Aiokafka

==================

Bugfix:

* Added error propagation from coordinator to main consumer. Before consumer
just stopped with error logged. (issue 294)
* Fix manual partition assignment, broken in 0.4.0 (issue 394)
* Fixed RecursionError in MessageAccumulator.add_message (issue 409)
* Update kafka-python to latest 1.4.3 and added support for Python3.7
* Dropped support for Python3.3 and Python3.4

Infrastructure:

* Added Kafka 1.0.2 broker for CI test runner
* Refactored travis CI build pipeline

==================

* Fix issue when offset commit error reports wrong partition in log (issue 353)
* Add ResourceWarning when Producer, Consumer or Connections are not closed
properly (issue 295)
* Fix Subscription None in GroupCoordinator._do_group_rejoin (issue 306)

==================

Major changes:

* Full refactor of the internals of AIOKafkaConsumer. Needed to avoid several
race conditions in code (PR 286, fixes 258, 264 and 261)
* Rewrote Records parsing protocol to allow implementation of newer protocol
versions later
* Added C extension for Records parsing protocol, boosting the speed of
produce/consume routines significantly
* Added an experimental batch producer API for unique cases, where user wants
to control batching himself (by shargan)


Minor changes:

* Add `timestamp` field to produced message's metadata. This is needed to find
LOG_APPEND_TIME configured timestamps.
* `Consumer.seek()` and similar API's now raise proper ``ValueError``'s on
validation failure instead of ``AssertionError``.


Bug fixes:

* Fix ``connections_max_idle_ms`` option, as earlier it was only applied to
bootstrap socket. (PR 299)
* Fix ``consumer.stop()`` side effect of logging an exception
ConsumerStoppedError (issue 263)
* Problem with Producer not able to recover from broker failure (issue 267)
* Traceback containing duplicate entries due to exception sharing (PR 247
by Artimi)
* Concurrent record consumption rasing `InvalidStateError('Exception is not
set.')` (PR 249 by aerkert)
* Don't fail ``GroupCoordinator._on_join_prepare()`` if ``commit_offset()``
throws exception (PR 230 by shargan)
* Send session_timeout_ms to GroupCoordinator constructor (PR 229 by shargan)

Big thanks to:

* shargan for Producer speed enhancements and the batch produce API
proposal/implementation.
* vineet-rh and other contributors for constant feedback on Consumer
problems, leading to the refactor mentioned above.

==================

* Added `AIOKafkaProducer.flush()` method. (PR 209 by vineet-rh)
* Fixed a bug with uvloop involving `float("inf")` for timeout. (PR 210 by
dmitry-moroz)
* Changed test runner to allow running tests on OSX. (PR 213 by shargan)

==================

* Moved all public structures and errors to `aiokafka` namespace. You will no
longer need to import from `kafka` namespace.
* Changed ConsumerRebalanceListener to support either function or coroutine
for `on_partitions_assigned` and `on_partitions_revoked` callbacks. (PR 190
by ask)
* Added support for `offsets_for_times`, `beginning_offsets`, `end_offsets`
API's. (issue 164)
* Coordinator requests are now sent using a separate socket. Fixes slow commit
issue. (issuer 137, issue 128)
* Added `seek_to_end`, `seek_to_beginning` API's. (issue 154)
* Updated documentation to provide more useful usage guide on both Consumer and
Producer interface.

==================

* Fixed retry problem in Producer, when buffer is not reset to 0 offset.
Thanks to ngavrysh for the fix in Tubular/aiokafka fork. (issue 184)
* Fixed how Producer handles retries on Leader node failure. It just did not
work before... Thanks to blugowski for the help in locating the problem.
(issue 176, issue 173)
* Fixed degrade in v0.2.2 on Consumer with no group_id. (issue 166)