Skip to content
  • v4.0
  • 9d19992
  • Compare
    Choose a tag to compare
    Search for a tag
  • v4.0
  • 9d19992
  • Compare
    Choose a tag to compare
    Search for a tag

@mdumandag mdumandag released this Jan 13, 2021 · 13 commits to master since this release

This document includes the new features, enhancements, and fixed issues for the Hazelcast Python Client 4.0 release.

New Features

  • Hazelcast IMDG 4.0 Changes: Client protocol enhancements, architectural changes. #217
  • Ownerless Client: Simpler design to track member leaves and joins to the cluster. #217
  • CP Subsystem Support: New concurrency APIs including AtomicLong, AtomicReference, FencedLock, Semaphore, and CountDownLatch. Hazelcast IMDG 4.0 introduces CP Subsystem powered by Raft consensus protocol. It provides CP concurrency primitives with respect to the CAP principle, i.e., they always maintain linearizability and prefer consistency to availability during network partitions and client or server failures. #223, #224, #225, #226, #227
  • Backup Acknowledgment: Python client now supports backup acks (aka boomerang backups). So far the client was waiting for the sync backups to complete on the member. Boomerang backups decrease the number of network hops, thus improving the performance. #222
  • Paging Predicates: Support for Paging Predicates which allows you to fetch keys, values, or entries of a map page by page is implemented. #232


  • Configuration Redesign: Programmatic configuration was simplified. The configuration is now represented with keyword arguments. #219
  • Google Style Docstrings: Docstrings now follow the Google Python Style Guide to improve the readability. Also, with this change, IDEs are better at showing the type-hints from the docstrings. #221
  • Logging: Hazelcast Python client was adding custom handlers to the loggers, which is not recommended by the standard library documentation. Custom handlers are now removed and other logging practices recommended by the standard library documentation is followed. #229
  • Performance: The performance of the client for the non-blocking mode is improved significantly. #242, #264
  • Documentation: API and detailed client documentation is moved to and README is simplified. #266
  • Predicates: Predicates are now properly documented, have shorter names, and moved from the serialization module. #239
  • Some private APIs were exposed to users mistakenly. They are removed from the public API and API documentation. #217, #221
  • Serialization of strings is now significantly faster. #217
  • Default serialization now supports the native bytearray type. #218
  • Default serialization now supports Java UUID type. #218
  • It is now possible to add or remove listeners in a non-blocking way. #248
  • Hot path of the client is optimized. #218, #241, #243, #244
  • Missing set_ttl method is added to the Map. #215
  • Missing ObjectDataInput#skip_bytes method is implemented. #254
  • Client now uses seconds as the time resolution for all APIs. #260
  • Hazelcast Cloud discovery documentation is improved. #263
  • Improved the responsiveness of the client in case of the reconnection scenarios. #274
  • Enum-like configurations now can be customized using string literals. #273
  • Source code of the client is now formatted using black. #284, #285, #286, #287, #288, #289


  • The member list could be broken after Hot Restart enabled member restart. Now the client handles the restart correctly. #231
  • Client now handles class definition registrations with the same class id but with different factory ids. #235
  • Anchor links were not working on the PyPI. README is converted to reStructuredText format to make them work. #237
  • The possible race on handling timers is fixed. #240
  • Client was not able to set the results of retried invocations in some cases. Now the client sets the results of the retried invocations correctly #268 #269
  • Client now handles the memberlist correctly in split-brain scenarios. #272

Breaking Changes

  • AP-based concurrency primitives (Lock, Semaphore, AtomicLong, AtomicReference, CountDownLatch) were removed from the client. Instead, CP-based versions are implemented. #223, #224, #225, #226, #227
  • Programmatic configuration API was changed to work with keyword arguments. Certain configuration options were renamed. Also, the configuration options are now flat. #219
  • Logging configuration API was changed according to recommended practices by the standard library. Hazelcast Python client does not add any handlers now. Also, one has to enable logging to see client logs. #229
  • Certain predicates are renamed. Also, they are now at the hazelcast.predicate module, instead of the hazelcast.serialization.predicate. #239
  • Some private APIs were exposed to users mistakenly. They are removed from the public API and API documentation. #217, #221, #293
  • Adding or removing listeners can be done in a non-blocking way now. Therefore, these APIs are returning Future[str] and Future[bool] instead of str and bool. #248
  • hazelcast.exception module is renamed as hazelcast.errors. #217

Known Issues

  • Client statistics format in Python client v4.0.0b1 is compatible with Management Center v4.0. Support for format compatible with Management Center v4.2020.08 and newer will be added in one of the upcoming minor releases of the client.
Assets 2