test_nanny_worker_port_range (GH#5980) Matthew Rocklin

• Use tempfile directory in cluster fixture (GH#5825) Florian Jetter

• Drop setuptools from distributed recipe (GH#5963) jakirkham

2022.03.0

Released on March 18, 2022

New Features

• Support dumping cluster state to URL (GH#5863) Gabe Joseph

Enhancements

• Prevent data duplication on unspill (GH#5936) crusaderky

• Encapsulate spill buffer and memory_monitor (GH#5904) crusaderky

• Drop pkg_resources in favour of importlib.metadata (GH#5923) Thomas Grainger

• Worker State Machine refactor: redesign TaskState and scheduler messages (GH#5922) crusaderky

• Tidying of OpenSSL 1.0.2/Python 3.9 (and earlier) handling (GH#5854) jakirkham

• zict type annotations (GH#5905) crusaderky

• Add key to compute failed message (GH#5928) Florian Jetter

• Change default log format to include timestamp (GH#5897) Florian Jetter

• Improve type annotations in worker.py (GH#5814) crusaderky

Bug Fixes

• Fix progress_stream teardown (GH#5823) Thomas Grainger

• Handle concurrent or failing handshakes in InProcListener (GH#5903) Thomas Grainger

• Make log_event threadsafe (GH#5946) Gabe Joseph

Documentation

• Fixes to documentation regarding plugins (GH#5940) crendoncoiled

• Some updates to scheduling policies docs (GH#5911) Gabe Joseph

Maintenance

• Fix test_nanny_worker_port_range hangs on Windows (GH#5956) crusaderky

• (REVERTED) Unblock event loop while waiting for ThreadpoolExecutor to shut down (GH#5883) Florian Jetter

• Revert GH#5883 (GH#5961) crusaderky

• Invert event_name check in test-report job (GH#5959) jakirkham

• Only run gh-pages workflow on dask/distributed (GH#5942) jakirkham

• absolufy-imports - No relative imports - PEP8 (GH#5924) Florian Jetter

• Fix track_features for distributed pre-releases (GH#5927) Charles Blackmon-Luca

• Xfail test_submit_different_names (GH#5916) Florian Jetter

• Fix distributed pre-release’s distributed-impl constraint (GH#5867) Charles Blackmon-Luca

• Mock process memory readings in test_worker.py (v2) (GH#5878) crusaderky

• Drop unused _round_robin global variable (GH#5881) jakirkham

• Add GitHub URL for PyPi (GH#5886) Andrii Oriekhov

• Mark xfail COMPILED tests skipif instead (GH#5884) Florian Jetter

2022.02.1

Released on February 25, 2022

New Features

• Add the ability for Client to run preload code (GH#5773) Bryan W. Weber

Enhancements

• Optionally use NumPy to allocate buffers (GH#5750) jakirkham

• Add git hash to distributed-impl version (GH#5865) Charles Blackmon-Luca

• Immediately raise exception when trying to connect to a closed cluster (GH#5855) Florian Jetter

• Lazily get dask version information (GH#5822) Thomas Grainger

• Remove the requirements to add comm to every handler (GH#5820) Florian Jetter

• Raise on unclosed comms in check_instances (GH#5836) Florian Jetter

• Python 3.8 f-strings (GH#5828) crusaderky

• Constrained spill (GH#5543) Naty Clementi

• Measure actual spilled bytes, not output of sizeof() (GH#5805) crusaderky

• Remove redundant str() conversions (GH#5810) crusaderky

• Cluster dump now excludes run_spec by default (GH#5803) Florian Jetter

• Dump more objects with dump_cluster_state (GH#5806) crusaderky

• Do not connect to any sockets on import (GH#5808) Florian Jetter

Bug Fixes

• Avoid deadlock when two tasks are concurrently waiting for an unresolved ActorFuture (GH#5709) Thomas Grainger

Deprecations

• Drop Python 3.7 (GH#5683) James Bourbeau

• Remove support for UCX < 1.11.1 (GH#5859) Peter Andreas Entschev

Documentation

• Fix typo in memory types documentation relative links (GH#5845) James Bourbeau

• Document and test spill->target hysteresis cycle (GH#5813) crusaderky

Maintenance

• Fix flaky test_remove_replicas_while_computing (GH#5860) crusaderky

• Fix time based test_assert_worker_story_malformed_story parameterize (GH#5856) Thomas Grainger

• Remove xfail from test_no_unnecessary_imports_on_worker (GH#5862) crusaderky

• Start building pre-releases with cythonized scheduler (GH#5831) Charles Blackmon-Luca

• Do not mark tests xfail if they don’t come up in time (GH#5824) Florian Jetter

• Use gen_cluster where possible in test_dask_worker.py (GH#5842) Florian Jetter

• Generate junit report when pytest-timeout kills pytest (GH#5832) crusaderky

• Decrease timeout-minutes for GHA jobs (GH#5837) Florian Jetter

• Fix some timeouts (GH#5647) Florian Jetter

• Bump pre-release version to be greater than stable releases (GH#5816) Charles Blackmon-Luca

• Do not run schedule jobs on forks (GH#5821) Florian Jetter

• Remove pillow<9 pin in CI (GH#5775) Thomas Grainger

• Show scheduled test runs in report (GH#5812) Ian Rose

• Add obvious exclusions with pragma statement (GH#5801) Sarah Charlotte Johnson

• Add coverage exclusions for cli files (GH#5800) Sarah Charlotte Johnson

• Add pragma statements (GH#5749) Sarah Charlotte Johnson

• Remove pragma: no cover from distributed.cli.dask_ssh (GH#5809) Thomas Grainger

• Add pragma - worker.py, client.py, stealing.py (GH#5827) Sarah Charlotte Johnson

• Relax distributed / dask-core dependencies for pre-releases (GH#5802) Charles Blackmon-Luca

• Remove test_ucx_config_w_env_var flaky condition (GH#5765) Peter Andreas Entschev

2022.02.0

Released on February 11, 2022

Note

This is the last release with support for Python 3.7

Enhancements

• Update client.scheduler_info in wait_for_workers (GH#5785) Matthew Rocklin

• Increase robustness to TimeoutError during connect (GH#5096) Florian Jetter

• Respect KeyboardInterrupt in sync (GH#5758) Thomas Grainger

• Add workflow / recipe to generate Dask/distributed pre-releases (GH#5636) Charles Blackmon-Luca

• Review Scheduler / Worker display repr (GH#5746) crusaderky

• AMM: Graceful Worker Retirement (GH#5381) crusaderky

• AMM: tentatively stabilize flaky tests around worker pause (GH#5735) crusaderky

• AMM: speed up and stabilize test_memory (GH#5737) crusaderky

• Defer pandas import on worker in P2P shuffle (GH#5695) Gabe Joseph

Bug Fixes

• Fix for distributed.worker.memory.target=False and spill=0.7 (GH#5788) crusaderky

• Transition flight to missing if no who_has (GH#5653) Florian Jetter

Deprecations

• Remove deprecated ncores (GH#5780) crusaderky

• Deprecate registering plugins by class (GH#5699) Thomas Grainger

• Deprecate --nprocs option for dask-worker CLI (GH#5641) Bryan W. Weber

Documentation

• Fix imbalanced backticks (GH#5784) Matthias Bussonnier

Maintenance

• xfail test_worker_reconnects_mid_compute (GH#5797) crusaderky

• Fix linting CI build (GH#5794) James Bourbeau

• Update pre-commit versions (GH#5782) James Bourbeau

• Reactivate pytest_resourceleaks (GH#5771) crusaderky

• Set test assumption for test_client_timeout (GH#5790) Florian Jetter

• Remove client timeout from test_ucx_config_w_env_var (GH#5792) Florian Jetter

• Remove test_failed_worker_without_warning (GH#5789) Florian Jetter

• Fix longitudinal report (GH#5783) Ian Rose

• Fix flaky test_robust_to_bad_sizeof_estimates (GH#5753) crusaderky

• Revert “Pin coverage to 6.2 (GH#5716)” (GH#5770) Thomas Grainger

• Trigger test runs periodically to increases failure statistics (GH#5769) Florian Jetter

• More fault tolerant test report (GH#5732) Ian Rose

• Pin pillow<9 to work around torch incompatibility (GH#5755) Thomas Grainger

• Overhaul check_process_leak (GH#5739) crusaderky

• Fix flaky test_exit_callback test (GH#5713) Jim Crist-Harif

• Generate tests summary (GH#5710) crusaderky

• Upload different architectured pre-releases separately (GH#5741) Charles Blackmon-Luca

• Ignore non-test directories (GH#5720) Gabe Joseph

• Bump gpuCI PYTHON_VER to 3.9 (GH#5738) Charles Blackmon-Luca

• Regression: threads noted down before they start (GH#5796) crusaderky

2022.01.1

Released on January 28, 2022

New Features

• P2P shuffle skeleton (GH#5520) Gabe Joseph

Enhancements

• Fix <Task pending name='...' coro=<Client._handle_report()> (GH#5721) Thomas Grainger

• Add distributed.client.security-loader config (GH#5693) Jim Crist-Harif

• Avoid Client._handle_report cancelling itself on Client._close (GH#5672) Thomas Grainger

• Paused workers shouldn’t steal tasks (GH#5665) crusaderky

• Add option for timestamps from output of Node.get_logs (GH#4932) Charles Blackmon-Luca

• Don’t use time.time() or IOLoop.time() (GH#5661) crusaderky

Bug Fixes

• Raise plugin exceptions on Worker.start() (GH#4298) Peter Andreas Entschev

Documentation

• Fixing docstrings (GH#5696) Julia Signell

• Fix typo in Client.run docstring (GH#5687) Thomas Grainger

• Update client.py docstrings (GH#5670) Tim Harris

Maintenance

• Skip shuffle tests if pandas / dask.dataframe not installed (GH#5730) James Bourbeau

• Improve test coverage (GH#5655) Sarah Charlotte Johnson

• Test report improvements (GH#5714) Ian Rose

• P2P shuffle: ignore row order in tests (GH#5706) Gabe Joseph

• Fix flaky test_no_reconnect[--no-nanny] (GH#5686) Thomas Grainger

• Pin coverage to 6.2 (GH#5716) Thomas Grainger

• Check for new name of timeouts artifact and be more fault tolerant (GH#5707) Ian Rose

• Revisit rebalance unit tests (GH#5697) crusaderky

• Update comment in rearrange_by_column_p2p (GH#5701) James Bourbeau

• Update gpuCI RAPIDS_VER to 22.04 (GH#5676)

• Fix groupby test after meta requirements got stricter in Dask PR#8563 (GH#5694) Julia Signell

• Fix flaky test_close_gracefully and test_lifetime (GH#5677) crusaderky

• Fix flaky test_workspace_concurrency (GH#5690) crusaderky

• Fix flaky test_shuffle_extension.py::test_get_partition (GH#5689) Gabe Joseph

• Fix flaky test_dump_cluster_unresponsive_remote_worker (GH#5679) crusaderky

• Dump cluster state on all test failures (GH#5674) crusaderky

• Update license format (GH#5652) James Bourbeau

• Fix flaky test_drop_with_paused_workers_with_running_tasks_3_4 (GH#5673) crusaderky

• Do not raise an exception if the GitHub token cannot be found (GH#5668) Florian Jetter

2022.01.0

Released on January 14, 2022

New Features

• Task group stacked area chart (GH#5320) Ian Rose

• Support configuring TLS min/max version (GH#5594) Jim Crist-Harif

• Use asyncio for TCP/TLS comms (GH#5450) Jim Crist-Harif

Enhancements

• Close comm on CancelledError (GH#5656) crusaderky

• Don’t drop from the only running worker (GH#5626) crusaderky

• Transfer priority (GH#5625) crusaderky

• Add RPC call for getting task prefixes (GH#5617) Benjamin Zaitlen

• Long running occupancy (GH#5395) Florian Jetter

• Handle errors on individual workers in run/broadcast (GH#5590) crusaderky

• Allow work stealing in case there are heterogeneous resources for thief and victim (GH#5573) Florian Jetter

• Disable NVML monitoring on WSL (GH#5568) Charles Blackmon-Luca

Bug Fixes

• Ensure uniqueness of steal stimulus ID (GH#5620) Florian Jetter

• Fix KeyError: 'startstops' in performance report (GH#5608) Gabe Joseph

• Story timestamps can be slightly in the future (GH#5612) crusaderky

• Prevent RecursionError in Worker._to_dict (GH#5591) crusaderky

• Ensure distributed can be imported in thread (GH#5593) Jim Crist-Harif

Documentation

• Fix changelog section hyperlinks (GH#5638) Aneesh Nema

• Fix typo in unpublish_dataset example invocation (GH#5615) Deepyaman Datta

• Fix typo in test report badge in README (GH#5586) James Bourbeau

Maintenance

• Cosmetic changes to distributed.comm (GH#5657) crusaderky

• Consolidate broken comm testing utilities (GH#5654) James Bourbeau

• Fix concurrency assumptions for test_worker_reconnects_mid_compute (GH#5623) Florian Jetter

• Handle Bokeh 3.0 CDSView change (GH#5643) Bryan Van de Ven

• Use packaging rather than distutils to get version (GH#5624) Julia Signell

• XFAIL tls explicit comm close test on py3.7 (GH#5639) Jim Crist-Harif

• Mark some additional ucx-py tests for GPU (GH#5603) Charles Blackmon-Luca

• Rename ensure_default_get and add test (GH#5609) Naty Clementi

• Remove render_mode kwarg from boekh LabelSets (GH#5616) Garry O’Donnell

• Add lambda support to assert_worker_story (GH#5618) crusaderky

• Ignore file not found warning for timeout artifact (GH#5619) Florian Jetter

• Improved cluster state dump in @gen_cluster (GH#5592) crusaderky

• Work around SSL failures on MacOS CI (GH#5606) crusaderky

• Bump gpuCI CUDA_VER to 11.5 (GH#5604) Charles Blackmon-Luca

• assert_worker_story (GH#5598) crusaderky

• distributed.versions code refresh (GH#5600) crusaderky

• Updates to gpuCI and test_ucx_config_w_env_var (GH#5595) James Bourbeau

• Replace blacklist/whitelist with blocklist/allowlist (GH#5589) crusaderky

• Distributed test report (GH#5583) Ian Rose

• AMM: cosmetic tweaks (GH#5584) crusaderky

2021.12.0

Released on December 10, 2021

Enhancements

• Support pytest fixures and parametrize with gen_test (GH#5532) Fábio Rosado

• Allow idempotent scheduler plugins to be registered via the RPC (GH#5545) Jacob Tomlinson

• AMM logging (GH#5530) crusaderky

• Raise error if asyncssh isn’t installed when using SSHCluster (GH#5535) Fábio Rosado

• Allow None in UCX configuration schema (GH#5534) Fábio Rosado

• Add distributed.comm.ucx.create-cuda-context config (GH#5526) Peter Andreas Entschev

Bug fixes

• Allow unknown tasks to be stolen (GH#5572) Florian Jetter

• Further RecursionError fixes in recursive_to_repr (GH#5579) crusaderky

• Revisit recursive_to_dict (GH#5557) crusaderky

• Handle UCXUnreachable exception (GH#5556) Peter Andreas Entschev

Documentation

• Separate Coordination section in API docs (GH#5412) Gabe Joseph

• Improved documentation for processing state and paused workers (GH#4985) Maximilian Roos

• Fix typo in TaskGroupGraph.update_layout comment (GH#5536) Hristo Georgiev

• Update documentation for register_worker_plugin (GH#5533) crusaderky

Maintenance

• Mark test_gpu_monitoring_recent as flaky (GH#5540) Peter Andreas Entschev

• Await worker arrival in SSH test_nprocs (GH#5575) James Bourbeau

• AMM: Test that acquire-replicas of a task already in flight is a no-op (GH#5566) crusaderky

• Make sure artifacts are tagged with CI partition so they don’t race and overwrite each other (GH#5571) Ian Rose

• Minor refactoring and commentary in worker state machine (GH#5563) Florian Jetter

• Fix test_ucx_unreachable on UCX < 1.12 (GH#5562) Peter Andreas Entschev

• Bump Bokeh min version to 2.1.1 (GH#5548) Bryan Van de Ven

• Update gen_test tests to be more robust (GH#5551) James Bourbeau

• Skip test_ucx_unreachable if UCXUnreachable is unavailable (GH#5560) Peter Andreas Entschev

• Update gpuCI RAPIDS_VER to 22.02 (GH#5544)

• Add workflow to automate gpuCI updates (GH#5541) Charles Blackmon-Luca

• Actually support uvloop in distributed (GH#5531) Jim Crist-Harif

• Standardize UCX config separator to - (GH#5539) Peter Andreas Entschev

2021.11.2

Released on November 19, 2021

• Ensure cancelled error transition can properly release a key (GH#5528) Florian Jetter

• Refactor release key (GH#5507) Florian Jetter

• Fix deadlock caused by an erred task (executing->cancelled->error) (GH#5503) Florian Jetter

• Resolve KeyError-related deadlock (GH#5525) Florian Jetter

• Remove extra quotation in worker failure docs (GH#5518) James Bourbeau

• Ensure safe_sizeof warning is accurate (GH#5519) James Bourbeau

• Visualize cluster-wide memory usage over time (GH#5477) crusaderky

• AMM: redesign start/stop methods (GH#5476) crusaderky

• Preserve contextvars during comm offload (GH#5486) Gabe Joseph

• Deserialization: zero-copy merge subframes when possible (GH#5208) Gabe Joseph

• Add support for multiple workers per SSH connection (GH#5506) Jacob Tomlinson

• Client method to dump cluster state (GH#5470) Florian Jetter

2021.11.1

Released on November 8, 2021

• Revert “Avoid multiple blocking calls by gathering UCX frames” (GH#5505) Peter Andreas Entschev

2021.11.0

Released on November 5, 2021

• Fix cluster_info sync handling (GH#5488) Jim Crist-Harif

• Serialization family to preserve headers of the underlying dumps functions (GH#5380) Mads R. B. Kristensen

• Point users to Discourse (GH#5489) James Bourbeau

• Avoid multiple blocking calls by gathering UCX frames (GH#5487) Peter Andreas Entschev

• Update all UCX tests to use asyncio marker (GH#5484) Peter Andreas Entschev

• Register UCX close callback (GH#5474) Peter Andreas Entschev

• Use older version of pynvml.nvmlDeviceGetComputeRunningProcesses (GH#5469) Jacob Tomlinson

• Check for Futures from the wrong Client in gather (GH#5468) Gabe Joseph

• Fix performance_report when used with %%time or %%timeit magic (GH#5463) Erik Welch

• Scatter and replicate to avoid paused workers (GH#5441) crusaderky

• AMM to avoid paused workers (GH#5440) crusaderky

• Update changelog with LocalCluster host security note (GH#5462) Jim Crist-Harif

2021.10.0

Released on October 22, 2021

Note

This release fixed a potential security vulnerability relating to single-machine Dask clusters. Clusters started with dask.distributed.LocalCluster or dask.distributed.Client() (which defaults to using LocalCluster) would mistakenly configure their respective Dask workers to listen on external interfaces (typically with a randomly selected high port) rather than only on localhost. A Dask cluster created using this method AND running on a machine that has these ports exposed could be used by a sophisticated attacker to enable remote code execution. Users running on machines with standard firewalls in place should not be affected. This vulnerability is documented in CVE-2021-42343, and is fixed in this release (GH#5427). Thanks to Jean-Pierre van Riel for discovering and reporting the issue.

• Ensure resumed flight tasks are still fetched (GH#5426) Florian Jetter

• AMM high level documentation (GH#5456) crusaderky

• Provide stack for suspended coro in test timeout (GH#5446) Florian Jetter

• Handle UCXNotConnected error (GH#5449) Peter Andreas Entschev

• Don’t schedule tasks to paused workers (GH#5431) crusaderky

• Use pip install . instead of calling setup.py (GH#5442) Matthias Bussonnier

• Increase latency for stealing (GH#5390) Florian Jetter

• Type annotations for Worker and gen_cluster (GH#5438) crusaderky

• Ensure reconnecting workers do not loose required data (GH#5436) Florian Jetter

• Mark test_gather_dep* as xfail (GH#5432) crusaderky

• Remove zict-related skips (GH#5429) James Bourbeau

• Pass host through LocalCluster to workers (GH#5427) Jim Crist-Harif

• Fixes async warnings in UCX tests (GH#5396) Peter Andreas Entschev

• Resolve work stealing deadlock caused by race in move_task_confirm (GH#5379) Florian Jetter

• Add scroll to dashboard dropdown (GH#5418) Jacob Tomlinson

• Fix regression where unknown tasks were allowed to be stolen (GH#5392) Florian Jetter

• Enable mypy in CI 2/2 (GH#5348) crusaderky

• Rewrite test_client_timeout (GH#5397) crusaderky

• Simple SSHCluster example (GH#5349) Ray Bell

• Do not attempt to fetch keys which are no longer in flight (GH#5160) Florian Jetter

• Revisit Scheduler.add_plugin / Scheduler.remove_plugin (GH#5394) crusaderky

• Fix flaky test_WorkerPlugin_overwrite (GH#5398) crusaderky

• Active Memory Manager to use bulk comms (GH#5357) crusaderky

• Add coverage badge to README (GH#5382) James Bourbeau

• Mark test_stress_creation_and_deletion as xfail (GH#5393) James Bourbeau

• Mark test_worker_reconnects_mid_compute* tests as flaky (GH#5378) James Bourbeau

• Use new Dask docs theme (GH#5391) Jacob Tomlinson

• Remove pytest.mark.repeat from test_prometheus_collect_task_states (GH#5376) James Bourbeau

• Log original exception upon compute failure (GH#5387) Florian Jetter

• Add code coverage (GH#4670) James Bourbeau

• Fix zombie worker tasks after missing transition (GH#5316) Florian Jetter

• Add support for partial functions to iscoroutinefunction util (GH#5344) Michael Adkins

• Mark distributed/tests/test_client.py::test_profile_server as flaky (GH#5375) James Bourbeau

• Enable mypy in CI 1/2 (GH#5328) crusaderky

• Ensure dask-worker and dask-scheduler pick up preload configuration values (GH#5365) James Bourbeau

• Use dask-spec for SSHCluster (GH#5191) Charles Blackmon-Luca

• Update _cluster_info dict in __init__ (GH#5305) Jacob Tomlinson

• Use Dask temporary file utility (GH#5361) James Bourbeau

• Avoid deprecated random set sampling (GH#5360) James Bourbeau

• Add check for unsupported NVML metrics (GH#5343) Charles Blackmon-Luca

• Workers submit a reply to the scheduler if replica removal was rejected (GH#5356) Florian Jetter

• Pickle exception and traceback immediately (GH#5338) Mads R. B. Kristensen

• Reinstate: AMM ReduceReplicas to iterate only on replicated tasks (GH#5341) crusaderky

• Sync worker status to the scheduler; new ‘paused’ status (GH#5330) crusaderky

• Add pre-commit to environments (GH#5362) Ray Bell

• Worker State Machine Refactor: clean up dead handlers (GH#5359) crusaderky

• Bump RAPIDS_VER for gpuCI (GH#5358) Charles Blackmon-Luca

• Generate Cython HTML annotations (GH#5321) crusaderky

• Worker state machine refactor (GH#5046) Florian Jetter

• fsspec and s3fs git tips are incompatible (GH#5346) crusaderky

• Fix test_many_Progress and others (GH#5329) crusaderky

• Run multiple AMMs in parallel (GH#5339) crusaderky

• Enhance AMM docstrings (GH#5340) crusaderky

• Run pyupgrade in CI (GH#5327) crusaderky

• Fix typo in client side example foundations.rst (GH#5336) Genevieve Buckley

2021.09.1

Released on September 21, 2021

• Revert AMM ReduceReplicas and parallel AMMs updates (GH#5335) James Bourbeau

• Run multiple AMMs in parallel (GH#5315) crusaderky

• AMM ReduceReplicas to iterate only on replicated tasks (GH#5297) crusaderky

• Add type annotations to various functions within distributed.worker (GH#5290) Tom Forbes

• Mark test_ucx_config_w_env_var flaky on UCX < 1.11 (GH#5262) Peter Andreas Entschev

• Warn if CUDA context is created on incorrect device in UCX (GH#5308) Peter Andreas Entschev

• Remove redundant timeouts from test_client (GH#5314) crusaderky

• Allow Client to subscribe to events // Remote printing and warning (GH#5217) Florian Jetter

• Test pickle protocols 4 & 5 (GH#5313) jakirkham

• Fix-up test_pickle_empty (GH#5303) jakirkham

• Increase timeout for test_worker_reconnects_mid_compute_multiple_states_on_scheduler (GH#5304) Florian Jetter

• Add synced dict between cluster and scheduler to store cluster info (GH#5033) Jacob Tomlinson

• Update test_sub_submit_priority (GH#5301) James Bourbeau

• Revert “Add test setup fixture (GH#5242)” (GH#5300) James Bourbeau

• Fix flaky test_worker_reconnects_mid_compute (GH#5299) Florian Jetter

• Use gen_test in test_adaptive (GH#5298) crusaderky

• Increase worker.suspicious_counter threshold (GH#5228) Florian Jetter

• Active Memory Manager framework + discard excess replicas (GH#5111) crusaderky

• Add test setup fixture (GH#5242) James Bourbeau

2021.09.0

Released on September 3, 2021

• Fix add_plugin warnings (GH#5267) Doug Davis

• Add list around iterator in handle_missing_dep (GH#5285) Matthew Rocklin

• Jupyter-client 7 compatibility (GH#5286) Min RK

• Replace atop with blockwise (GH#5289) James Bourbeau

• Add pytest color to CI (GH#5276) James Bourbeau

• Fix test_map and others (GH#5278) crusaderky

• Use name argument with Scheduler.remove_plugin calls (GH#5260) Doug Davis

• Downgrade to jupyter_client 6 (GH#5273) crusaderky

• Migrate Security HTML repr to Jinja2 (GH#5264) Jacob Tomlinson

• Migrate ProcessInterface HTML repr to Jinja2 (GH#5263) Jacob Tomlinson

• Add support for diskless machines to system monitor (GH#5257) James Bourbeau

• Avoid during-iteration scheduler plugin changes (GH#5259) Doug Davis

• Remove GroupProgress scheduler plugin (GH#5256) James Bourbeau

• Properly check for ipv6 availability (GH#5255) crusaderky

• Improved IPv6 dask-worker support (GH#5197) Walt Woods

• Overwrite worker plugins (GH#5248) Matthew Rocklin

• Refactor scheduler plugins; store in a dictionary (GH#5120) Doug Davis

• Fix “then” -> “than” typo in docs (GH#5247) David Chudzicki

• Fix typo (remove extra verb “creates”) in docs (GH#5244) David Chudzicki

• Fix “fractiom” -> “fraction” typo in docstring (GH#5245) David Chudzicki

• Fix “schedulers” -> “scheduler” typo in docs (GH#5246) David Chudzicki

• Use non-histogram plots up to 100 workers (GH#5249) Matthew Rocklin

2021.08.1

Released on August 20, 2021

• Rename plots to fit in the labextension (GH#5239) Naty Clementi

• Log messages for CommClosedError now includes information about remote address (GH#5209) Florian Jetter

• Add target='_blank' for redirects of dashboard link (GH#5237) Naty Clementi

• Update computation code retrieval logic (GH#5236) James Bourbeau

• Minor polish on cfexecutor (GH#5233) crusaderky

• Use development version of dask in gpuCI build (GH#5232) James Bourbeau

• Use upstream dask.widgets (GH#5205) Jacob Tomlinson

• Fix flaky test_worker_reconnects_mid_compute (GH#5227) Florian Jetter

• Update WorkerPlugin docstring about usage of TaskState objects (GH#5226) Florian Jetter

• Worker Network Timeseries (GH#5129) Naty Clementi

• Add HTML Repr for ProcessInterface class and all its subclasses (GH#5181) Freyam Mehta

• Fix an issue where a reconnecting worker could cause an invalid transition (GH#5210) Florian Jetter

• Minor fixes for cfexecutor (GH#5177) Florian Jetter

• Add HTML Repr for Security class (GH#5178) Freyam Mehta

• Fix performance report sizing issue (GH#5213) Ian Rose

• Drop RMM compatibility code from RAPIDS < 0.11 (GH#5214) Peter Andreas Entschev

2021.08.0

Released on August 13, 2021

• Include addresses in closed comm repr (GH#5203) James Bourbeau

• Test nanny.environ precedence (GH#5204) Florian Jetter

• Migrating HTML reprs to jinja2 (GH#5188) Jacob Tomlinson

• Fix test_process_executor_kills_process flakyness (GH#5183) crusaderky

• Remove urllib3 as a dependency downloading preloads (GH#5199) Marcos Moyano

• Download preload urls in the Preload constructor (GH#5194) Marcos Moyano

• Avoid recursion error in profile.merge (GH#5195) Matthew Rocklin

• Add text exceptions to the Scheduler (GH#5148) Matthew Rocklin

• Use kwarg for Theme filename (GH#5190) Bryan Van de Ven

• Add a .git-ignore-revs file (GH#5187) Florian Jetter

• Replace not not with bool() (GH#5182) Jacob Tomlinson

• Resolve deadlock cause by transition error after fetching dependency (GH#5157) Florian Jetter

• Set z-index of data-table lower (GH#5175) Julia Signell

• Add no-worker - memory transition to scheduler (GH#5174) Florian Jetter

• Deprecate worker plugin overwrite policy (GH#5146) James Bourbeau

• Fix flaky tests in CI (GH#5168) crusaderky

• Instructions for jemalloc with brew on macOS (GH#4996) Gabe Joseph

• Bump RAPIDS_VER to 21.10 (GH#5165) Charles Blackmon-Luca

• Tweak verbiage around async functions (GH#5166) crusaderky

• Use Python 3 super() calls (GH#5167) crusaderky

• Support asynchronous tasks (GH#5151) Matthew Rocklin

• Rename total comm bytes and provide doc string (GH#5155) Florian Jetter

• Add GPU executor if GPU is present (GH#5123) Matthew Rocklin

• Fix RMM and UCX tests (GH#5158) Peter Andreas Entschev

• Remove excessive timeout of test_steal_during_task_deserialization (GH#5156) Florian Jetter

• Add gpuCI build script (GH#5147) Charles Blackmon-Luca

• Demote Worker.ensure_computing to function (GH#5153) Florian Jetter

2021.07.2

Released on July 30, 2021

• Fix a deadlock connected to task stealing and task deserialization (GH#5128) Florian Jetter

• Include maximum shard size in second to_frames method (GH#5145) Matthew Rocklin

• Minor dashboard style updates (GH#5143) Bryan Van de Ven

• Cap maximum shard size at the size of an integer (GH#5141) Matthew Rocklin

• Document automatic MALLOC_TRIM_THRESHOLD_ environment variable (GH#5139) James Bourbeau

• Mark ucx-py tests for GPU (GH#5133) Charles Blackmon-Luca

• Update individual profile plot sizing (GH#5131) James Bourbeau

• Handle NVMLError_Unknown in NVML diagnostics (GH#5121) Peter Andreas Entschev

• Unit tests to use a random port for the dashboard (GH#5060) crusaderky

• Ensure worker reconnect registers existing tasks properly (GH#5103) Florian Jetter

• Halve CI runtime! (GH#5074) crusaderky

• Add NannyPlugins (GH#5118) Matthew Rocklin

• Add WorkerNetworkBandwidth chart to dashboard (GH#5104) Naty Clementi

• Set nanny environment variables in config (GH#5098) Matthew Rocklin

• Read smaller frames to workaround OpenSSL bug (GH#5115) jakirkham

• Move UCX/RMM config variables to Distributed namespace (GH#4916) Charles Blackmon-Luca

• Allow ws(s) messages greater than 10Mb (GH#5110) Marcos Moyano

• Short-circuit root-ish check for many deps (GH#5113) Gabe Joseph

2021.07.1

Released on July 23, 2021

• Remove experimental feature warning from actors docs (GH#5108) James Bourbeau

• Keep dependents in worker dependency if TS is still known (GH#5034) Florian Jetter

• Add Scheduler.set_restrictions (GH#5101) Matthew Rocklin

• Make Actor futures awaitable and work with as_completed (GH#5092) Martin Durant

• Simplify test_secede_balances (GH#5071) Florian Jetter

• Computation class (GH#5001) Florian Jetter

• Some light dashboard cleanup (GH#5102) Bryan Van de Ven

• Don’t package tests (GH#5054) James Bourbeau

• Add pytest marker for GPU tests (GH#5023) Charles Blackmon-Luca

• Actor: don’t hold key references on workers (GH#4937) Gabe Joseph

• Collapse nav to hamburger sooner (GH#5094) Julia Signell

• Verify that actors survive pickling (GH#5086) Matthew Rocklin

• Reenable UCX-Py tests that used to segfault (GH#5076) Peter Andreas Entschev

• Better support ProcessPoolExecutors (GH#5063) Matthew Rocklin

• Simplify test_worker_heartbeat_after_cancel (GH#5067) Florian Jetter

• Avoid property validation in Bokeh (GH#5065) Matthew Rocklin

• Reduce default websocket frame size and make configurable (GH#5070) Ian Rose

• Disable pytest-timeout SIGALARM on MacOS (GH#5057) crusaderky

• rebalance() resilience to computations (GH#4968) crusaderky

• Improve CI stability (GH#5022) crusaderky

• Ensure heartbeats after cancelation do not raise KeyError s (GH#5053) Florian Jetter

• Add more useful exception message on TLS cert mismatch (GH#5040) Jacob Tomlinson

• Add bokeh mode parameter to performance reports (GH#5025) James Bourbeau

2021.07.0

Released on July 9, 2021

• Fix Nbytes jitter - less expensive (GH#5043) Naty Clementi

• Use native GH actions cancel feature (GH#5037) Florian Jetter

• Don’t require workers to report to scheduler if scheduler shuts down (GH#5032) Florian Jetter

• Add pandas to the list of checked packages for client.get_versions() (GH#5029) Ian Rose

• Move worker preload before scheduler address is set (GH#5024) Matthew Rocklin

• Fix flaky test_oversubscribing_leases (GH#5030) Florian Jetter

• Update scheduling policy docs for #4967 (GH#5018) Gabe Joseph

• Add echo handler to Server class (GH#5020) Matthew Rocklin

• Also include pngs when bundling package (GH#5016) Ian Rose

• Remove duplicated dashboard panes (GH#5017) Ian Rose

• Fix worker memory dashboard flickering (GH#4997) Naty Clementi

• Tabs on bottom left corner on dashboard (GH#5006) Naty Clementi

• Rename nbytes widgets (GH#4878) crusaderky

• Co-assign root-ish tasks (GH#4967) Gabe Joseph

• OSError tweaks (GH#5003) crusaderky

• Update imports to cudf.testing._utils (GH#5005) Peter Andreas Entschev

• Ensure shuffle split default durations uses proper prefix (GH#4991) Florian Jetter

• Follow up pyupgrade formatting (GH#4993) Florian Jetter

• Rename plot dropdown (GH#4992) James Bourbeau

• Pyupgrade (GH#4741) Florian Jetter

• Misc Sphinx tweaks (GH#4988) crusaderky

• No longer hold dependencies of erred tasks in memory #4918 Florian Jetter

• Add maximum shard size to config (GH#4986) Matthew Rocklin

• Ensure shuffle split operations are blacklisted from work stealing (GH#4964) Florian Jetter

• Add dropdown menu to access individual plots (GH#4984) Jacob Tomlinson

• Edited the path to scheduler.py (GH#4983) Freyam Mehta

• Task Group Graph Visualization (GH#4886) Naty Clementi

• Remove more internal references to deprecated utilities (GH#4971) James Bourbeau

• Restructure nbytes hover (GH#4952) Naty Clementi

• Except more errors in pynvml.nvmlInit() (GH#4970) gerrymanoim

• Add occupancy as individual plot (GH#4963) Naty Clementi

• Deprecate utilities which have moved to dask (GH#4966) James Bourbeau

• Ensure connectionpool does not leave comms if closed mid connect (GH#4951) Florian Jetter

• Add support for registering scheduler plugins from Client (GH#4808) Doug Davis

• Stealing dashboard fixes (GH#4948) Florian Jetter

• Allow requirements verification to be ignored when loading backends from entrypoints (GH#4961) Florian Jetter

• Add Log and Logs to API docs (GH#4946) James Bourbeau

• Support fixtures and pytest.mark.parametrize with gen_cluster (GH#4958) Gabe Joseph

2021.06.2

Released on June 22, 2021

• Revert refactor to utils.Log[s] and Cluster.get_logs (GH#4941) Charles Blackmon-Luca

• Use deprecation utility from Dask (GH#4924) James Bourbeau

• Add transition counter to Scheduler (GH#4934) Matthew Rocklin

• Remove nbytes_in_memory (GH#4930) Matthew Rocklin

2021.06.1

Released on June 18, 2021

• Fix deadlock in handle_missing_dep if additional replicas are available (GH#4929) Florian Jetter

• Add configuration to enable/disable NVML diagnostics (GH#4893) Peter Andreas Entschev

• Add scheduler log tab to performance reports (GH#4909) Charles Blackmon-Luca

• Add HTML repr to scheduler_info and incorporate into client and cluster reprs (GH#4857) Jacob Tomlinson

• Fix error state typo (GH#4898) James Bourbeau

• Allow actor exceptions to propagate (GH#4232) Martin Durant

• Remove importing apply from dask.compatibility (GH#4913) Elliott Sales de Andrade

• Use more informative default name for WorkerPlugin s (GH#4908) James Bourbeau

• Removed unused utility functions (GH#4911) James Bourbeau

• Locally rerun successfully completed futures (GH#4813) ArtinSarraf

• Forget erred tasks and fix deadlocks on worker (GH#4784) Florian Jetter

• Handle HTTPClientError in websocket connector (GH#4900) Marcos Moyano

• Update dask_cuda usage in SSHCluster docstring (GH#4894) James Bourbeau

• Remove tests for process_time and thread_time (GH#4895) James Bourbeau

• Flake8 config cleanup (GH#4888) Florian Jetter

• Don’t strip scheduler protocol when determining host (GH#4883) James Bourbeau

• Add more documentation on memory management (GH#4874) crusaderky

• Add range_query tests to NVML test suite (GH#4879) Charles Blackmon-Luca

• No longer cancel result future in async process when using timeouts (GH#4882) Florian Jetter

2021.06.0

Released on June 4, 2021

• Multiple worker executors (GH#4869) Mads R. B. Kristensen

• Ensure PyNVML works correctly when installed with no GPUs (GH#4873) Peter Andreas Entschev

• Show more in test summary (GH#4875) James Bourbeau

• Move SystemMonitor s GPU initialization back to constructor (GH#4866) Peter Andreas Entschev

• Mark test_server_comms_mark_active_handlers with pytest.mark.asyncio (GH#4876) James Bourbeau

• Who has has what html reprs v2 (GH#4865) Jacob Tomlinson

• O(1) rebalance (GH#4774) crusaderky

• Ensure repr and eq for cluster always works (GH#4799) Florian Jetter

2021.05.1

Released on May 28, 2021

• Drop usage of WhoHas & WhatHas from Client (GH#4863) jakirkham

• Ensure adaptive scaling is properly awaited and closed (GH#4720) Florian Jetter

• Fix WhoHas/ HasWhat async usage (GH#4860) Benjamin Zaitlen

• Add HTML reprs for Client.who_has and Client.has_what (GH#4853) Jacob Tomlinson

• Prevent accidentally starting multiple Worker s in the same process (GH#4852) crusaderky

• Add system tab to performance reports (GH#4561) Charles Blackmon-Luca

• Let servers close faster if there are no active handlers (GH#4805) Florian Jetter

• Fix UCX scrub config logging (GH#4850) Peter Andreas Entschev

• Ensure worker clients are closed (GH#3921) Florian Jetter

• Fix warning for attribute error when deleting a client (GH#4807) Florian Jetter

• Ensure exceptions are raised if workers are incorrectly started (GH#4733) Florian Jetter

• Update handling of UCX exceptions on endpoint closing (GH#4836) Peter Andreas Entschev

• Ensure busy workloads properly look up who_has (GH#4793) Florian Jetter

• Check distributed.scheduler.pickle in Scheduler.run_function (GH#4838) James Bourbeau

• Add performance_report to API docs (GH#4840) James Bourbeau

• Use dict _workers_dv in unordered use cases (GH#4826) jakirkham

• Bump pre-commit hook versions (GH#4835) James Bourbeau

• Do not mindlessly spawn workers when no memory limit is set (GH#4397) Torsten Wörtwein

• test_memory to use gen_cluster (GH#4811) crusaderky

• Increase timeout of gen_test to 30s (GH#4821) Florian Jetter

2021.05.0

Released on May 14, 2021

• Merge global annotations on the client (GH#4691) Mads R. B. Kristensen

• Add support for click 8 (GH#4810) James Bourbeau

• Add HTML reprs to some scheduler classes (GH#4795) James Bourbeau

• Use JupyterLab theme variables (GH#4796) Ian Rose

• Allow the dashboard to run on multiple ports (GH#4786) Jacob Tomlinson

• Remove release_dep from WorkerPlugin API (GH#4791) James Bourbeau

• Support for UCX 1.10+ (GH#4787) Peter Andreas Entschev

• Reduce complexity of test_gather_allow_worker_reconnect (GH#4739) Florian Jetter

• Fix doctests in utils.py (GH#4785) Jacob Tomlinson

• Ensure deps are actually logged in worker (GH#4753) Florian Jetter

• Add stacklevel keyword into performance_report() to allow for selecting calling code to be displayed (GH#4777) Nathan Danielsen

• Unregister worker plugin (GH#4748) Naty Clementi

• Fixes some pickling issues in the Cythonized Scheduler (GH#4768) jakirkham

• Improve graceful shutdown if nanny is involved (GH#4725) Florian Jetter

• Update cythonization in CI (GH#4764) James Bourbeau

• Use contextlib.nullcontext (GH#4763) James Bourbeau

• Cython fixes for MemoryState (GH#4761) jakirkham

• Fix errors in check_thread_leak (GH#4747) James Bourbeau

• Handle missing key case in report_on_key (GH#4755) jakirkham

• Drop temporary set variables s (GH#4758) jakirkham

2021.04.1

Released on April 23, 2021

• Avoid active_threads changing size during iteration (GH#4729) James Bourbeau

• Fix UnboundLocalError in AdaptiveCore.adapt() (GH#4731) Anderson Banihirwe

• Minor formatting updates for HTTP endpoints doc (GH#4736) James Bourbeau

• Unit test for metrics["memory"]=None (GH#4727) crusaderky

• Enable configuration of prometheus metrics namespace (GH#4722) Jacob Tomlinson

• Reintroduce weight function (GH#4723) James Bourbeau

• Add ready->memory to transitions in worker (GH#4728) Gil Forsyth

• Fix regressions in GH#4651 (GH#4719) crusaderky

• Add descriptions for UCX config options (GH#4683) Charles Blackmon-Luca

• Split RAM measure into dask keys/other old/other new (GH#4651) crusaderky

• Fix DeprecationWarning on Python 3.9 (GH#4717) George Sakkis

• ipython causes test_profile_nested_sizeof crash on windows (GH#4713) crusaderky

• Add iterate_collection argument to serialize (GH#4641) Richard J Zamora

• When closing Server, close all listeners (GH#4704) Florian Jetter

• Fix timeout in client.restart (GH#4690) Matteo De Wint

• Avoid repeatedly using the same worker on first task with quiet cluster (GH#4638) Doug Davis

• Grab func for finish case only if used (GH#4702) jakirkham

• Remove hostname check in test_dashboard (GH#4706) James Bourbeau

• Faster tests_semaphore::test_worker_dies (GH#4703) Florian Jetter

• Clean up test_dashboard (GH#4700) crusaderky

• Add timing information to TaskGroup (GH#4671) Matthew Rocklin

• Remove WSSConnector TLS presence check (GH#4695) Marcos Moyano

• Fix typo and remove unused time.time import (GH#4689) Hristo Georgiev

• Don’t initialize CUDA context in monitor (GH#4688) Charles Blackmon-Luca

• Add support for extra conn args for HTTP protocols (GH#4682) Marcos Moyano

• Adjust timings in test_threadpoolworkers (GH#4681) Florian Jetter

• Add GPU metrics to SystemMonitor (GH#4661) Charles Blackmon-Luca

• Removing dumps_msgpack() and loads_msgpack() (GH#4677) Mads R. B. Kristensen

• Expose worker SystemMonitor s to scheduler via RPC (GH#4657) Charles Blackmon-Luca

2021.04.0

Released on April 2, 2021

• Fix un-merged frames (GH#4666) Matthew Rocklin

• Add informative error message to install uvloop (GH#4664) Matthew Rocklin

• Remove incorrect comment regarding default LocalCluster creation (GH#4660) cameron16

• Treat empty/missing writeable as a no-op (GH#4659) jakirkham

• Avoid list mutation in pickle_loads (GH#4653) Matthew Rocklin

• Ignore OSError exception when scaling down (GH#4633) Gerald

• Add isort to pre-commit hooks, package resorting (GH#4647) Charles Blackmon-Luca

• Use powers-of-two when displaying RAM (GH#4649) crusaderky

• Support Websocket communication protocols (GH#4396) Marcos Moyano

• scheduler.py / worker.py code cleanup (GH#4626) crusaderky

• Update out-of-date references to config.yaml (GH#4643) Hristo Georgiev

• Suppress OSError on SpecCluster shutdown (GH#4567) Jacob Tomlinson

• Replace conda with mamba (GH#4585) crusaderky

• Expand documentation on pure functions (GH#4644) James Lamb

2021.03.1

Released on March 26, 2021

• Add standalone dashboard page for GPU usage (GH#4556) Jacob Tomlinson

• Handle stream is None case in TCP comm finalizer (GH#4631) James Bourbeau

• Include LIST_PICKLE in NumPy array serialization (GH#4632) James Bourbeau

• Rename annotation plugin in test_highlevelgraph.py (GH#4618) James Bourbeau

• UCX use nbytes instead of len (GH#4621) Mads R. B. Kristensen

• Skip NumPy and pandas tests if not importable (GH#4563) Ben Greiner

• Remove utils.shutting_down in favor of sys.is_finalizing (GH#4624) James Bourbeau

• Handle async clients when closing (GH#4623) Matthew Rocklin

• Drop log from remove_key_from_stealable (GH#4609) jakirkham

• Introduce events log length config option (GH#4615) Fabian Gebhart

• Upstream config serialization and inheritance (GH#4372) Jacob Tomlinson

• Add check to scheduler creation in SpecCluster (GH#4605) Jacob Tomlinson

• Make length of events deque configurable (GH#4604) Fabian Gebhart

• Add explicit fetch state to worker TaskState (GH#4470) Gil Forsyth

• Update develop.rst (GH#4603) Florian Jetter

• pickle_loads(): Handle empty memoryview (GH#4595) Mads R. B. Kristensen

• Switch documentation builds for PRs to readthedocs (GH#4599) James Bourbeau

• Track frame sizes along with frames (GH#4593) jakirkham

• Add support for a list of keys when using batch_size in client.map (GH#4592) Sultan Orazbayev

• If SpecCluster fails to start attempt to gracefully close out again (GH#4590) Jacob Tomlinson

• Multi-lock extension (GH#4503) Mads R. B. Kristensen

• Update PipInstall plugin command (GH#4584) James Bourbeau

• IPython magics: remove deprecated ioloop workarounds (GH#4530) Min RK

• Add GitHub actions workflow to cancel duplicate builds (GH#4581) James Bourbeau

• Remove outdated macOS build badge from README (GH#4576) James Bourbeau

• Dask master -> main (GH#4569) Julia Signell

• Drop support for Python 3.6 (GH#4390) James Bourbeau

• Add docstring for dashboard_link property (GH#4572) Doug Davis

• Change default branch from master to main (GH#4495) Julia Signell

• Msgpack handles extract serialize (GH#4531) Mads R. B. Kristensen

2021.03.0

Released on March 5, 2021

Note

This is the first release with support for Python 3.9 and the last release with support for Python 3.6

• tcp.write(): cast memoryview to byte itemsize (GH#4555) Mads R. B. Kristensen

• Refcount the thread_state.asynchronous flag (GH#4557) Mads R. B. Kristensen

• Python 3.9 (GH#4460) crusaderky

• Better bokeh defaults for dashboard (GH#4554) Benjamin Zaitlen

• Expose system monitor dashboard as individual plot for lab extension (GH#4540) Jacob Tomlinson

• Pass on original temp dir from nanny to worker (GH#4549) Martin Durant

• Serialize and split (GH#4541) Mads R. B. Kristensen

• Use the new HLG pack/unpack API in Dask (GH#4489) Mads R. B. Kristensen

• Handle annotations for culled tasks (GH#4544) Tom Augspurger

• Make sphinx autosummary and autoclass consistent (GH#4367) Casey Clements

• Move _transition* to SchedulerState (GH#4545) jakirkham

• Migrate from travis to GitHub actions (GH#4504) crusaderky

• Move new_task to SchedulerState (GH#4527) jakirkham

• Batch more Scheduler sends (GH#4526) jakirkham

• transition_memory_released and get_nbytes() optimizations (GH#4516) jakirkham

• Pin black pre-commit (GH#4533) James Bourbeau

• Read & write all frames in one pass (GH#4506) jakirkham

• Skip stream.write call for empty frames (GH#4507) jakirkham

• Prepend frame metadata header (GH#4505) jakirkham

• transition_processing_memory optimizations, etc. (GH#4487) jakirkham

• Attempt to get client from worker in Queue and Variable (GH#4490) James Bourbeau

• Use main branch for zict (GH#4499) jakirkham

• Use a callback to close TCP Comms, rather than check every time (GH#4453) Matthew Rocklin

2021.02.0

Released on February 5, 2021

• Bump minimum Dask to 2021.02.0 (GH#4486) James Bourbeau

• Update TaskState documentation about dependents attribute (GH#4440) Florian Jetter

• DOC: Autoreformat all functions docstrings (GH#4475) Matthias Bussonnier

• Use cached version of is_coroutine_function in stream handling to (GH#4481) Ian Rose

• Optimize transitions (GH#4451) jakirkham

• Create PULL_REQUEST_TEMPLATE.md (GH#4476) Ray Bell

• DOC: typo, directives ends with 2 colons :: (GH#4472) Matthias Bussonnier

• DOC: Proper numpydoc syntax for distributed/protocol/*.py (GH#4473) Matthias Bussonnier

• Update pytest.skip usage in test_server_listen (GH#4467) James Bourbeau

• Unify annotations (GH#4406) Ian Rose

• Added worker resources from config (GH#4456) Tom Augspurger

• Fix var name in worker validation func (GH#4457) Gil Forsyth

• Refactor task_groups & task_prefixes (GH#4452) jakirkham

• Use parent._tasks in heartbeat (GH#4450) jakirkham

• Refactor SchedulerState from Scheduler (GH#4365) jakirkham

2021.01.1

Released on January 22, 2021

• Make system monitor interval configurable (GH#4447) Matthew Rocklin

• Add uvloop config value (GH#4448) Matthew Rocklin

• Additional optimizations to stealing (GH#4445) jakirkham

• Give clusters names (GH#4426) Jacob Tomlinson

• Use worker comm pool in Semaphore (GH#4195) Florian Jetter

• Set runspec on all new tasks to avoid deadlocks (GH#4432) Florian Jetter

• Support TaskState objects in story methods (GH#4434) Matthew Rocklin

• Support missing event loop in Client.asynchronous (GH#4436) Matthew Rocklin

• Don’t require network to inspect tests (GH#4433) Matthew Rocklin

2021.01.0

Released on January 15, 2021

• Add time started to scheduler info (GH#4425) Jacob Tomlinson

• Log adaptive error (GH#4422) Jacob Tomlinson

• Xfail normalization tests (GH#4411) Jacob Tomlinson

• Use dumps_msgpack and loads_msgpack when packing high level graphs (GH#4409) Mads R. B. Kristensen

• Add nprocs auto option to dask-worker CLI (GH#4377) Jacob Tomlinson

• Type annotation of _reevaluate_occupancy_worker (GH#4398) jakirkham

• Type TaskGroup in active_states (GH#4408) jakirkham

• Fix test_as_current_is_thread_local (GH#4402) jakirkham

• Use list comprehensions to bind TaskGroup type (GH#4401) jakirkham

• Make tests pass after 2028 (GH#4403) Bernhard M. Wiedemann

• Fix compilation warnings, decide_worker now a C func, stealing improvements (GH#4375) jakirkham

• Drop custom __eq__ from Status (GH#4270) jakirkham

• test_performance_report: skip without bokeh (GH#4388) Bruno Pagani

• Nanny now respects dask settings from ctx mgr (GH#4378) Florian Jetter

• Better task duration estimates for outliers (GH#4213) selshowk

• Dask internal inherit config (GH#4364) Jacob Tomlinson

• Provide setup.py option to profile Cython code (GH#4362) jakirkham

• Optimizations of *State and Task* objects and stealing (GH#4358) jakirkham

• Cast SortedDict s to dict s in a few key places & other minor changes (GH#4355) jakirkham

• Use task annotation priorities for user-level priorities (GH#4354) James Bourbeau

• Added docs to highlevelgraph pack/unpack (GH#4352) Mads R. B. Kristensen

• Optimizations in notable functions used by transitions (GH#4351) jakirkham

• Silence exception when releasing futures on process shutdown (GH#4309) Benjamin Zaitlen

2020.12.0

Released on December 10, 2020

Highlights

• Switched to CalVer for versioning scheme.

• The scheduler can now receives Dask HighLevelGraph s instead of raw dictionary task graphs. This allows for a much more efficient communication of task graphs from the client to the scheduler.

• Added support for using custom Layer-level annotations like priority, retries, etc. with the dask.annotations context manager.

• Updated minimum supported version of Dask to 2020.12.0.

• Added many type annotations and updates to allow for gradually Cythonizing the scheduler.

All changes

• Some common optimizations across transitions (GH#4348) jakirkham

• Drop serialize extension (GH#4344) jakirkham

• Log duplicate workers in scheduler (GH#4338) Matthew Rocklin

• Annotation of some comm related methods in the Scheduler (GH#4341) jakirkham

• Optimize assert in validate_waiting (GH#4342) jakirkham

• Optimize decide_worker (GH#4332) jakirkham

• Store occupancy in _reevaluate_occupancy_worker (GH#4337) jakirkham

• Handle WorkerState memory_limit of None (GH#4335) jakirkham

• Use bint to annotate boolean attributes (GH#4334) jakirkham

• Optionally use offload executor in worker (GH#4307) Matthew Rocklin

• Optimize send_task_to_worker (GH#4331) jakirkham

• Optimize valid_workers (GH#4329) jakirkham

• Store occupancy in transition_waiting_processing (GH#4330) jakirkham

• Optimize get_comm_cost (GH#4328) jakirkham

• Use .pop(...) to remove key (GH#4327) jakirkham

• Use operator.attrgetter on WorkerState.address (GH#4324) jakirkham

• Annotate Task* objects for Cythonization (GH#4302) jakirkham

• Ensure retire_workers always return a dict (GH#4323) jakirkham

• Some Cython fixes for WorkerState (GH#4321) jakirkham

• Optimize WorkerState.__eq__ (GH#4320) jakirkham

• Swap order of TaskGroup and TaskPrefix (GH#4319) jakirkham

• Check traceback object can be unpickled (GH#4299) jakirkham

• Move TaskGroup & TaskPrefix before TaskState (GH#4318) jakirkham

• Remove empty test_highgraph.py file (GH#4313) James Bourbeau

• Ensure that retire_workers returns a dict (GH#4315) Matthew Rocklin

• Annotate WorkerState for Cythonization (GH#4294) jakirkham

• Close comm on low-level errors (GH#4239) jochen-ott-by

• Coerce new TaskState.nbytes value to int (GH#4311) jakirkham

• Remove offload try/except for thread_name_prefix keyword (GH#4308) James Bourbeau

• Fix pip install issue on CI (GH#4310) jakirkham

• Transmit Layer annotations to scheduler (GH#4279) Simon Perkins

• Ignores any compiled files generated by Cython (GH#4301) jakirkham

• Protect against missing key in get_metrics (GH#4300) Matthew Rocklin

• Provide option to build Distributed with Cython (GH#4292) jakirkham

• Set WorkerState.processing w/dict in clean (GH#4295) jakirkham

• Annotate ClientState for Cythonization (GH#4290) jakirkham

• Annotate check_idle_saturated for Cythonization (GH#4289) jakirkham

• Avoid flicker in TaskStream with “Scheduler is empty” message (GH#4284) Matthew Rocklin

• Make gather_dep robust to missing tasks (GH#4285) Matthew Rocklin

• Annotate extract_serialize (for Cythonization) (GH#4283) jakirkham

• Move nbytes from Worker’s state to TaskState (GH#4274) Gil Forsyth

• Drop extra type check in _extract_serialize (GH#4281) jakirkham

• Move Status to top-level import (GH#4280) Matthew Rocklin

• Add __hash__ and __eq__ for TaskState (GH#4278) jakirkham

• Add __hash__ and __eq__ for ClientState (GH#4276) jakirkham

• Collect report’s client_key``s in a ``list (GH#4275) jakirkham

• Precompute hash for WorkerState (GH#4271) jakirkham

• Use Status Enum in remove_worker (GH#4269) jakirkham

• Add aggregated topic logs and log_event method (GH#4230) James Bourbeau

• Find the set of workers instead of their frequency (GH#4267) jakirkham

• Use set.update to include other comms (GH#4268) jakirkham

• Support string timeouts in sync (GH#4266) James Bourbeau

• Use dask.utils.stringify() instead of distributed.utils.tokey() (GH#4255) Mads R. B. Kristensen

• Use .items() to walk through keys and values (GH#4261) jakirkham

• Simplify frame length packing in TCP write (GH#4257) jakirkham

• Comm/tcp listener: do not pass comm with failed handshake to comm_handler (GH#4240) jochen-ott-by

• Fuse steps in extract_serialize (GH#4254) jakirkham

• Drop test_sklearn (GH#4253) jakirkham

• Document task priority tie breaking (GH#4252) James Bourbeau

• __dask_distributed_pack__(): client argument (GH#4248) Mads R. B. Kristensen

• Configurable timeouts for worker_client and get_client (GH#4146) GeethanjaliEswaran

• Add dask/distributed versions to performance_report (GH#4249) Matthew Rocklin

• Update miniconda GitHub action (GH#4250) James Bourbeau

• UCX closing ignore error (GH#4236) Mads R. B. Kristensen

• Redirect to dask-worker cli documentation (GH#4247) Timost

• Upload file worker plugin (GH#4238) Ian Rose

• Create dependency TaskState as needed in gather_dep (GH#4241) Gil Forsyth

• Instantiate plugin if needed in register_worker_plugin (GH#4198) Julia Signell

• Allow actors to call actors on the same worker (GH#4225) Martin Durant

• Special case profile thread in leaked thread check (GH#4229) James Bourbeau

• Use intersection() on a set instead of dict_keys in update_graph (GH#4227) Mads R. B. Kristensen

• Communicate HighLevelGraphs directly to the Scheduler (GH#4140) Mads R. B. Kristensen

• Add get_task_metadata context manager (GH#4216) James Bourbeau

• Task state logs and data fix (GH#4206) Gil Forsyth

• Send active task durations from worker to scheduler (GH#4192) James Bourbeau

• Fix state check in test_close_gracefully (GH#4203) Gil Forsyth

• Avoid materializing layers in Client.compute() (GH#4196) Mads R. B. Kristensen

• Add TaskState metadata (GH#4191) James Bourbeau

• Fix regression in task stealing for already released keys (GH#4182) Florian Jetter

• Fix _graph_to_futures bug for futures-based dependencies (GH#4178) Richard J Zamora

• High level graph dumps/loads support (GH#4174) Mads R. B. Kristensen

• Implement pass HighLevelGraphs through _graph_to_futures (GH#4139) Mads R. B. Kristensen

• Support async preload click commands (GH#4170) James Bourbeau

• dask-worker cli memory limit option doc fix (GH#4172) marwan116

• Add TaskState to worker.py (GH#4107) Gil Forsyth

• Increase robustness of Semaphore.release (GH#4151) Lucas Rademaker

• Skip batched comm test win / tornado5 (GH#4166) Tom Augspurger

• Set Zict buffer target to maxsize when memory_target_fraction is False (GH#4156) Krishan Bhasin

• Add PipInstall WorkerPlugin (GH#3216) Matthew Rocklin

• Log KilledWorker events in the scheduler (GH#4157) Matthew Rocklin

• Fix test_gpu_metrics failure (GH#4154) jakirkham

2.30.1 - 2020-11-03

• Pin pytest-asyncio version (GH#4212) James Bourbeau

• Replace AsyncProcess exit handler by weakref.finalize (GH#4184) Peter Andreas Entschev

• Remove hard coded connect handshake timeouts (GH#4176) Florian Jetter

2.30.0 - 2020-10-06

• Support SubgraphCallable in str_graph() (GH#4148) Mads R. B. Kristensen

• Handle exceptions in BatchedSend (GH#4135) Tom Augspurger

• Fix for missing : in autosummary docs (GH#4143) Gil Forsyth

• Limit GPU metrics to visible devices only (GH#3810) Jacob Tomlinson

2.29.0 - 2020-10-02

• Use pandas.testing (GH#4138) jakirkham

• Fix a few typos (GH#4131) Pav A

• Return right away in Cluster.close if cluster is already closed (GH#4116) Tom Rochette

• Update async doc with example on .compute() vs client.compute() (GH#4137) Benjamin Zaitlen

• Correctly tear down LoopRunner in Client (GH#4112) Sergey Kozlov

• Simplify Client._graph_to_futures() (GH#4127) Mads R. B. Kristensen

• Cleanup new exception traceback (GH#4125) Krishan Bhasin

• Stop writing config files by default (GH#4123) Matthew Rocklin

2.28.0 - 2020-09-25

• Fix SSL connection_args for progressbar connect (GH#4122) jennalc

2.27.0 - 2020-09-18

• Fix registering a worker plugin with name arg (GH#4105) Nick Evans

• Support different remote_python paths on cluster nodes (GH#4085) Abdulelah Bin Mahfoodh

• Allow RuntimeError s when closing global clients (GH#4115) Matthew Rocklin

• Match pre-commit in dask (GH#4049) Julia Signell

• Update super usage (GH#4110) Poruri Sai Rahul

2.26.0 - 2020-09-11

• Add logging for adaptive start and stop (GH#4101) Matthew Rocklin

• Don’t close a nannied worker if it hasn’t yet started (GH#4093) Matthew Rocklin

• Respect timeouts when closing clients synchronously (GH#4096) Matthew Rocklin

• Log when downloading a preload script (GH#4094) Matthew Rocklin

• dask-worker --nprocs accepts negative values (GH#4089) Dror Speiser

• Support zero-worker clients (GH#4090) Matthew Rocklin

• Exclude fire-and-forget client from metrics (GH#4078) Tom Augspurger

• Drop Serialized.deserialize() method (GH#4073) jakirkham

• Add timeout= keyword to Client.wait_for_workers method (GH#4087) Matthew Rocklin

2.25.0 - 2020-08-28

• Update for black (GH#4081) Tom Augspurger

• Provide informative error when connecting an older version of Dask (GH#4076) Matthew Rocklin

• Simplify pack_frames (GH#4068) jakirkham

• Simplify frame_split_size (GH#4067) jakirkham

• Use list.insert to add prelude up front (GH#4066) jakirkham

• Graph helper text (GH#4064) Julia Signell

• Graph dashboard: Reset container data if task number is too large (GH#4056) Florian Jetter

• Ensure semaphore picks correct IOLoop for threadpool workers (GH#4060) Florian Jetter

• Add cluster log method (GH#4051) Jacob Tomlinson

• Cleanup more exception tracebacks (GH#4054) Krishan Bhasin

• Improve documentation of scheduler.locks options (GH#4062) Florian Jetter

2.24.0 - 2020-08-22

• Move toolbar to above and fix y axis (#4043) Julia Signell

• Make behavior clearer for how to get worker dashboard (#4047) Julia Signell

• Worker dashboard clean up (#4046) Julia Signell

• Add a default argument to the datasets and a possibility to override datasets (#4052) Nils Braun

• Discover HTTP endpoints (#3744) Martin Durant

2.23.0 - 2020-08-14

• Tidy up exception traceback in TCP Comms (GH#4042) Krishan Bhasin

• Angle on the x-axis labels (GH#4030) Mathieu Dugré

• Always set RMM’s strides in the header (GH#4039) jakirkham

• Fix documentation upload_file (GH#4038) Roberto Panai

• Update UCX tests for new handshake step (GH#4036) jakirkham

• Add test for informative errors in serialization cases (GH#4029) Matthew Rocklin

• Add compression, pickle protocol to comm contexts (GH#4019) Matthew Rocklin

• Make GPU plots robust to not having GPUs (GH#4008) Matthew Rocklin

• Update PendingDeprecationWarning with correct version number (GH#4025) Matthias Bussonnier

• Install PyTorch on CI (GH#4017) jakirkham

• Try getting cluster dashboard_link before asking scheduler (GH#4018) Matthew Rocklin

• Ignore writeable frames with builtin array (GH#4016) jakirkham

• Just extend frames2 by frames (GH#4015) jakirkham

• Serialize builtin array (GH#4013) jakirkham

• Use cuDF’s assert_eq (GH#4014) jakirkham

• Clear function cache whenever we upload a new file (GH#3993) Jack Xiaosong Xu

• Emit warning when assign/comparing string with Status Enum (GH#3875) Matthias Bussonnier

• Track mutable frames (GH#4004) jakirkham

• Improve bytes and bytearray serialization (GH#4009) jakirkham

• Fix memory histogram values in dashboard (GH#4006) Willi Rath

2.22.0 - 2020-07-31

• Only call frame_split_size when there are frames (GH#3996) jakirkham

• Fix failing test_bandwidth (GH#3999) jakirkham

• Handle sum of memory percentage when memory_limit is 0 (GH#3984) Julia Signell

• Drop msgpack pre-0.5.2 compat code (GH#3977) jakirkham

• Revert to localhost for local IP if no network available (GH#3991) Matthew Rocklin

• Add missing backtick in inline directive. (GH#3988) Matthias Bussonnier

• Warn when threads_per_worker is set to zero (GH#3986) Julia Signell

• Use memoryview in unpack_frames (GH#3980) jakirkham

• Iterate over list of comms (GH#3959) Matthew Rocklin

• Streamline pack_frames/unpack_frames frames (GH#3973) jakirkham

• Always attempt to create dask-worker-space folder and continue if it exists (GH#3972) Jendrik Jördening

• Use merge_frames with host memory only (GH#3971) jakirkham

• Simplify pack_frames_prelude (GH#3961) jakirkham

• Use continuation prompt for proper example parsing (GH#3966) Matthias Bussonnier

• Ensure writable frames (GH#3967) jakirkham

2.21.0 - 2020-07-17

• Fix data replication error (GH#3963) Andrew Fulton

• Treat falsey local directory as None (GH#3964) Tom Augspurger

• Unpin numpydoc now that 1.1 is released (GH#3957) Gil Forsyth

• Error hard when Dask has mismatched versions or lz4 installed (GH#3936) Matthew Rocklin

• Skip coercing to bytes in merge_frames (GH#3960) jakirkham

• UCX: reuse endpoints in order to fix NVLINK issue (GH#3953) Mads R. B. Kristensen

• Optionally use pickle5 (GH#3849) jakirkham

• Update time per task chart with filtering and pie (GH#3933) Benjamin Zaitlen

• UCX: explicit shutdown message (GH#3950) Mads R. B. Kristensen

• Avoid too aggressive retry of connections (GH#3944) Matthias Bussonnier

• Parse timeouts in Client.sync (GH#3952) Matthew Rocklin

• Synchronize on non-trivial CUDA frame transmission (GH#3949) jakirkham

• Serialize memoryview with shape and format (GH#3947) jakirkham

• Move scheduler_comm into Cluster.__init__ (GH#3945) Matthew Rocklin

2.20.0 - 2020-07-02

• Link issue on using async with executor_submit (GH#3939) jakirkham

• Make dashboard server listens on all IPs by default even when interface is set explicitly (GH#3941) Loïc Estève

• Update logic for worker removal in check ttl (GH#3927) Benjamin Zaitlen

• Close a created cluster quietly (GH#3935) Matthew Rocklin

• Ensure Worker.run* handles kwargs correctly (GH#3937) jakirkham

• Restore Scheduler.time_started for Dask Gateway (GH#3934) Tom Augspurger

• Fix exception handling in _wait_until_connected (GH#3912) Alexander Clausen

• Make local directory if it does not exist (GH#3928) Matthew Rocklin

• Install vanilla status route if bokeh dependency is not satisfied (GH#3844) joshreback

• Make Worker.delete_data sync (GH#3922) Peter Andreas Entschev

• Fix ensure_bytes import location (GH#3919) jakirkham

• Fix race condition in repeated calls to cluster.adapt() (GH#3915) Jacob Tomlinson

2.19.0 - 2020-06-19

• Notify worker plugins when a task is released (GH#3817) Nick Evans

• Update heartbeat checks in scheduler (GH#3896) Benjamin Zaitlen

• Make encryption default if Security is given arguments (GH#3887) Matthew Rocklin

• Show cpu_fraction on hover for dashboard workers circle plot. (GH#3906) Loïc Estève

• Prune virtual client on variable deletion (GH#3910) Marco Neumann

• Fix total aggregated metrics in dashboard (GH#3897) Loïc Estève

• Support Bokeh 2.1 (GH#3904) Matthew Rocklin

• Update related-work.rst (GH#3889) DomHudson

• Skip test_pid_file in older versions of Python (GH#3888) Matthew Rocklin

• Replace stream= with comm= in handlers (GH#3860) Julien Jerphanion

• Check hosts for None value in SSH cluster. (GH#3883) Matthias Bussonnier

• Allow dictionaries in security= keywords (GH#3874) Matthew Rocklin

• Use pickle protocol 5 with NumPy object arrays (GH#3871) jakirkham

• Cast any frame to uint8 (same type as bytes) (GH#3870) jakirkham

• Use Enum for worker, scheduler and nanny status. (GH#3853) Matthias Bussonnier

• Drop legacy buffer_interface assignment (GH#3869) jakirkham

• Drop old frame splitting in NumPy serialization (GH#3868) jakirkham

• Drop no longer needed local import pickle (GH#3865) jakirkham

• Fix typo in feed’s log message (GH#3867) jakirkham

• Tidy pickle (GH#3866) jakirkham

• Handle empty times in task stream (GH#3862) Benjamin Zaitlen

• Change asyncssh objects to sphinx references (GH#3861) Jacob Tomlinson

• Improve SSHCluster docstring for connect_options (GH#3859) Jacob Tomlinson

• Validate address parameter in client constructor (GH#3842) joshreback

• Use SpecCluster name in worker names (GH#3855) Loïc Estève

• Allow async add_worker and remove_worker plugin methods (GH#3847) James Bourbeau

2.18.0 - 2020-06-05

• Merge frames in deserialize_bytes (GH#3639) John Kirkham

• Allow SSHCluster to take a list of connect_options (GH#3854) Jacob Tomlinson

• Add favicon to performance report (GH#3852) Jacob Tomlinson

• Add dashboard plots for the amount of time spent per key and for transfer/serialization (GH#3792) Benjamin Zaitlen

• Fix variable name in journey of a task documentation (GH#3840) Matthias Bussonnier

• Fix typo in journey of a task doc (GH#3838) James Bourbeau

• Register dask_cudf serializers (GH#3832) John Kirkham

• Fix key check in rebalance missing keys (GH#3834) Jacob Tomlinson

• Allow collection of partial profile information in case of exceptions (GH#3773) Florian Jetter

2.17.0 - 2020-05-26

• Record the time since the last run task on the scheduler (GH#3830) Matthew Rocklin

• Set colour of nbytes pane based on thresholds (GH#3805) Krishan Bhasin

• Include total number of tasks in the performance report (GH#3822) Abdulelah Bin Mahfoodh

• Allow to pass in task key strings in the worker restrictions (GH#3826) Nils Braun

• Control de/ser offload (GH#3793) Martin Durant

• Parse timeout parameters in Variable/Event/Lock to support text timeouts (GH#3825) Nils Braun

• Don’t send empty dependencies (GH#3423) Jakub Beránek

• Add distributed Dask Event that mimics threading.Event (GH#3821) Nils Braun

• Enhance VersionMismatchWarning messages (GH#3786) Abdulelah Bin Mahfoodh

• Support Pickle’s protocol 5 (GH#3784) jakirkham

• Replace utils.ignoring with contextlib.suppress (GH#3819) Nils Braun

• Make re-creating conda environments from the CI output easier (GH#3816) Lucas Rademaker

• Add prometheus metrics for semaphore (GH#3757) Lucas Rademaker

• Fix worker plugin called with superseded transition (GH#3812) Nick Evans

• Add retries to server listen (GH#3801) Jacob Tomlinson

• Remove commented out lines from scheduler.py (GH#3803) James Bourbeau

• Fix RuntimeWarning for never awaited coroutine when using distributed.Semaphore (GH#3713) Florian Jetter

• Fix profile thread leakage during test teardown on some platforms (GH#3795) Florian Jetter

• Await self before handling comms (GH#3788) Matthew Rocklin

• Fix typo in Cluster docstring (GH#3787) Scott Sanderson

2.16.0 - 2020-05-08

• Client.get_dataset to always create Futures attached to itself (GH#3729) crusaderky

• Remove dev-requirements since it is unused (GH#3782) Julia Signell

• Use bokeh column for /system instead of custom css (GH#3781) Julia Signell

• Attempt to fix test_preload_remote_module on windows (GH#3775) James Bourbeau

• Fix broadcast for TLS comms (GH#3766) Florian Jetter

• Don’t validate http preloads locally (GH#3768) Rami Chowdhury

• Allow range of ports to be specified for Workers (GH#3704) James Bourbeau

• Add UCX support for RDMACM (GH#3759) Peter Andreas Entschev

• Support web addresses in preload (GH#3755) Matthew Rocklin

2.15.2 - 2020-05-01

• Connect to dashboard when address provided (GH#3758) Tom Augspurger

• Move test_gpu_metrics test (GH#3721) Tom Augspurger

• Nanny closing worker on KeyboardInterrupt (GH#3747) Mads R. B. Kristensen

• Replace OrderedDict with dict in scheduler (GH#3740) Matthew Rocklin

• Fix exception handling typo (GH#3751) Jonas Haag

2.15.1 - 2020-04-28

• Ensure BokehTornado uses prefix (GH#3746) James Bourbeau

• Warn if cluster closes before starting (GH#3735) Matthew Rocklin

• Memoryview serialisation (GH#3743) Martin Durant

• Allows logging config under distributed key (GH#2952) Dillon Niederhut

2.15.0 - 2020-04-24

• Reinstate support for legacy @gen_cluster functions (GH#3738) crusaderky

• Relax NumPy requirement in UCX (GH#3731) jakirkham

• Add Configuration Schema (GH#3696) Matthew Rocklin

• Reuse CI scripts for local installation process (GH#3698) crusaderky

• Use PeriodicCallback class from tornado (GH#3725) James Bourbeau

• Add remote_python option in ssh cmd (GH#3709) Abdulelah Bin Mahfoodh

• Configurable polling interval for cluster widget (GH#3723) Julia Signell

• Fix copy-paste in docs (GH#3728) Julia Signell

• Replace gen.coroutine with async-await in tests (GH#3706) crusaderky

• Fix flaky test_oversubscribing_leases (GH#3726) Florian Jetter

• Add batch_size to Client.map (GH#3650) Tom Augspurger

• Adjust semaphore test timeouts (GH#3720) Florian Jetter

• Dask-serialize dicts longer than five elements (GH#3689) Richard J Zamora

• Force threads_per_worker (GH#3715) crusaderky

• Idempotent semaphore acquire with retries (GH#3690) Florian Jetter

• Always use readinto in TCP (GH#3711) jakirkham

• Avoid DeprecationWarning from pandas (GH#3712) Tom Augspurger

• Allow modification of distributed.comm.retry at runtime (GH#3705) Florian Jetter

• Do not log an error on unset variable delete (GH#3652) Jonathan J. Helmus

• Add remote_python keyword to the new SSHCluster (GH#3701) Abdulelah Bin Mahfoodh

• Replace Example with Examples in docstrings (GH#3697) Matthew Rocklin

• Add Cluster __enter__ and __exit__ methods (GH#3699) Matthew Rocklin

• Fix propagating inherit config in SSHCluster for non-bash shells (GH#3688) Abdulelah Bin Mahfoodh

• Add Client.wait_to_workers to Client autosummary table (GH#3692) James Bourbeau

• Replace Bokeh Server with Tornado HTTPServer (GH#3658) Matthew Rocklin

• Fix dask-ssh after removing local-directory from dask_scheduler cli (GH#3684) Abdulelah Bin Mahfoodh

• Support preload modules in Nanny (GH#3678) Matthew Rocklin

• Refactor semaphore internals: make _get_lease synchronous (GH#3679) Lucas Rademaker

• Don’t make task graphs too big (GH#3671) Martin Durant

• Pass through connection/listen_args as splatted keywords (GH#3674) Matthew Rocklin

• Run preload at import, start, and teardown (GH#3673) Matthew Rocklin

• Use relative URL in scheduler dashboard (GH#3676) Nicholas Smith

• Expose Security object as public API (GH#3675) Matthew Rocklin

• Add zoom tools to profile plots (GH#3672) James Bourbeau

• Update Scheduler.rebalance return value when data is missing (GH#3670) James Bourbeau

2.14.0 - 2020-04-03

• Enable more UCX tests (GH#3667) jakirkham

• Remove openssl 1.1.1d pin for Travis (GH#3668) Jonathan J. Helmus

• More documentation for Semaphore (GH#3664) Florian Jetter

• Get CUDA context to finalize Numba DeviceNDArray (GH#3666) jakirkham

• Add Resources option to get_task_stream and call output_file (GH#3653) Prasun Anand

• Add Semaphore extension (GH#3573) Lucas Rademaker

• Replace ncores with nthreads in work stealing tests (GH#3615) James Bourbeau

• Clean up some test warnings (GH#3662) Matthew Rocklin

• Write “why killed” docs (GH#3596) Martin Durant

• Update Python version checking (GH#3660) James Bourbeau

• Add newlines to ensure code formatting for retire_workers (GH#3661) Rami Chowdhury

• Clean up performance report test (GH#3655) Matthew Rocklin

• Avoid diagnostics time in performance report (GH#3654) Matthew Rocklin

• Introduce config for default task duration (GH#3642) Gabriel Sailer

• UCX simplify receiving frames in comm (GH#3651) jakirkham

• Bump checkout GitHub action to v2 (GH#3649) James Bourbeau

• Handle exception in faulthandler (GH#3646) Jacob Tomlinson

• Add prometheus metric for suspicious tasks (GH#3550) Gabriel Sailer

• Remove local-directory keyword (GH#3620) Prasun Anand

• Don’t create output Futures in Client when there are mixed Client Futures (GH#3643) James Bourbeau

• Add link to contributing.md (GH#3621) Prasun Anand

• Update bokeh dependency in CI builds (GH#3637) James Bourbeau

2.13.0 - 2020-03-25

• UCX synchronize default stream only on CUDA frames (GH#3638) Peter Andreas Entschev

• Add as_completed.clear method (GH#3617) Matthew Rocklin

• Drop unused line from pack_frames_prelude (GH#3634) John Kirkham

• Add logging message when closing idle dask scheduler (GH#3632) Matthew Rocklin

• Include frame lengths of CUDA objects in header["lengths"] (GH#3631) John Kirkham

• Ensure Client connection pool semaphore attaches to the Client event loop (GH#3546) James Bourbeau

• Remove dead stealing code (GH#3619) Florian Jetter

• Check nbytes and types before reading data (GH#3628) John Kirkham

• Ensure that we don’t steal blacklisted fast tasks (GH#3591) Florian Jetter

• Support async Listener.stop functions (GH#3613) Matthew Rocklin

• Add str/repr methods to as_completed (GH#3618) Matthew Rocklin

• Add backoff to comm connect attempts. (GH#3496) Matthias Urlichs

• Make Listeners awaitable (GH#3611) Matthew Rocklin

• Increase number of visible mantissas in dashboard plots (GH#3585) Scott Sievert

• Pin openssl to 1.1.1d for Travis (GH#3602) Jacob Tomlinson

• Replace tornado.queues with asyncio.queues (GH#3607) James Bourbeau

• Remove dill from CI environments (GH#3608) Loïc Estève

• Fix linting errors (GH#3604) James Bourbeau

• Synchronize default CUDA stream before UCX send/recv (GH#3598) Peter Andreas Entschev

• Add configuration for Adaptive arguments (GH#3509) Gabriel Sailer

• Change Adaptive docs to reference adaptive_target (GH#3597) Julia Signell

• Optionally compress on a frame-by-frame basis (GH#3586) Matthew Rocklin

• Add Python version to version check (GH#3567) James Bourbeau

• Import tlz (GH#3579) John Kirkham

• Pin numpydoc to avoid double escaped * (GH#3530) Gil Forsyth

• Avoid performance_report crashing when a worker dies mid-compute (GH#3575) Krishan Bhasin

• Pin bokeh in CI builds (GH#3570) James Bourbeau

• Disable fast fail on GitHub Actions Windows CI (GH#3569) James Bourbeau

• Fix typo in Client.shutdown docstring (GH#3562) John Kirkham

• Add local_directory option to dask-ssh (GH#3554) Abdulelah Bin Mahfoodh

2.12.0 - 2020-03-06

• Update TaskGroup remove logic (GH#3557) James Bourbeau

• Fix-up CuPy sparse serialization (GH#3556) John Kirkham

• API docs for LocalCluster and SpecCluster (GH#3548) Tom Augspurger

• Serialize sparse arrays (GH#3545) John Kirkham

• Allow tasks with restrictions to be stolen (GH#3069) Stan Seibert

• Use UCX default configuration instead of raising (GH#3544) Peter Andreas Entschev

• Support using other serializers with register_generic (GH#3536) John Kirkham

• DOC: update to async await (GH#3543) Tom Augspurger

• Use pytest.raises in test_ucx_config.py (GH#3541) John Kirkham

• Fix/more ucx config options (GH#3539) Benjamin Zaitlen

• Update heartbeat CommClosedError error handling (GH#3529) James Bourbeau

• Use makedirs when constructing local_directory (GH#3538) John Kirkham

• Mark None as MessagePack serializable (GH#3537) John Kirkham

• Mark bool as MessagePack serializable (GH#3535) John Kirkham

• Use ‘temporary-directory’ from dask.config for Nanny’s directory (GH#3531) John Kirkham

• Add try-except around getting source code in performance report (GH#3505) Matthew Rocklin

• Fix typo in docstring (GH#3528) Davis Bennett

• Make work stealing callback time configurable (GH#3523) Lucas Rademaker

• RMM/UCX Config Flags (GH#3515) Benjamin Zaitlen

• Revise develop-docs: conda env example (GH#3406) Darren Weber

• Remove import ucp from the top of ucx.py (GH#3510) Peter Andreas Entschev

• Rename logs to get_logs (GH#3473) Jacob Tomlinson

• Stop keep alives when worker reconnecting to the scheduler (GH#3493) Jacob Tomlinson

2.11.0 - 2020-02-19

• Add dask serialization of CUDA objects (GH#3482) John Kirkham

• Suppress cuML ImportError (GH#3499) John Kirkham

• Msgpack 1.0 compatibility (GH#3494) James Bourbeau

• Register cuML serializers (GH#3485) John Kirkham

• Check exact equality for worker state (GH#3483) Brett Naul

• Serialize 1-D, contiguous, uint8 CUDA frames (GH#3475) John Kirkham

• Update NumPy array serialization to handle non-contiguous slices (GH#3474) James Bourbeau

• Propose fix for collection based resources docs (GH#3480) Chris Roat

• Remove --verbose flag from CI runs (GH#3484) Matthew Rocklin

• Do not duplicate messages in scheduler report (GH#3477) Jakub Beránek

• Register Dask cuDF serializers (GH#3478) John Kirkham

• Add support for Python 3.8 (GH#3249) James Bourbeau

• Add last seen column to worker table and highlight errant workers (GH#3468) kaelgreco

• Change default value of local_directory from empty string to None (GH#3441) condoratberlin

• Clear old docs (GH#3458) Matthew Rocklin

• Change default multiprocessing behavior to spawn (GH#3461) Matthew Rocklin

• Split dashboard host on additional slashes to handle inproc (GH#3466) Jacob Tomlinson

• Update locality.rst (GH#3470) Dustin Tindall

• Minor gen.Return cleanup (GH#3469) James Bourbeau

• Update comparison logic for worker state (GH#3321) rockwellw

• Update minimum tblib version to 1.6.0 (GH#3451) James Bourbeau

• Add total row to workers plot in dashboard (GH#3464) Julia Signell

• Workaround RecursionError on profile data (GH#3455) Tom Augspurger

• Include code and summary in performance report (GH#3462) Matthew Rocklin

• Skip test_open_close_many_workers on Python 3.6 (GH#3459) Matthew Rocklin

• Support serializing/deserializing rmm.DeviceBuffer s (GH#3442) John Kirkham

• Always add new TaskGroup to TaskPrefix (GH#3322) James Bourbeau

• Rerun black on the code base (GH#3444) John Kirkham

• Ensure __causes__ s of exceptions raised on workers are serialized (GH#3430) Alex Adamson

• Adjust numba.cuda import and add check (GH#3446) John Kirkham

• Fix name of Numba serialization test (GH#3447) John Kirkham

• Checks for command parameters in ssh2 (GH#3078) Peter Andreas Entschev

• Update worker_kwargs description in LocalCluster constructor (GH#3438) James Bourbeau

• Ensure scheduler updates task and worker states after successful worker data deletion (GH#3401) James Bourbeau

• Avoid loop= keyword in asyncio coordination primitives (GH#3437) Matthew Rocklin

• Call pip as a module to avoid warnings (GH#3436) Cyril Shcherbin

• Add documentation of parameters in coordination primitives (GH#3434) Søren Fuglede Jørgensen

• Replace tornado.locks with asyncio for Events/Locks/Conditions/Semaphore (GH#3397) Matthew Rocklin

• Remove object from class hierarchy (GH#3432) Anderson Banihirwe

• Add dashboard_link property to Client (GH#3429) Jacob Tomlinson

• Allow memory monitor to evict data more aggressively (GH#3424) fjetter

• Make _get_ip return an IP address when defaulting (GH#3418) Pierre Glaser

• Support version checking with older versions of Dask (GH#3390) Igor Gotlibovych

• Add Mac OS build to CI (GH#3358) James Bourbeau

2.10.0 - 2020-01-28

• Fixed ZeroDivisionError in dashboard when no workers were present (GH#3407) James Bourbeau

• Respect the dashboard-prefix when redirecting from the root (GH#3387) Chrysostomos Nanakos

• Allow enabling / disabling work-stealing after the cluster has started (GH#3410) John Kirkham

• Support *args and **kwargs in offload (GH#3392) Matthew Rocklin

• Add lifecycle hooks to SchedulerPlugin (GH#3391) Matthew Rocklin

2.9.3 - 2020-01-17

• Raise RuntimeError if no running loop (GH#3385) James Bourbeau

• Fix get_running_loop import (GH#3383) James Bourbeau

• Get JavaScript document location instead of window and handle proxied url (GH#3382) Jacob Tomlinson

2.9.2 - 2020-01-16

• Move Windows CI to GitHub Actions (GH#3373) Jacob Tomlinson

• Add client join and leave hooks (GH#3371) Jacob Tomlinson

• Add cluster map dashboard (GH#3361) Jacob Tomlinson

• Close connection comm on retry (GH#3365) James Bourbeau

• Fix scheduler state in case of worker name collision (GH#3366) byjott

• Add --worker-class option to dask-worker CLI (GH#3364) James Bourbeau

• Remove locale check that fails on OS X (GH#3360) Jacob Tomlinson

• Rework version checking (GH#2627) Matthew Rocklin

• Add websocket scheduler plugin (GH#3335) Jacob Tomlinson

• Return task in dask-worker on_signal function (GH#3354) James Bourbeau

• Fix failures on mixed integer/string worker names (GH#3352) Benedikt Reinartz

• Avoid calling nbytes multiple times when sending data (GH#3349) Markus Mohrhard

• Avoid setting event loop policy if within IPython kernel and no running event loop (GH#3336) Mana Borwornpadungkitti

• Relax intermittent failing test_profile_server (GH#3346) Matthew Rocklin

2.9.1 - 2019-12-27

• Add lock around dumps_function cache (GH#3337) Matthew Rocklin

• Add setuptools to dependencies (GH#3320) James Bourbeau

• Use TaskPrefix.name in Graph layout (GH#3328) Matthew Rocklin

• Add missing “ in performance report example (GH#3329) John Kirkham

• Add performance report docs and color definitions to docs (GH#3325) Benjamin Zaitlen

• Switch startstops to dicts and add worker name to transfer (GH#3319) Jacob Tomlinson

• Add plugin entry point for out-of-tree comms library (GH#3305) Patrick Sodré

• All scheduler task states in prometheus (GH#3307) fjetter

• Use worker name in logs (GH#3309) Stephan Erb

• Add TaskGroup and TaskPrefix scheduler state (GH#3262) Matthew Rocklin

• Update latencies with heartbeats (GH#3310) fjetter

• Update inlining Futures in task graph in Client._graph_to_futures (GH#3303) James Bourbeau

• Use hostname as default IP address rather than localhost (GH#3308) Matthew Rocklin

• Clean up flaky test_nanny_throttle (GH#3295) Tom Augspurger

• Add lock to scheduler for sensitive operations (GH#3259) Matthew Rocklin

• Log address for each of the Scheduler listerners (GH#3306) Matthew Rocklin

• Make ConnectionPool.close asynchronous (GH#3304) Matthew Rocklin

2.9.0 - 2019-12-06

• Add dask-spec CLI tool (GH#3090) Matthew Rocklin

• Connectionpool: don’t hand out closed connections (GH#3301) byjott

• Retry operations on network issues (GH#3294) byjott

• Skip Security.temporary() tests if cryptography not installed (GH#3302) James Bourbeau

• Support multiple listeners in the scheduler (GH#3288) Matthew Rocklin

• Updates RMM comment to the correct release (GH#3299) John Kirkham

• Add title to performance_report (GH#3298) Matthew Rocklin

• Forgot to fix slow test (GH#3297) Benjamin Zaitlen

• Update SSHCluster docstring parameters (GH#3296) James Bourbeau

• worker.close() awaits batched_stream.close() (GH#3291) Mads R. B. Kristensen

• Fix asynchronous listener in UCX (GH#3292) Benjamin Zaitlen

• Avoid repeatedly adding deps to already in memory stack (GH#3293) James Bourbeau

• xfail ucx empty object typed dataframe (GH#3279) Benjamin Zaitlen

• Fix distributed.wait documentation (GH#3289) Tom Rochette

• Move Python 3 syntax tests into main tests (GH#3281) Matthew Rocklin

• xfail test_workspace_concurrency for Python 3.6 (GH#3283) Matthew Rocklin

• Add performance_report context manager for static report generation (GH#3282) Matthew Rocklin

• Update function serialization caches with custom LRU class (GH#3260) James Bourbeau

• Make Listener.start asynchronous (GH#3278) Matthew Rocklin

• Remove dask-submit and dask-remote (GH#3280) Matthew Rocklin

• Worker profile server (GH#3274) Matthew Rocklin

• Improve bandwidth workers plot (GH#3273) Matthew Rocklin

• Make profile coroutines consistent between Scheduler and Worker (GH#3277) Matthew Rocklin

• Enable saving profile information from server threads (GH#3271) Matthew Rocklin

• Remove memory use plot (GH#3269) Matthew Rocklin

• Add offload size to configuration (GH#3270) Matthew Rocklin

• Fix layout scaling on profile plots (GH#3268) Jacob Tomlinson

• Set x_range in CPU plot based on the number of threads (GH#3266) Matthew Rocklin

• Use base-2 values for byte-valued axes in dashboard (GH#3267) Matthew Rocklin

• Robust gather in case of connection failures (GH#3246) fjetter

• Use DeviceBuffer from newer RMM releases (GH#3261) John Kirkham

• Fix dev requirements for pytest (GH#3264) Elliott Sales de Andrade

• Add validate options to configuration (GH#3258) Matthew Rocklin

2.8.1 - 2019-11-22

• Fix hanging worker when the scheduler leaves (GH#3250) Tom Augspurger

• Fix NumPy writeable serialization bug (GH#3253) James Bourbeau

• Skip numba.cuda tests if CUDA is not available (GH#3255) Peter Andreas Entschev

• Add new dashboard plot for memory use by key (GH#3243) Matthew Rocklin

• Fix array.shape() -> array.shape (GH#3247) Jed Brown

• Fixed typos in pubsub.py (GH#3244) He Jia

• Fixed cupy array going out of scope (GH#3240) Mads R. B. Kristensen

• Remove gen.coroutine usage in scheduler (GH#3242) Jim Crist-Harif

• Use inspect.isawaitable where relevant (GH#3241) Jim Crist-Harif

2.8.0 - 2019-11-14

• Add UCX config values (GH#3135) Matthew Rocklin

• Relax test_MultiWorker (GH#3210) Matthew Rocklin

• Avoid ucp.init at import time (GH#3211) Matthew Rocklin

• Clean up rpc to avoid intermittent test failure (GH#3215) Matthew Rocklin

• Respect protocol if given to Scheduler (GH#3212) Matthew Rocklin

• Use legend_field= keyword in bokeh plots (GH#3218) Matthew Rocklin

• Cache psutil.Process object in Nanny (GH#3207) Matthew Rocklin

• Replace gen.sleep with asyncio.sleep (GH#3208) Matthew Rocklin

• Avoid offloading serialization for small messages (GH#3224) Matthew Rocklin

• Add desired_workers metric (GH#3221) Gabriel Sailer

• Fail fast when importing distributed.comm.ucx (GH#3228) Matthew Rocklin

• Add module name to Future repr (GH#3231) Matthew Rocklin

• Add name to Pub/Sub repr (GH#3235) Matthew Rocklin

• Import CPU_COUNT from dask.system (GH#3199) James Bourbeau

• Efficiently serialize zero strided NumPy arrays (GH#3180) James Bourbeau

• Cache function deserialization in workers (GH#3234) Matthew Rocklin

• Respect ordering of futures in futures_of (GH#3236) Matthew Rocklin

• Bump dask dependency to 2.7.0 (GH#3237) James Bourbeau

• Avoid setting inf x_range (GH#3229) rockwellw

• Clear task stream based on recent behavior (GH#3200) Matthew Rocklin

• Use the percentage field for profile plots (GH#3238) Matthew Rocklin

2.7.0 - 2019-11-08

This release drops support for Python 3.5

• Adds badges to README.rst [skip ci] (GH#3152) James Bourbeau

• Don’t overwrite self.address if it is present (GH#3153) Gil Forsyth

• Remove outdated references to debug scheduler and worker bokeh pages. (GH#3160) darindf

• Update CONTRIBUTING.md (GH#3159) Jacob Tomlinson

• Add Prometheus metric for a worker’s executing tasks count (GH#3163) darindf

• Update Prometheus documentation (GH#3165) darindf

• Fix Numba serialization when strides is None (GH#3166) Peter Andreas Entschev

• Await cluster in Adaptive.recommendations (GH#3168) Simon Boothroyd

• Support automatic TLS (GH#3164) Jim Crist

• Avoid swamping high-memory workers with data requests (GH#3071) Tom Augspurger

• Update UCX variables to use sockcm by default (GH#3177) Peter Andreas Entschev

• Get protocol in Nanny/Worker from scheduler address (GH#3175) Peter Andreas Entschev

• Add worker and tasks state for Prometheus data collection (GH#3174) darindf

• Use async def functions for offload to/from_frames (GH#3171) Mads R. B. Kristensen

• Subprocesses inherit the global dask config (GH#3192) Mads R. B. Kristensen

• XFail test_open_close_many_workers (GH#3194) Matthew Rocklin

• Drop Python 3.5 (GH#3179) James Bourbeau

• UCX: avoid double init after fork (GH#3178) Mads R. B. Kristensen

• Silence warning when importing while offline (GH#3203) James A. Bednar

• Adds docs to Client methods for resources, actors, and traverse (GH#2851) IPetrik

• Add test for concurrent scatter operations (GH#2244) Matthew Rocklin

• Expand async docs (GH#2293) Dave Hirschfeld

• Add PatchedDeviceArray to drop stride attribute for cupy<7.0 (GH#3198) Richard J Zamora

2.6.0 - 2019-10-15

• Refactor dashboard module (GH#3138) Jacob Tomlinson

• Use setuptools.find_packages in setup.py (GH#3150) Matthew Rocklin

• Move death timeout logic up to Node.start (GH#3115) Matthew Rocklin

• Only include metric in WorkerTable if it is a scalar (GH#3140) Matthew Rocklin

• Add Nanny(config={...}) keyword (GH#3134) Matthew Rocklin

• Xfail test_worksapce_concurrency on Python 3.6 (GH#3132) Matthew Rocklin

• Extend Worker plugin API with transition method (GH#2994) matthieubulte

• Raise exception if the user passes in unused keywords to Client (GH#3117) Jonathan De Troye

• Move new SSHCluster to top level (GH#3128) Matthew Rocklin

• Bump dask dependency (GH#3124) Jim Crist

2.5.2 - 2019-10-04

• Make dask-worker close quietly when given sigint signal (GH#3116) Matthew Rocklin

• Replace use of tornado.gen with asyncio in dask-worker (GH#3114) Matthew Rocklin

• UCX: allocate CUDA arrays using RMM and Numba (GH#3109) Mads R. B. Kristensen

• Support calling cluster.scale as async method (GH#3110) Jim Crist

• Identify lost workers in SpecCluster based on address not name (GH#3088) James Bourbeau

• Add Client.shutdown method (GH#3106) Matthew Rocklin

• Collect worker-worker and type bandwidth information (GH#3094) Matthew Rocklin

• Send noise over the wire to keep dask-ssh connection alive (GH#3105) Gil Forsyth

• Retry scheduler connect multiple times (GH#3104) Jacob Tomlinson

• Add favicon of logo to the dashboard (GH#3095) James Bourbeau

• Remove utils.py functions for their dask/utils.py equivalents (GH#3042) Matthew Rocklin

• Lower default bokeh log level (GH#3087) Philipp Rudiger

• Check if self.cluster.scheduler is a local scheduler (GH#3099) Jacob Tomlinson

2.5.1 - 2019-09-27

• Support clusters that don’t have .security or ._close methods (GH#3100) Matthew Rocklin

2.5.0 - 2019-09-27

• Use the new UCX Python bindings (GH#3059) Mads R. B. Kristensen

• Fix worker preload config (GH#3027) byjott

• Fix widget with spec that generates multiple workers (GH#3067) Loïc Estève

• Make Client.get_versions async friendly (GH#3064) Jacob Tomlinson

• Add configuration option for longer error tracebacks (GH#3086) Daniel Farrell

• Have Client get Security from passed Cluster (GH#3079) Matthew Rocklin

• Respect Cluster.dashboard_link in Client._repr_html_ if it exists (GH#3077) Matthew Rocklin

• Add monitoring with dask cluster docs (GH#3072) Arpit Solanki

• Protocol of cupy and numba handles serialization exclusively (GH#3047) Mads R. B. Kristensen

• Allow specification of worker type in SSHCLuster (GH#3061) Jacob Tomlinson

• Use Cluster.scheduler_info for workers= value in repr (GH#3058) Matthew Rocklin

• Allow SpecCluster to scale by memory and cores (GH#3057) Matthew Rocklin

• Allow full script in preload inputs (GH#3052) Matthew Rocklin

• Check multiple cgroups dirs, ceil fractional cpus (GH#3056) Jim Crist

• Add blurb about disabling work stealing (GH#3055) Chris White

2.4.0 - 2019-09-13

• Remove six (GH#3045) Matthew Rocklin

• Add missing test data to sdist tarball (GH#3050) Elliott Sales de Andrade

• Use mock from unittest standard library (GH#3049) Elliott Sales de Andrade

• Use cgroups resource limits to determine default threads and memory (GH#3039) Jim Crist

• Move task deserialization to immediately before task execution (GH#3015) James Bourbeau

• Drop joblib shim module in distributed (GH#3040) John Kirkham

• Redirect configuration doc page (GH#3038) Matthew Rocklin

• Support --name 0 and --nprocs keywords in dask-worker cli (GH#3037) Matthew Rocklin

• Remove lost workers from SpecCluster.workers (GH#2990) Guillaume Eynard-Bontemps

• Clean up test_local.py::test_defaults (GH#3017) Matthew Rocklin

• Replace print statement in Queue.__init__ with debug message (GH#3035) Mikhail Akimov

• Set the x_range limit of the Meory utilization plot to memory-limit (GH#3034) Matthew Rocklin

• Rely on cudf codebase for cudf serialization (GH#2998) Benjamin Zaitlen

• Add fallback html repr for Cluster (GH#3023) Jim Crist

• Add support for zstandard compression to comms (GH#2970) Abael He

• Avoid collision when using os.environ in dashboard_link (GH#3021) Matthew Rocklin

• Fix ConnectionPool limit handling (GH#3005) byjott

• Support Spec jobs that generate multiple workers (GH#3013) Matthew Rocklin

• Tweak Logs styling (GH#3012) Jim Crist

• Better name for cudf deserialization function name (GH#3008) Benjamin Zaitlen

• Make spec.ProcessInterface a valid no-op worker (GH#3004) Matthew Rocklin

• Return dictionaries from new_worker_spec rather than name/worker pairs (GH#3000) Matthew Rocklin

• Fix minor typo in documentation (GH#3002) Mohammad Noor

• Permit more keyword options when scaling with cores and memory (GH#2997) Matthew Rocklin

• Add cuda_ipc to UCX environment for NVLink (GH#2996) Benjamin Zaitlen

• Add threads= and memory= to Cluster and Client reprs (GH#2995) Matthew Rocklin

• Fix PyNVML initialization (GH#2993) Richard J Zamora

2.3.2 - 2019-08-23

• Skip exceptions in startup information (GH#2991) Jacob Tomlinson

2.3.1 - 2019-08-22

• Add support for separate external address for SpecCluster scheduler (GH#2963) Jacob Tomlinson

• Defer cudf serialization/deserialization to that library (GH#2881) Benjamin Zaitlen

• Workaround for hanging test now calls ucp.fin() (GH#2967) Mads R. B. Kristensen

• Remove unnecessary bullet point (GH#2972) Pav A

• Directly import progress from diagnostics.progressbar (GH#2975) Matthew Rocklin

• Handle buffer protocol objects in ensure_bytes (GH#2969) Tom Augspurger

• Fix documentatation syntax and tree (GH#2981) Pav A

• Improve get_ip_interface error message when interface does not exist (GH#2964) Loïc Estève

• Add cores= and memory= keywords to scale (GH#2974) Matthew Rocklin

• Make workers robust to bad custom metrics (GH#2984) Matthew Rocklin

2.3.0 - 2019-08-16

• Except all exceptions when checking pynvml (GH#2961) Matthew Rocklin

• Pass serialization down through small base collections (GH#2948) Peter Andreas Entschev

• Use pytest.warning(Warning) rather than Exception (GH#2958) Matthew Rocklin

• Allow server_kwargs to override defaults in dashboard (GH#2955) Bruce Merry

• Update utils_perf.py (GH#2954) Shayan Amani

• Normalize names with str in retire_workers (GH#2949) Matthew Rocklin

• Update client.py (GH#2951) Shayan Amani

• Add GPUCurrentLoad dashboard plots (GH#2944) Matthew Rocklin

• Pass GPU diagnostics from worker to scheduler (GH#2932) Matthew Rocklin

• Import from collections.abc (GH#2938) Jim Crist

• Fixes Worker docstring formatting (GH#2939) James Bourbeau

• Redirect setup docs to docs.dask.org (GH#2936) Matthew Rocklin

• Wrap offload in gen.coroutine (GH#2934) Matthew Rocklin

• Change TCP.close to a coroutine to avoid task pending warning (GH#2930) Matthew Rocklin

• Fixup black string normalization (GH#2929) Jim Crist

• Move core functionality from SpecCluster to Cluster (GH#2913) Matthew Rocklin

• Add aenter/aexit protocols to ProcessInterface (GH#2927) Matthew Rocklin

• Add real-time CPU utilization plot to dashboard (GH#2922) Matthew Rocklin

• Always kill processes in clean tests, even if we don’t check (GH#2924) Matthew Rocklin

• Add timeouts to processes in SSH tests (GH#2925) Matthew Rocklin

• Add documentation around spec.ProcessInterface (GH#2923) Matthew Rocklin

• Cleanup async warnings in tests (GH#2920) Matthew Rocklin

• Give 404 when requesting nonexistent tasks or workers (GH#2921) Martin Durant

• Raise informative warning when rescheduling an unknown task (GH#2916) James Bourbeau

• Fix docstring (GH#2917) Martin Durant

• Add keep-alive message between worker and scheduler (GH#2907) Matthew Rocklin

• Rewrite Adaptive/SpecCluster to support slowly arriving workers (GH#2904) Matthew Rocklin

• Call heartbeat rather than reconnect on disconnection (GH#2906) Matthew Rocklin

2.2.0 - 2019-07-31

• Respect security configuration in LocalCluster (GH#2822) Russ Bubley

• Add Nanny to worker docs (GH#2826) Christian Hudon

• Don’t make False add-keys report to scheduler (GH#2421) tjb900

• Include type name in SpecCluster repr (GH#2834) Jacob Tomlinson

• Extend prometheus metrics endpoint (GH#2833) Gabriel Sailer

• Add alternative SSHCluster implementation (GH#2827) Matthew Rocklin

• Dont reuse closed worker in get_worker (GH#2841) Pierre Glaser

• SpecCluster: move init logic into start (GH#2850) Jacob Tomlinson

• Document distributed.Reschedule in API docs (GH#2860) James Bourbeau

• Add fsspec to installation of test builds (GH#2859) Martin Durant

• Make await/start more consistent across Scheduler/Worker/Nanny (GH#2831) Matthew Rocklin

• Add cleanup fixture for asyncio tests (GH#2866) Matthew Rocklin

• Use only remote connection to scheduler in Adaptive (GH#2865) Matthew Rocklin

• Add Server.finished async function (GH#2864) Matthew Rocklin

• Align text and remove bullets in Client HTML repr (GH#2867) Matthew Rocklin

• Test dask-scheduler –idle-timeout flag (GH#2862) Matthew Rocklin

• Remove Client.upload_environment (GH#2877) Jim Crist

• Replace gen.coroutine with async/await in core (GH#2871) Matthew Rocklin

• Forcefully kill all processes before each test (GH#2882) Matthew Rocklin

• Cleanup Security class and configuration (GH#2873) Jim Crist

• Remove unused variable in SpecCluster scale down (GH#2870) Jacob Tomlinson

• Add SpecCluster ProcessInterface (GH#2874) Jacob Tomlinson

• Add Log(str) and Logs(dict) classes for nice HTML reprs (GH#2875) Jacob Tomlinson

• Pass Client._asynchronous to Cluster._asynchronous (GH#2890) Matthew Rocklin

• Add default logs method to Spec Cluster (GH#2889) Matthew Rocklin

• Add processes keyword back into clean (GH#2891) Matthew Rocklin

• Update black (GH#2901) Matthew Rocklin

• Move Worker.local_dir attribute to Worker.local_directory (GH#2900) Matthew Rocklin

• Link from TapTools to worker info pages in dashboard (GH#2894) Matthew Rocklin

• Avoid exception in Client._ensure_connected if closed (GH#2893) Matthew Rocklin

• Convert Pythonic kwargs to CLI Keywords for SSHCluster (GH#2898) Matthew Rocklin

• Use kwargs in CLI (GH#2899) Matthew Rocklin

• Name SSHClusters by providing name= keyword to SpecCluster (GH#2903) Matthew Rocklin

• Request feed of worker information from Scheduler to SpecCluster (GH#2902) Matthew Rocklin

• Clear out compatibillity file (GH#2896) Matthew Rocklin

• Remove future imports (GH#2897) Matthew Rocklin

• Use click’s show_default=True in relevant places (GH#2838) Christian Hudon

• Close workers more gracefully (GH#2905) Matthew Rocklin

• Close workers gracefully with –lifetime keywords (GH#2892) Matthew Rocklin

• Add closing <li> tags to Client._repr_html_ (GH#2911) Matthew Rocklin

• Add endline spacing in Logs._repr_html_ (GH#2912) Matthew Rocklin

2.1.0 - 2019-07-08

• Fix typo that prevented error message (GH#2825) Russ Bubley

• Remove dask-mpi (GH#2824) Matthew Rocklin

• Updates to use update_graph in task journey docs (GH#2821) James Bourbeau

• Fix Client repr with memory_info=None (GH#2816) Matthew Rocklin

• Fix case where key, rather than TaskState, could end up in ts.waiting_on (GH#2819) tjb900

• Use Keyword-only arguments (GH#2814) Matthew Rocklin

• Relax check for worker references in cluster context manager (GH#2813) Matthew Rocklin

• Add HTTPS support for the dashboard (GH#2812) Jim Crist

• Use dask.utils.format_bytes (GH#2810) Tom Augspurger

2.0.1 - 2019-06-26

We neglected to include python_requires= in our setup.py file, resulting in confusion for Python 2 users who erroneously get packages for 2.0.0. This is fixed in 2.0.1 and we have removed the 2.0.0 files from PyPI.

• Add python_requires entry to setup.py (GH#2807) Matthew Rocklin

• Correctly manage tasks beyond deque limit in TaskStream plot (GH#2797) Matthew Rocklin

• Fix diagnostics page for memory_limit=None (GH#2770) Brett Naul

2.0.0 - 2019-06-25

• Drop support for Python 2

• Relax warnings before release (GH#2796) Matthew Rocklin

• Deprecate –bokeh/–no-bokeh CLI (GH#2800) Tom Augspurger

• Typo in bokeh service_kwargs for dask-worker (GH#2783) Tom Augspurger

• Update command line cli options docs (GH#2794) James Bourbeau

• Remove “experimental” from TLS docs (GH#2793) James Bourbeau

• Add warnings around ncores= keywords (GH#2791) Matthew Rocklin

• Add –version option to scheduler and worker CLI (GH#2782) Tom Augspurger

• Raise when workers initialization times out (GH#2784) Tom Augspurger

• Replace ncores with nthreads throughout codebase (GH#2758) Matthew Rocklin

• Add unknown pytest markers (GH#2764) Tom Augspurger

• Delay lookup of allowed failures. (GH#2761) Tom Augspurger

• Change address -> worker in ColumnDataSource for nbytes plot (GH#2755) Matthew Rocklin

• Remove module state in Prometheus Handlers (GH#2760) Matthew Rocklin

• Add stress test for UCX (GH#2759) Matthew Rocklin

• Add nanny logs (GH#2744) Tom Augspurger

• Move some of the adaptive logic into the scheduler (GH#2735) Matthew Rocklin

• Add SpecCluster.new_worker_spec method (GH#2751) Matthew Rocklin

• Worker dashboard fixes (GH#2747) Matthew Rocklin

• Add async context managers to scheduler/worker classes (GH#2745) Matthew Rocklin

• Fix the resource key representation before sending graphs (GH#2733) Michael Spiegel

• Allow user to configure whether workers are daemon. (GH#2739) Caleb

• Pin pytest >=4 with pip in appveyor and python 3.5 (GH#2737) Matthew Rocklin

• Add Experimental UCX Comm (GH#2591) Ben Zaitlen Tom Augspurger Matthew Rocklin

• Close nannies gracefully (GH#2731) Matthew Rocklin

• add kwargs to progressbars (GH#2638) Manuel Garrido

• Add back LocalCluster.__repr__. (GH#2732) Loïc Estève

• Move bokeh module to dashboard (GH#2724) Matthew Rocklin

• Close clusters at exit (GH#2730) Matthew Rocklin

• Add SchedulerPlugin TaskState example (GH#2622) Matt Nicolls

• Add SpecificationCluster (GH#2675) Matthew Rocklin

• Replace register_worker_callbacks with worker plugins (GH#2453) Matthew Rocklin

• Proxy worker dashboards from scheduler dashboard (GH#2715) Ben Zaitlen

• Add docstring to Scheduler.check_idle_saturated (GH#2721) Matthew Rocklin

• Refer to LocalCluster in Client docstring (GH#2719) Matthew Rocklin

• Remove special casing of Scikit-Learn BaseEstimator serialization (GH#2713) Matthew Rocklin

• Fix two typos in Pub class docstring (GH#2714) Magnus Nord

• Support uploading files with multiple modules (GH#2587) Sam Grayson

• Change the main workers bokeh page to /status (GH#2689) Ben Zaitlen

• Cleanly stop periodic callbacks in Client (GH#2705) Matthew Rocklin

• Disable pan tool for the Progress, Byte Stored and Tasks Processing plot (GH#2703) Mathieu Dugré

• Except errors in Nanny’s memory monitor if process no longer exists (GH#2701) Matthew Rocklin

• Handle heartbeat when worker has just left (GH#2702) Matthew Rocklin

• Modify styling of histograms for many-worker dashboard plots (GH#2695) Mathieu Dugré

• Add method to wait for n workers before continuing (GH#2688) Daniel Farrell

• Support computation on delayed(None) (GH#2697) Matthew Rocklin

• Cleanup localcluster (GH#2693) Matthew Rocklin

• Use ‘temporary-directory’ from dask.config for Worker’s directory (GH#2654) Matthew Rocklin

• Remove support for Iterators and Queues (GH#2671) Matthew Rocklin

1.28.1 - 2019-05-13

This is a small bugfix release due to a config change upstream.

• Use config accessor method for “scheduler-address” (GH#2676) James Bourbeau

1.28.0 - 2019-05-08

• Add Type Attribute to TaskState (GH#2657) Matthew Rocklin

• Add waiting task count to progress title bar (GH#2663) James Bourbeau

• DOC: Clean up reference to cluster object (GH#2664) K.-Michael Aye

• Allow scheduler to politely close workers as part of shutdown (GH#2651) Matthew Rocklin

• Check direct_to_workers before using get_worker in Client (GH#2656) Matthew Rocklin

• Fixed comment regarding keeping existing level if less verbose (GH#2655) Brett Randall

• Add idle timeout to scheduler (GH#2652) Matthew Rocklin

• Avoid deprecation warnings (GH#2653) Matthew Rocklin

• Use an LRU cache for deserialized functions (GH#2623) Matthew Rocklin

• Rename Worker._close to Worker.close (GH#2650) Matthew Rocklin

• Add Comm closed bookkeeping (GH#2648) Matthew Rocklin

• Explain LocalCluster behavior in Client docstring (GH#2647) Matthew Rocklin

• Add last worker into KilledWorker exception to help debug (GH#2610) @plbertrand

• Set working worker class for dask-ssh (GH#2646) Martin Durant

• Add as_completed methods to docs (GH#2642) Jim Crist

• Add timeout to Client._reconnect (GH#2639) Jim Crist

• Limit test_spill_by_default memory, reenable it (GH#2633) Peter Andreas Entschev

• Use proper address in worker -> nanny comms (GH#2640) Jim Crist

• Fix deserialization of bytes chunks larger than 64MB (GH#2637) Peter Andreas Entschev

1.27.1 - 2019-04-29

• Adaptive: recommend close workers when any are idle (GH#2330) Michael Delgado

• Increase GC thresholds (GH#2624) Matthew Rocklin

• Add interface= keyword to LocalCluster (GH#2629) Matthew Rocklin

• Add worker_class argument to LocalCluster (GH#2625) Matthew Rocklin

• Remove Python 2.7 from testing matrix (GH#2631) Matthew Rocklin

• Add number of trials to diskutils test (GH#2630) Matthew Rocklin

• Fix parameter name in LocalCluster docstring (GH#2626) Loïc Estève

• Integrate stacktrace for low-level profiling (GH#2575) Peter Andreas Entschev

• Apply Black to standardize code styling (GH#2614) Matthew Rocklin

• added missing whitespace to start_worker cmd (GH#2613) condoratberlin

• Updated logging module doc links from docs.python.org/2 to docs.python.org/3. (GH#2635) Brett Randall

1.27.0 - 2019-04-12

• Add basic health endpoints to scheduler and worker bokeh. (GH#2607) amerkel2

• Improved description accuracy of –memory-limit option. (GH#2601) Brett Randall

• Check self.dependencies when looking at dependent tasks in memory (GH#2606) deepthirajagopalan7

• Add RabbitMQ SchedulerPlugin example (GH#2604) Matt Nicolls

• add resources to scheduler update_graph plugin (GH#2603) Matt Nicolls

• Use ensure_bytes in serialize_error (GH#2588) Matthew Rocklin

• Specify data storage explicitly from Worker constructor (GH#2600) Matthew Rocklin

• Change bokeh port keywords to dashboard_address (GH#2589) Matthew Rocklin

• .detach_() pytorch tensor to serialize data as numpy array. (GH#2586) Muammar El Khatib

• Add warning if creating scratch directories takes a long time (GH#2561) Matthew Rocklin

• Fix typo in pub-sub doc. (GH#2599) Loïc Estève

• Allow return_when=’FIRST_COMPLETED’ in wait (GH#2598) Nikos Tsaousis

• Forward kwargs through Nanny to Worker (GH#2596) Brian Chu

• Use ensure_dict instead of dict (GH#2594) James Bourbeau

• Specify protocol in LocalCluster (GH#2489) Matthew Rocklin

1.26.1 - 2019-03-29

• Fix LocalCluster to not overallocate memory when overcommitting threads per worker (GH#2541) George Sakkis

• Make closing resilient to lacking an address (GH#2542) Matthew Rocklin

• fix typo in comment (GH#2546) Brett Jurman

• Fix double init of prometheus metrics (GH#2544) Marco Neumann

• Skip test_duplicate_clients without bokeh. (GH#2553) Elliott Sales de Andrade

• Add blocked_handlers to servers (GH#2556) Chris White

• Always yield Server.handle_comm coroutine (GH#2559) Tom Augspurger

• Use yaml.safe_load (GH#2566) Matthew Rocklin

• Fetch executables from build root. (GH#2551) Elliott Sales de Andrade

• Fix Torando 6 test failures (GH#2570) Matthew Rocklin

• Fix test_sync_closed_loop (GH#2572) Matthew Rocklin

1.26.0 - 2019-02-25

• Update style to fix recent flake8 update (GH#2500) (GH#2509) Matthew Rocklin

• Fix typo in gen_cluster log message (GH#2503) Loïc Estève

• Allow KeyError when closing event loop (GH#2498) Matthew Rocklin

• Avoid thread testing for TCP ThreadPoolExecutor (GH#2510) Matthew Rocklin

• Find Futures inside SubgraphCallable (GH#2505) Jim Crist

• Avoid AttributeError when closing and sending a message (GH#2514) Matthew Rocklin

• Add deprecation warning to dask_mpi.py (GH#2522) Julia Kent

• Relax statistical profiling test (GH#2527) Matthew Rocklin

• Support alternative –remote-dask-worker SSHCluster() and dask-ssh CLI (GH#2526) Adam Beberg

• Iterate over full list of plugins in transition (GH#2518) Matthew Rocklin

• Create Prometheus Endpoint (GH#2499) Adam Beberg

• Use pytest.importorskip for prometheus test (GH#2533) Matthew Rocklin

• MAINT skip prometheus test when no installed (GH#2534) Olivier Grisel

• Fix intermittent testing failures (GH#2535) Matthew Rocklin

• Avoid using nprocs keyword in dask-ssh if set to one (GH#2531) Matthew Rocklin

• Bump minimum Tornado version to 5.0

1.25.3 - 2019-01-31

• Fix excess threading on missing connections (GH#2403) Daniel Farrell

• Fix typo in doc (GH#2457) Loïc Estève

• Start fewer but larger workers with LocalCluster (GH#2452) Matthew Rocklin

• Check for non-zero length first in read loop (GH#2465) John Kirkham

• DOC: Use of local cluster in script (GH#2462) Peter Killick

• DOC/API: Signature for base class write / read (GH#2472) Tom Augspurger

• Support Pytest 4 in Tests (GH#2478) Adam Beberg

• Ensure async behavior in event loop with LocalCluster (GH#2484) Matthew Rocklin

• Fix spurious CancelledError (GH#2485) Loïc Estève

• Properly reset dask.config scheduler and shuffle when closing the client (GH#2475) George Sakkis

• Make it more explicit that resources are per worker. (GH#2470) Loïc Estève

• Remove references to center (GH#2488) Matthew Rocklin

• Expand client clearing timeout to 10s in testing (GH#2493) Matthew Rocklin

• Propagate key keyword in progressbar (GH#2492) Matthew Rocklin

• Use provided cluster’s IOLoop if present in Client (GH#2494) Matthew Rocklin

1.25.2 - 2019-01-04

• Clean up LocalCluster logging better in async mode (GH#2448) Matthew Rocklin

• Add short error message if bokeh cannot be imported (GH#2444) Dirk Petersen

• Add optional environment variables to Nanny (GH#2431) Matthew Rocklin

• Make the direct keyword docstring entries uniform (GH#2441) Matthew Rocklin

• Make LocalCluster.close async friendly (GH#2437) Matthew Rocklin

• gather_dep: don’t request dependencies we already found out we don’t want (GH#2428) tjb900

• Add parameters to Client.run docstring (GH#2429) Matthew Rocklin

• Support coroutines and async-def functions in run/run_scheduler (GH#2427) Matthew Rocklin

• Name threads in ThreadPoolExecutors (GH#2408) Matthew Rocklin

1.25.1 - 2018-12-15

• Serialize numpy.ma.masked objects properly (GH#2384) Jim Crist

• Turn off bokeh property validation in dashboard (GH#2387) Jim Crist

• Fully initialize WorkerState objects (GH#2388) Jim Crist

• Fix typo in scheduler docstring (GH#2393) Russ Bubley

• DOC: fix typo in distributed.worker.Worker docstring (GH#2395) Loïc Estève

• Remove clients and workers from event log after removal (GH#2394) tjb900

• Support msgpack 0.6.0 by providing length keywords (GH#2399) tjb900

• Use async-await on large messages test (GH#2404) Matthew Rocklin

• Fix race condition in normalize_collection (GH#2386) Jim Crist

• Fix redict collection after HighLevelGraph fix upstream (GH#2413) Matthew Rocklin

• Add a blocking argument to Lock.acquire() (GH#2412) Stephan Hoyer

• Fix long traceback test (GH#2417) Matthew Rocklin

• Update x509 certificates to current OpenSSL standards. (GH#2418) Diane Trout

1.25.0 - 2018-11-28

• Fixed the 404 error on the Scheduler Dashboard homepage (GH#2361) Michael Wheeler

• Consolidate two Worker classes into one (GH#2363) Matthew Rocklin

• Avoid warnings in pyarrow and msgpack (GH#2364) Matthew Rocklin

• Avoid race condition in Actor’s Future (GH#2374) Matthew Rocklin

• Support missing packages keyword in Client.get_versions (GH#2379) Matthew Rocklin

• Fixup serializing masked arrays (GH#2373) Jim Crist

1.24.2 - 2018-11-15

• Add support for Bokeh 1.0 (GH#2348) (GH#2356) Matthew Rocklin

• Fix regression that dropped support for Tornado 4 (GH#2353) Roy Wedge

• Avoid deprecation warnings (GH#2355) (GH#2357) Matthew Rocklin

• Fix typo in worker documentation (GH#2349) Tom Rochette

1.24.1 - 2018-11-09

• Use tornado’s builtin AnyThreadLoopEventPolicy (GH#2326) Matthew Rocklin

• Adjust TLS tests for openssl 1.1 (GH#2331) Marius van Niekerk

• Avoid setting event loop policy if within Jupyter notebook server (GH#2343) Matthew Rocklin

• Add preload script to conf (GH#2325) Guillaume Eynard-Bontemps

• Add serializer for Numpy masked arrays (GH#2335) Peter Killick

• Use psutil.Process.oneshot (GH#2339) NotSqrt

• Use worker SSL context when getting client from worker. (GH#2301) Anonymous

1.24.0 - 2018-10-26

• Remove Joblib Dask Backend from codebase (GH#2298) Matthew Rocklin

• Include worker tls protocol in Scheduler.restart (GH#2295) Matthew Rocklin

• Adapt to new Bokeh selection for 1.0 (GH#2292) Matthew Rocklin

• Add explicit retry method to Future and Client (GH#2299) Matthew Rocklin

• Point to main worker page in bokeh links (GH#2300) Matthew Rocklin

• Limit concurrency when gathering many times (GH#2303) Matthew Rocklin

• Add tls_cluster pytest fixture (GH#2302) Matthew Rocklin

• Convert ConnectionPool.open and active to properties (GH#2304) Matthew Rocklin

• change export_tb to format_tb (GH#2306) Eric Ma

• Redirect joblib page to dask-ml (GH#2307) Matthew Rocklin

• Include unserializable object in error message (GH#2310) Matthew Rocklin

• Import Mapping, Iterator, Set from collections.abc in Python 3 (GH#2315) Gaurav Sheni

• Extend Client.scatter docstring (GH#2320) Eric Ma

• Update for new flake8 (GH#2321) Matthew Rocklin

1.23.3 - 2018-10-05

• Err in dask serialization if not a NotImplementedError (GH#2251) Matthew Rocklin

• Protect against key missing from priority in GraphLayout (GH#2259) Matthew Rocklin

• Do not pull data twice in Client.gather (GH#2263) Adam Klein

• Add pytest fixture for cluster tests (GH#2262) Matthew Rocklin

• Cleanup bokeh callbacks (GH#2261) (GH#2278) Matthew Rocklin

• Fix bokeh error for memory_limit=None (GH#2255) Brett Naul

• Place large keywords into task graph in Client.map (GH#2281) Matthew Rocklin

• Remove redundant blosc threading code from protocol.numpy (GH#2284) Mike Gevaert

• Add ncores to workertable (GH#2289) Matthew Rocklin

• Support upload_file on files with no extension (GH#2290) Matthew Rocklin

1.23.2 - 2018-09-17

• Discard dependent rather than remove (GH#2250) Matthew Rocklin

• Use dask_sphinx_theme Matthew Rocklin

• Drop the Bokeh index page (GH#2241) John Kirkham

• Revert change to keep link relative (GH#2242) Matthew Rocklin

• docs: Fix broken AWS link in setup.rst file (GH#2240) Vladyslav Moisieienkov

• Return cancelled futures in as_completed (GH#2233) Chris White

1.23.1 - 2018-09-06

• Raise informative error when mixing futures between clients (GH#2227) Matthew Rocklin

• add byte_keys to unpack_remotedata call (GH#2232) Matthew Rocklin

• Add documentation for gist/rawgit for get_task_stream (GH#2236) Matthew Rocklin

• Quiet Client.close by waiting for scheduler stop signal (GH#2237) Matthew Rocklin

• Display system graphs nicely on different screen sizes (GH#2239) Derek Ludwig

• Mutate passed in workers dict in TaskStreamPlugin.rectangles (GH#2238) Matthew Rocklin

1.23.0 - 2018-08-30

• Add direct_to_workers to Client Matthew Rocklin

• Add Scheduler.proxy to workers Matthew Rocklin

• Implement Actors Matthew Rocklin

• Fix tooltip (GH#2168) Loïc Estève

• Fix scale / avoid returning coroutines (GH#2171) Joe Hamman

• Clarify dask-worker –nprocs (GH#2173) Yu Feng

• Concatenate all bytes of small messages in TCP comms (GH#2172) Matthew Rocklin

• Add dashboard_link property (GH#2176) Jacob Tomlinson

• Always offload to_frames (GH#2170) Matthew Rocklin

• Warn if desired port is already in use (GH#2191) (GH#2199) Matthew Rocklin

• Add profile page for event loop thread (GH#2144) Matthew Rocklin

• Use dispatch for dask serialization, also add sklearn, pytorch (GH#2175) Matthew Rocklin

• Handle corner cases with busy signal (GH#2182) Matthew Rocklin

• Check self.dependencies when looking at tasks in memory (GH#2196) Matthew Rocklin

• Add ability to log additional custom metrics from each worker (GH#2169) Loïc Estève

• Fix formatting when port is a tuple (GH#2204) Loïc Estève

• Describe what ZeroMQ is (GH#2211) Mike DePalatis

• Tiny typo fix (GH#2214) Anderson Banihirwe

• Add Python 3.7 to travis.yml (GH#2203) Matthew Rocklin

• Add plot= keyword to get_task_stream (GH#2198) Matthew Rocklin

• Add support for optional versions in Client.get_versions (GH#2216) Matthew Rocklin

• Add routes for solo bokeh figures in dashboard (GH#2185) Matthew Rocklin

• Be resilient to missing dep after busy signal (GH#2217) Matthew Rocklin

• Use CSS Grid to layout status page on the dashboard (GH#2213) Derek Ludwig and Luke Canavan

• Fix deserialization of queues on main ioloop thread (GH#2221) Matthew Rocklin

• Add a worker initialization function (GH#2201) Guillaume Eynard-Bontemps

• Collapse navbar in dashboard (GH#2223) Luke Canavan

1.22.1 - 2018-08-03

• Add worker_class= keyword to Nanny to support different worker types (GH#2147) Martin Durant

• Cleanup intermittent worker failures (GH#2152) (GH#2146) Matthew Rocklin

• Fix msgpack PendingDeprecationWarning for encoding=’utf-8’ (GH#2153) Olivier Grisel

• Make bokeh coloring deterministic using hash function (GH#2143) Matthew Rocklin

• Allow client to query the task stream plot (GH#2122) Matthew Rocklin

• Use PID and counter in thread names (GH#2084) (GH#2128) Dror Birkman

• Test that worker restrictions are cleared after cancellation (GH#2107) Matthew Rocklin

• Expand resources in graph_to_futures (GH#2131) Matthew Rocklin

• Add custom serialization support for pyarrow (GH#2115) Dave Hirschfeld

• Update dask-scheduler cli help text for preload (GH#2120) Matt Nicolls

• Added another nested parallelism test (GH#1710) Tom Augspurger

• insert newline by default after TextProgressBar (GH#1976) Phil Tooley

• Retire workers from scale (GH#2104) Matthew Rocklin

• Allow worker to refuse data requests with busy signal (GH#2092) Matthew Rocklin

• Don’t forget released keys (GH#2098) Matthew Rocklin

• Update example for stopping a worker (GH#2088) John Kirkham

• removed hardcoded value of memory terminate fraction from a log message (GH#2096) Bartosz Marcinkowski

• Adjust worker doc after change in config file location and treatment (GH#2094) Aurélien Ponte

• Prefer gathering data from same host (GH#2090) Matthew Rocklin

• Handle exceptions on deserialized comm with text error (GH#2093) Matthew Rocklin

• Fix typo in docstring (GH#2087) Loïc Estève

• Provide communication context to serialization functions (GH#2054) Matthew Rocklin

• Allow name to be explicitly passed in publish_dataset (GH#1995) Marius van Niekerk

• Avoid accessing Worker.scheduler_delay around yield point (GH#2074) Matthew Rocklin

• Support TB and PB in format bytes (GH#2072) Matthew Rocklin

• Add test for as_completed for loops in Python 2 (GH#2071) Matthew Rocklin

• Allow adaptive to exist without a cluster (GH#2064) Matthew Rocklin

• Have worker data transfer wait until recipient acknowledges (GH#2052) Matthew Rocklin

• Support async def functions in Client.sync (GH#2070) Matthew Rocklin

• Add asynchronous parameter to docstring of LocalCluster Matthew Rocklin

• Normalize address before comparison (GH#2066) Tom Augspurger

• Use ConnectionPool for Worker.scheduler Matthew Rocklin

• Avoid reference cycle in str_graph Matthew Rocklin

• Pull data outside of while loop in gather (GH#2059) Matthew Rocklin

1.22.0 - 2018-06-14

• Overhaul configuration (GH#1948) Matthew Rocklin

• Replace get= keyword with scheduler= (GH#1959) Matthew Rocklin

• Use tuples in msgpack (GH#2000) Matthew Rocklin and Marius van Niekerk

• Unify handling of high-volume connections (GH#1970) Matthew Rocklin

• Automatically scatter large arguments in joblib connector (GH#2020) (GH#2030) Olivier Grisel

• Turn click Python 3 locales failure into a warning (GH#2001) Matthew Rocklin

• Rely on dask implementation of sizeof (GH#2042) Matthew Rocklin

• Replace deprecated workers.iloc with workers.values() (GH#2013) Grant Jenks

• Introduce serialization families (GH#1912) Matthew Rocklin

• Add PubSub (GH#1999) Matthew Rocklin

• Add Dask stylesheet to documentation Matthew Rocklin

• Avoid recomputation on partially-complete results (GH#1840) Matthew Rocklin

• Use sys.prefix in popen for testing (GH#1954) Matthew Rocklin

• Include yaml files in manifest Matthew Rocklin

• Use self.sync so Client.processing works in asynchronous context (GH#1962) Henry Doupe

• Fix bug with bad repr on closed client (GH#1965) Matthew Rocklin

• Parse –death-timeout keyword in dask-worker (GH#1967) Matthew Rocklin

• Support serializers in BatchedSend (GH#1964) Matthew Rocklin

• Use normal serialization mechanisms to serialize published datasets (GH#1972) Matthew Rocklin

• Add security support to LocalCluster. (GH#1855) Marius van Niekerk

• add ConnectionPool.remove method (GH#1977) Tony Lorenzo

• Cleanly close workers when scheduler closes (GH#1981) Matthew Rocklin

• Add .pyz support in upload_file (GH#1781) @bmaisson

• add comm to packages (GH#1980) Matthew Rocklin

• Replace dask.set_options with dask.config.set Matthew Rocklin

• Exclude versions of sortedcontainers which do not have .iloc. (GH#1993) Russ Bubley

• Exclude gc statistics under PyPy (GH#1997) Marius van Niekerk

• Manage recent config and dataframe changes in dask (GH#2009) Matthew Rocklin

• Cleanup lingering clients in tests (GH#2012) Matthew Rocklin

• Use timeouts during Client._ensure_connected (GH#2011) Martin Durant

• Avoid reference cycle in joblib backend (GH#2014) Matthew Rocklin, also Olivier Grisel

• DOC: fixed test example (GH#2017) Tom Augspurger

• Add worker_key parameter to Adaptive (GH#1992) Matthew Rocklin

• Prioritize tasks with their true keys, before stringifying (GH#2006) Matthew Rocklin

• Serialize worker exceptions through normal channels (GH#2016) Matthew Rocklin

• Include exception in progress bar (GH#2028) Matthew Rocklin

• Avoid logging orphaned futures in All (GH#2008) Matthew Rocklin

• Don’t use spill-to-disk dictionary if we’re not spilling to disk Matthew Rocklin

• Only avoid recomputation if key exists (GH#2036) Matthew Rocklin

• Use client connection and serialization arguments in progress (GH#2035) Matthew Rocklin

• Rejoin worker client on closing context manager (GH#2041) Matthew Rocklin

• Avoid forgetting erred tasks when losing dependencies (GH#2047) Matthew Rocklin

• Avoid collisions in graph_layout (GH#2050) Matthew Rocklin

• Avoid recursively calling bokeh callback in profile plot (GH#2048) Matthew Rocklin

1.21.8 - 2018-05-03

• Remove errant print statement (GH#1957) Matthew Rocklin

• Only add reevaluate_occupancy callback once (GH#1953) Tony Lorenzo

1.21.7 - 2018-05-02

• Newline needed for doctest rendering (GH#1917) Loïc Estève

• Support Client._repr_html_ when in async mode (GH#1909) Matthew Rocklin

• Add parameters to dask-ssh command (GH#1910) Irene Rodriguez

• Sanitize get_dataset trace (GH#1888) John Kirkham

• Fix bug where queues would not clean up cleanly (GH#1922) Matthew Rocklin

• Delete cached file safely in upload file (GH#1921) Matthew Rocklin

• Accept KeyError when closing tornado IOLoop in tests (GH#1937) Matthew Rocklin

• Quiet the client and scheduler when gather(…, errors=’skip’) (GH#1936) Matthew Rocklin

• Clarify couldn’t gather keys warning (GH#1942) Kenneth Koski

• Support submit keywords in joblib (GH#1947) Matthew Rocklin

• Avoid use of external resources in bokeh server (GH#1934) Matthew Rocklin

• Drop __contains__ from Datasets (GH#1889) John Kirkham

• Fix bug with queue timeouts (GH#1950) Matthew Rocklin

• Replace msgpack-python by msgpack (GH#1927) Loïc Estève

1.21.6 - 2018-04-06

• Fix numeric environment variable configuration (GH#1885) Joseph Atkins-Kurkish

• support bytearrays in older lz4 library (GH#1886) Matthew Rocklin

• Remove started timeout in nanny (GH#1852) Matthew Rocklin

• Don’t log errors in sync (GH#1894) Matthew Rocklin

• downgrade stale lock warning to info logging level (GH#1890) Matthew Rocklin

• Fix UnboundLocalError for key (GH#1900) John Kirkham

• Resolve deployment issues in Python 2 (GH#1905) Matthew Rocklin

• Support retries and priority in Client.get method (GH#1902) Matthew Rocklin

• Add additional attributes to task page if applicable (GH#1901) Matthew Rocklin

• Add count method to as_completed (GH#1897) Matthew Rocklin

• Extend default timeout to 10s (GH#1904) Matthew Rocklin

1.21.5 - 2018-03-31

• Increase default allowable tick time to 3s (GH#1854) Matthew Rocklin

• Handle errant workers when another worker has data (GH#1853) Matthew Rocklin

• Close multiprocessing queue in Nanny to reduce open file descriptors (GH#1862) Matthew Rocklin

• Extend nanny started timeout to 30s, make configurable (GH#1865) Matthew Rocklin

• Comment out the default config file (GH#1871) Matthew Rocklin

• Update to fix bokeh 0.12.15 update errors (GH#1872) Matthew Rocklin

• Downgrade Event Loop unresponsive warning to INFO level (GH#1870) Matthew Rocklin

• Add fifo timeout to control priority generation (GH#1828) Matthew Rocklin

• Add retire_workers API to Client (GH#1876) Matthew Rocklin

• Catch NoSuchProcess error in Nanny.memory_monitor (GH#1877) Matthew Rocklin

• Add uid to nanny queue communications (GH#1880) Matthew Rocklin

1.21.4 - 2018-03-21

• Avoid passing bytearrays to snappy decompression (GH#1831) Matthew Rocklin

• Specify IOLoop in Adaptive (GH#1841) Matthew Rocklin

• Use connect-timeout config value throughout client (GH#1839) Matthew Rocklin

• Support direct= keyword argument in Client.get (GH#1845) Matthew Rocklin

1.21.3 - 2018-03-08

• Add cluster superclass and improve adaptivity (GH#1813) Matthew Rocklin

• Fixup tests and support Python 2 for Tornado 5.0 (GH#1818) Matthew Rocklin

• Fix bug in recreate_error when dependencies are dropped (GH#1815) Matthew Rocklin

• Add worker time to live in Scheduler (GH#1811) Matthew Rocklin

• Scale adaptive based on total_occupancy (GH#1807) Matthew Rocklin

• Support calling compute within worker_client (GH#1814) Matthew Rocklin

• Add percentage to profile plot (GH#1817) Brett Naul

• Overwrite option for remote python in dask-ssh (GH#1812) Sven Kreiss

1.21.2 - 2018-03-05

• Fix bug where we didn’t check idle/saturated when stealing (GH#1801) Matthew Rocklin

• Fix bug where client was noisy when scheduler closed unexpectedly (GH#1806) Matthew Rocklin

• Use string-based timedeltas (like '500 ms') everywhere (GH#1804) Matthew Rocklin

• Keep logs in scheduler and worker even if silenced (GH#1803) Matthew Rocklin

• Support minimum, maximum, wait_count keywords in Adaptive (GH#1797) Jacob Tomlinson and Matthew Rocklin

• Support async protocols for LocalCluster, replace start= with asynchronous= (GH#1798) Matthew Rocklin

• Avoid restarting workers when nanny waits on scheduler (GH#1793) Matthew Rocklin

• Use IOStream.read_into() when available (GH#1477) Antoine Pitrou

• Reduce LocalCluster logging threshold from CRITICAL to WARN (GH#1785) Andy Jones

• Add futures_of to API docs (GH#1783) John Kirkham

• Make diagnostics link in client configurable (GH#1810) Matthew Rocklin

1.21.1 - 2018-02-22

• Fixed an uncaught exception in distributed.joblib with a LocalCluster using only threads (GH#1775) Tom Augspurger

• Format bytes in info worker page (GH#1752) Matthew Rocklin

• Add pass-through arguments for scheduler/worker –preload modules. (GH#1634) Alex Ford

• Use new LZ4 API (GH#1757) Thrasibule

• Replace dask.optimize with dask.optimization (GH#1754) Matthew Rocklin

• Add graph layout engine and bokeh plot (GH#1756) Matthew Rocklin

• Only expand name with –nprocs if name exists (GH#1776) Matthew Rocklin

• specify IOLoop for stealing PeriodicCallback (GH#1777) Matthew Rocklin

• Fixed distributed.joblib with no processes Tom Augspurger

• Use set.discard to avoid KeyErrors in stealing (GH#1766) Matthew Rocklin

• Avoid KeyError when task has been released during steal (GH#1765) Matthew Rocklin

• Add versions routes to avoid the use of run in Client.get_versions (GH#1773) Matthew Rocklin

• Add write_scheduler_file to Client (GH#1778) Joe Hamman

• Default host to tls:// if tls information provided (GH#1780) Matthew Rocklin

1.21.0 - 2018-02-09

• Refactor scheduler to use TaskState objects rather than dictionaries (GH#1594) Antoine Pitrou

• Plot CPU fraction of total in workers page (GH#1624) Matthew Rocklin

• Use thread CPU time in Throttled GC (GH#1625) Antoine Pitrou

• Fix bug with memory_limit=None (GH#1639) Matthew Rocklin

• Add futures_of to top level api (GH#1646) Matthew Rocklin

• Warn on serializing large data in Client (GH#1636) Matthew Rocklin

• Fix intermittent windows failure when removing lock file (GH#1652) Antoine Pitrou

• Add diagnosis and logging of poor GC Behavior (GH#1635) Antoine Pitrou

• Add client-scheduler heartbeats (GH#1657) Matthew Rocklin

• Return dictionary of worker info in retire_workers (GH#1659) Matthew Rocklin

• Ensure dumps_function works with unhashable functions (GH#1662) Matthew Rocklin

• Collect client name ids rom client-name config variable (GH#1664) Matthew Rocklin

• Allow simultaneous use of –name and –nprocs in dask-worker (GH#1665) Matthew Rocklin

• Add support for grouped adaptive scaling and adaptive behavior overrides (GH#1632) Alex Ford

• Share scheduler RPC between worker and client (GH#1673) Matthew Rocklin

• Allow retries= in ClientExecutor (GH#1672) @rqx

• Improve documentation for get_client and dask.compute examples (GH#1638) Scott Sievert

• Support DASK_SCHEDULER_ADDRESS environment variable in worker (GH#1680) Matthew Rocklin

• Support tuple-keys in retries (GH#1681) Matthew Rocklin

• Use relative links in bokeh dashboard (GH#1682) Matthew Rocklin

• Make message log length configurable, default to zero (GH#1691) Matthew Rocklin

• Deprecate Client.shutdown (GH#1699) Matthew Rocklin

• Add warning in configuration docs to install pyyaml (GH#1701) Cornelius Riemenschneider

• Handle nested parallelism in distributed.joblib (GH#1705) Tom Augspurger

• Don’t wait for Worker.executor to shutdown cleanly when restarting process (GH#1708) Matthew Rocklin

• Add support for user defined priorities (GH#1651) Matthew Rocklin

• Catch and log OSErrors around worker lock files (GH#1714) Matthew Rocklin

• Remove worker prioritization. Coincides with changes to dask.order (GH#1730) Matthew Rocklin

• Use process-measured memory rather than nbytes in Bokeh dashboard (GH#1737) Matthew Rocklin

• Enable serialization of Locks (GH#1738) Matthew Rocklin

• Support Tornado 5 beta (GH#1735) Matthew Rocklin

• Cleanup remote_magic client cache after tests (GH#1743) Min RK

• Allow service ports to be specified as (host, port) (GH#1744) Bruce Merry

1.20.2 - 2017-12-07

• Clear deque handlers after each test (GH#1586) Antoine Pitrou

• Handle deserialization in FutureState.set_error (GH#1592) Matthew Rocklin

• Add process leak checker to tests (GH#1596) Antoine Pitrou

• Customize process title for subprocess (GH#1590) Antoine Pitrou

• Make linting a separate CI job (GH#1599) Antoine Pitrou

• Fix error from get_client() with no global client (GH#1595) Daniel Li

• Remove Worker.host_health, correct WorkerTable metrics (GH#1600) Matthew Rocklin

• Don’t mark tasks as suspicious when retire_workers called. Addresses (GH#1607) Russ Bubley

• Do not include processing workers in workers_to_close (GH#1609) Russ Bubley

• Disallow simultaneous scale up and down in Adaptive (GH#1608) Russ Bubley

• Parse bytestrings in –memory-limit (GH#1615) Matthew Rocklin

• Use environment variable for scheduler address if present (GH#1610) Matthew Rocklin

• Fix deprecation warning from logger.warn (GH#1616) Brett Naul

1.20.1 - 2017-11-26

• Wrap import ssl statements with try-except block for ssl-crippled environments, (GH#1570) Xander Johnson

• Support zero memory-limit in Nanny (GH#1571) Matthew Rocklin

• Avoid PeriodicCallback double starts (GH#1573) Matthew Rocklin

• Add disposable workspace facility (GH#1543) Antoine Pitrou

• Use format_time in task_stream plots (GH#1575) Matthew Rocklin

• Avoid delayed finalize calls in compute (GH#1577) Matthew Rocklin

• Doc fix about secede (GH#1583) Scott Sievert

• Add tracemalloc option when tracking test leaks (GH#1585) Antoine Pitrou

• Add JSON routes to Bokeh server (GH#1584) Matthew Rocklin

• Handle exceptions cleanly in Variables and Queues (GH#1580) Matthew Rocklin

1.20.0 - 2017-11-17

• Drop use of pandas.msgpack (GH#1473) Matthew Rocklin

• Add methods to get/set scheduler metadata Matthew Rocklin

• Add distributed lock Matthew Rocklin

• Add reschedule exception for worker tasks Matthew Rocklin

• Fix nbytes() for bytearrays Matthew Rocklin

• Capture scheduler and worker logs Matthew Rocklin

• Garbage collect after data eviction on high worker memory usage (GH#1488) Olivier Grisel

• Add scheduler HTML routes to bokeh server (GH#1478) (GH#1514) Matthew Rocklin

• Add pytest plugin to test for resource leaks (GH#1499) Antoine Pitrou

• Improve documentation for scheduler states (GH#1498) Antoine Pitrou

• Correct warn_if_longer timeout in ThrottledGC (GH#1496) Fabian Keller

• Catch race condition in as_completed on cancelled futures (GH#1507) Matthew Rocklin

• Transactional work stealing (GH#1489) (GH#1528) Matthew Rocklin

• Avoid forkserver in PyPy (GH#1509) Matthew Rocklin

• Add dict access to get/set datasets (GH#1508) Mike DePalatis

• Support Tornado 5 (GH#1509) (GH#1512) (GH#1518) (GH#1534) Antoine Pitrou

• Move thread_state in Dask (GH#1523) Jim Crist

• Use new Dask collections interface (GH#1513) Matthew Rocklin

• Add nanny flag to dask-mpi Matthew Rocklin

• Remove JSON-based HTTP servers Matthew Rocklin

• Avoid doing I/O in repr/str (GH#1536) Matthew Rocklin

• Fix URL for MPI4Py project (GH#1546) Ian Hopkinson

• Allow automatic retries of a failed task (GH#1524) Antoine Pitrou

• Clean and accelerate tests (GH#1548) (GH#1549) (GH#1552) (GH#1553) (GH#1560) (GH#1564) Antoine Pitrou

• Move HDFS functionality to the hdfs3 library (GH#1561) Jim Crist

• Fix bug when using events page with no events (GH#1562) @rbubley

• Improve diagnostic naming of tasks within tuples (GH#1566) Kelvyn Yang

1.19.3 - 2017-10-16

• Handle None case in profile.identity (GH#1456)

• Asyncio rewrite (GH#1458)

• Add rejoin function partner to secede (GH#1462)

• Nested compute (GH#1465)

• Use LooseVersion when comparing Bokeh versions (GH#1470)

1.19.2 - 2017-10-06

• as_completed doesn’t block on cancelled futures (GH#1436)

• Notify waiting threads/coroutines on cancellation (GH#1438)

• Set Future(inform=True) as default (GH#1437)

• Rename Scheduler.transition_story to story (GH#1445)

• Future uses default client by default (GH#1449)

• Add keys= keyword to Client.call_stack (GH#1446)

• Add get_current_task to worker (GH#1444)

• Ensure that Client remains asynchronous before ioloop starts (GH#1452)

• Remove “click for worker page” in bokeh plot (GH#1453)

• Add Client.current() (GH#1450)

• Clean handling of restart timeouts (GH#1442)

1.19.1 - 2017-09-25

• Fix tool issues with TaskStream plot (GH#1425)

• Move profile module to top level (GH#1423)

1.19.0 - 2017-09-24

• Avoid storing messages in message log (GH#1361)

• fileConfig does not disable existing loggers (GH#1380)

• Offload upload_file disk I/O to separate thread (GH#1383)

• Add missing SSLContext (GH#1385)

• Collect worker thread information from sys._curent_frames (GH#1387)

• Add nanny timeout (GH#1395)

• Restart worker if memory use goes above 95% (GH#1397)

• Track workers memory use with psutil (GH#1398)

• Track scheduler delay times in workers (GH#1400)

• Add time slider to profile plot (GH#1403)

• Change memory-limit keyword to refer to maximum number of bytes (GH#1405)

• Add cancel(force=) keyword (GH#1408)

1.18.2 - 2017-09-02

• Silently pass on cancelled futures in as_completed (GH#1366)

• Fix unicode keys error in Python 2 (GH#1370)

• Support numeric worker names

• Add dask-mpi executable (GH#1367)

1.18.1 - 2017-08-25

• Clean up forgotten keys in fire-and-forget workloads (GH#1250)

• Handle missing extensions (GH#1263)

• Allow recreate_exception on persisted collections (GH#1253)

• Add asynchronous= keyword to blocking client methods (GH#1272)

• Restrict to horizontal panning in bokeh plots (GH#1274)

• Rename client.shutdown to client.close (GH#1275)

• Avoid blocking on event loop (GH#1270)

• Avoid cloudpickle errors for Client.get_versions (GH#1279)

• Yield on Tornado IOStream.write futures (GH#1289)

• Assume async behavior if inside a sync statement (GH#1284)

• Avoid error messages on closing (GH#1297), (GH#1296) (GH#1318) (GH#1319)

• Add timeout= keyword to get_client (GH#1290)

• Respect timeouts when restarting (GH#1304)

• Clean file descriptor and memory leaks in tests (GH#1317)

• Deprecate Executor (GH#1302)

• Add timeout to ThreadPoolExecutor.shutdown (GH#1330)

• Clean up AsyncProcess handling (GH#1324)

• Allow unicode keys in Python 2 scheduler (GH#1328)

• Avoid leaking stolen data (GH#1326)

• Improve error handling on failed nanny starts (GH#1337), (GH#1331)

• Make Adaptive more flexible

• Support --contact-address and --listen-address in worker (GH#1278)

• Remove old dworker, dscheduler executables (GH#1355)

• Exit workers if nanny process fails (GH#1345)

• Auto pep8 and flake (GH#1353)

1.18.0 - 2017-07-08

• Multi-threading safety (GH#1191), (GH#1228), (GH#1229)

• Improve handling of byte counting (GH#1198) (GH#1224)

• Add get_client, secede functions, refactor worker-client relationship (GH#1201)

• Allow logging configuration using logging.dictConfig() (GH#1206) (GH#1211)

• Offload serialization and deserialization to separate thread (GH#1218)

• Support fire-and-forget tasks (GH#1221)

• Support bytestrings as keys (for Julia) (GH#1234)

• Resolve testing corner-cases (GH#1236), (GH#1237), (GH#1240), (GH#1241), (GH#1242), (GH#1244)

• Automatic use of scatter/gather(direct=True) in more cases (GH#1239)

1.17.1 - 2017-06-14

• Remove Python 3.4 testing from travis-ci (GH#1157)

• Remove ZMQ Support (GH#1160)

• Fix memoryview nbytes issue in Python 2.7 (GH#1165)

• Re-enable counters (GH#1168)

• Improve scheduler.restart (GH#1175)

1.17.0 - 2017-06-09

• Reevaluate worker occupancy periodically during scheduler downtime (GH#1038) (GH#1101)

• Add AioClient asyncio-compatible client API (GH#1029) (GH#1092) (GH#1099)

• Update Keras serializer (GH#1067)

• Support TLS/SSL connections for security (GH#866) (GH#1034)

• Always create new worker directory when passed --local-directory (GH#1079)

• Support pre-scattering data when using joblib frontend (GH#1022)

• Make workers more robust to failure of sizeof function (GH#1108) and writing to disk (GH#1096)

• Add is_empty and update methods to as_completed (GH#1113)

• Remove _get coroutine and replace with get(..., sync=False) (GH#1109)

• Improve API compatibility with async/await syntax (GH#1115) (GH#1124)

• Add distributed Queues (GH#1117) and shared Variables (GH#1128) to enable inter-client coordination

• Support direct client-to-worker scattering and gathering (GH#1130) as well as performance enhancements when scattering data

• Style improvements for bokeh web dashboards (GH#1126) (GH#1141) as well as a removal of the external bokeh process

• HTML reprs for Future and Client objects (GH#1136)

• Support nested collections in client.compute (GH#1144)

• Use normal client API in asynchronous mode (GH#1152)

• Remove old distributed.collections submodule (GH#1153)

1.16.3 - 2017-05-05

• Add bokeh template files to MANIFEST (GH#1063)

• Don’t set worker_client.get as default get (GH#1061)

• Clean up logging on Client().shutdown() (GH#1055)

1.16.2 - 2017-05-03

• Support async with Client syntax (GH#1053)

• Use internal bokeh server for default diagnostics server (GH#1047)

• Improve styling of bokeh plots when empty (GH#1046) (GH#1037)

• Support efficient serialization for sparse arrays (GH#1040)

• Prioritize newly arrived work in worker (GH#1035)

• Prescatter data with joblib backend (GH#1022)

• Make client.restart more robust to worker failure (GH#1018)

• Support preloading a module or script in dask-worker or dask-scheduler processes (GH#1016)

• Specify network interface in command line interface (GH#1007)

• Client.scatter supports a single element (GH#1003)

• Use blosc compression on all memoryviews passing through comms (GH#998)

• Add concurrent.futures-compatible Executor (GH#997)

• Add as_completed.batches method and return results (GH#994) (GH#971)

• Allow worker_clients to optionally stay within the thread pool (GH#993)

• Add bytes-stored and tasks-processing diagnostic histograms (GH#990)

• Run supports non-msgpack-serializable results (GH#965)

1.16.1 - 2017-03-22

• Use inproc transport in LocalCluster (GH#919)

• Add structured and queryable cluster event logs (GH#922)

• Use connection pool for inter-worker communication (GH#935)

• Robustly shut down spawned worker processes at shutdown (GH#928)

• Worker death timeout (GH#940)

• More visual reporting of exceptions in progressbar (GH#941)

• Render disk and serialization events to task stream visual (GH#943)

• Support async for / await protocol (GH#952)

• Ensure random generators are re-seeded in worker processes (GH#953)

• Upload sourcecode as zip module (GH#886)

• Replay remote exceptions in local process (GH#894)

1.16.0 - 2017-02-24

• First come first served priorities on client submissions (GH#840)

• Can specify Bokeh internal ports (GH#850)

• Allow stolen tasks to return from either worker (GH#853), (GH#875)

• Add worker resource constraints during execution (GH#857)

• Send small data through Channels (GH#858)

• Better estimates for SciPy sparse matrix memory costs (GH#863)

• Avoid stealing long running tasks (GH#873)

• Maintain fortran ordering of NumPy arrays (GH#876)

• Add --scheduler-file keyword to dask-scheduler (GH#877)

• Add serializer for Keras models (GH#878)

• Support uploading modules from zip files (GH#886)

• Improve titles of Bokeh dashboards (GH#895)

1.15.2 - 2017-01-27

• Fix a bug where arrays with large dtypes or shapes were being improperly compressed (GH#830 GH#832 GH#833)

• Extend as_completed to accept new futures during iteration (GH#829)

• Add --nohost keyword to dask-ssh startup utility (GH#827)

• Support scheduler shutdown of remote workers, useful for adaptive clusters (:pr: 811 GH#816 GH#821)

• Add Client.run_on_scheduler method for running debug functions on the scheduler (GH#808)

1.15.1 - 2017-01-11

• Make compatible with Bokeh 0.12.4 (GH#803)

• Avoid compressing arrays if not helpful (GH#777)

• Optimize inter-worker data transfer (GH#770) (GH#790)

• Add –local-directory keyword to worker (GH#788)

• Enable workers to arrive to the cluster with their own data. Useful if a worker leaves and comes back (GH#785)

• Resolve thread safety bug when using local_client (GH#802)

• Resolve scheduling issues in worker (GH#804)

1.15.0 - 2017-01-02

• Major Worker refactor (GH#704)

• Major Scheduler refactor (GH#717) (GH#722) (GH#724) (GH#742) (GH#743

• Add check (default is False) option to Client.get_versions to raise if the versions don’t match on client, scheduler & workers (GH#664)

• Future.add_done_callback executes in separate thread (GH#656)

• Clean up numpy s