v2.4.0

This minor release exposes two new capabilities: to read from a specific time or logical offset and to publish to multiple topics using a single Producer object.

What's Changed

• Allow specifying logical offset or datetime start position on Consumer.stream() by @alchzh in #65
• Multi-topic publishing by @cnweaver in #69 . Users can continue to specify a single topic when constructing a Producer and have calls to write() automatically send messages to that topic, or can now specify the target topic when calling write().

Full Changelog: v2.3.2...v2.4.0

v2.3.2

This is a bug-fix release which updates the list of compatible dependency versions to avoid recent versions of librdkafka/confluent-kafka with known communications problems.

What's Changed

• Update confluent-kafka version to skip recent buggy releases by @lpsinger in #67
• Update doc strings and comments for confluent-kafka workarounds by @lpsinger in #68

Full Changelog: v2.3.1...v2.3.2

v2.3.1

This is a bug-fix release which reverts to using no compression by default when sending messages. Advanced users who wish to use compression can continue to do so by configuring the compression_type ProducerConfig option.

v2.3.0

This release adds new minor features (thanks to @lpsinger), but is otherwise fully interface backwards-compatible:

• #61: Enable Zstd compression of messages by default, with the option to choose any compression method supported by Apache Kafka.
• #62: Add the option to adjust the producer message.max.bytes. This can be important to be able to send messages larger than the default maximum, when the configured maximum on the broker is also larger than the default.

v2.2.0

This release includes bug fixes and improvements since v2.1.0:

• #60: The build system has been updated with a pyproject.toml file.
• #58: adc.Consumer now supports a stop method which can be used to end the stream loop conveniently from another thread.
• #55: adc.Consumer._stream_forever handles errors raised by librdkafka better.

v2.1.0: Add support for OIDC/OAUTHBEARER Authentication

This release includes #53 by @lpsinger which enables the new(ly enhanced) OIDC authentication mechanism supported by the latest (version 3.1.0) Apache Kafka release. When not using this new feature this release remains compatible with older Kafka versions.

Also included is a bug-fix to avoid crashing when attempting to read a topic with multiple partitions 'to its end' while new messages are being actively published.

v1.2.0: Improve KafkaException to show more low-level detail

This release includes #43, which adds more detail to the adc.errors.KafkaException class by exposing the low-level confluent_kafka.KafkaError directly on the exception so that callers can make judgements about the retriability of errors.

v1.1.0: Obey stored offsets

This release includes the changes from #42, so that offsets which are stored are used. This changes the external API in some backwards-incompatible ways, so perhaps this ought to be a 2.0.0 release if we were sticklers for semver, but we have few enough users to be pretty sure nobody was using the explicit offsets API.

v1.0.2: Fix conda build

This release fixes a misconfiguration of the conda build. It does not change anything user-facing.

v1.0.1: Higher default producer timeout

The default timeout for producing a message is raised in this release to 10 seconds from 2 seconds. Connecting to a Kafka broker with TLS and SCRAM authentication can take at least 2s normally with a round-trip latency of 100ms between the producer and broker, which resulted in the first message timing out consistently. 10s is a more reasonable default for producing to a broker over the internet.

In addition, #38 fixes a minor formatting bug when warning about reading from a LATEST offset, and #39 removes fastavro as a dependency because it is no longer used since v1.0.0.