Skip to content

Latest commit

 

History

History
999 lines (800 loc) · 48 KB

CHANGELOG.md

File metadata and controls

999 lines (800 loc) · 48 KB
Raw

librdkafka v1.9.2

librdkafka v1.9.2 is a maintenance release:

  • The SASL OAUTHBEAR OIDC POST field was sometimes truncated by one byte (#3192).
  • The bundled version of OpenSSL has been upgraded to version 1.1.1q for non-Windows builds. Windows builds remain on OpenSSL 1.1.1n for the time being.
  • The bundled version of Curl has been upgraded to version 7.84.0.

librdkafka v1.9.1

librdkafka v1.9.1 is a maintenance release:

  • The librdkafka.redist NuGet package now contains OSX M1/arm64 builds.
  • Self-contained static libraries can now be built on OSX M1 too, thanks to disabling curl's configure runtime check.

librdkafka v1.9.0

librdkafka v1.9.0 is a feature release:

  • Added KIP-768 OUATHBEARER OIDC support (by @jliunyu, #3560)
  • Added KIP-140 Admin API ACL support (by @emasab, #2676)

Upgrade considerations

  • Consumer: rd_kafka_offsets_store() (et.al) will now return an error for any partition that is not currently assigned (through rd_kafka_*assign()). This prevents a race condition where an application would store offsets after the assigned partitions had been revoked (which resets the stored offset), that could cause these old stored offsets to be committed later when the same partitions were assigned to this consumer again - effectively overwriting any committed offsets by any consumers that were assigned the same partitions previously. This would typically result in the offsets rewinding and messages to be reprocessed. As an extra effort to avoid this situation the stored offset is now also reset when partitions are assigned (through rd_kafka_*assign()). Applications that explicitly call ..offset*_store() will now need to handle the case where RD_KAFKA_RESP_ERR__STATE is returned in the per-partition .err field - meaning the partition is no longer assigned to this consumer and the offset could not be stored for commit.

Enhancements

  • Improved producer queue scheduling. Fixes the performance regression introduced in v1.7.0 for some produce patterns. (#3538, #2912)
  • Windows: Added native Win32 IO/Queue scheduling. This removes the internal TCP loopback connections that were previously used for timely queue wakeups.
  • Added socket.connection.setup.timeout.ms (default 30s). The maximum time allowed for broker connection setups (TCP connection as well as SSL and SASL handshakes) is now limited to this value. This fixes the issue with stalled broker connections in the case of network or load balancer problems. The Java clients has an exponential backoff to this timeout which is limited by socket.connection.setup.timeout.max.ms - this was not implemented in librdkafka due to differences in connection handling and ERR__ALL_BROKERS_DOWN error reporting. Having a lower initial connection setup timeout and then increase the timeout for the next attempt would yield possibly false-positive ERR__ALL_BROKERS_DOWN too early.
  • SASL OAUTHBEARER refresh callbacks can now be scheduled for execution on librdkafka's background thread. This solves the problem where an application has a custom SASL OAUTHBEARER refresh callback and thus needs to call rd_kafka_poll() (et.al.) at least once to trigger the refresh callback before being able to connect to brokers. With the new rd_kafka_conf_enable_sasl_queue() configuration API and rd_kafka_sasl_background_callbacks_enable() the refresh callbacks can now be triggered automatically on the librdkafka background thread.
  • rd_kafka_queue_get_background() now creates the background thread if not already created.
  • Added rd_kafka_consumer_close_queue() and rd_kafka_consumer_closed(). This allow applications and language bindings to implement asynchronous consumer close.
  • Bundled zlib upgraded to version 1.2.12.
  • Bundled OpenSSL upgraded to 1.1.1n.
  • Added test.mock.broker.rtt to simulate RTT/latency for mock brokers.

Fixes

General fixes

  • Fix various 1 second delays due to internal broker threads blocking on IO even though there are events to handle. These delays could be seen randomly in any of the non produce/consume request APIs, such as commit_transaction(), list_groups(), etc.
  • Windows: some applications would crash with an error message like no OPENSSL_Applink() written to the console if ssl.keystore.location was configured. This regression was introduced in v1.8.0 due to use of vcpkgs and how keystore file was read. #3554.
  • Windows 32-bit only: 64-bit atomic reads were in fact not atomic and could in rare circumstances yield incorrect values. One manifestation of this issue was the max.poll.interval.ms consumer timer expiring even though the application was polling according to profile. Fixed by @WhiteWind (#3815).
  • rd_kafka_clusterid() would previously fail with timeout if called on cluster with no visible topics (#3620). The clusterid is now returned as soon as metadata has been retrieved.
  • Fix hang in rd_kafka_list_groups() if there are no available brokers to connect to (#3705).
  • Millisecond timeouts (timeout_ms) in various APIs, such as rd_kafka_poll(), was limited to roughly 36 hours before wrapping. (#3034)
  • If a metadata request triggered by rd_kafka_metadata() or consumer group rebalancing encountered a non-retriable error it would not be propagated to the caller and thus cause a stall or timeout, this has now been fixed. (@aiquestion, #3625)
  • AdminAPI DeleteGroups() and DeleteConsumerGroupOffsets(): if the given coordinator connection was not up by the time these calls were initiated and the first connection attempt failed then no further connection attempts were performed, ulimately leading to the calls timing out. This is now fixed by keep retrying to connect to the group coordinator until the connection is successful or the call times out. Additionally, the coordinator will be now re-queried once per second until the coordinator comes up or the call times out, to detect change in coordinators.
  • Mock cluster rd_kafka_mock_broker_set_down() would previously accept and then disconnect new connections, it now refuses new connections.

Consumer fixes

  • rd_kafka_offsets_store() (et.al) will now return an error for any partition that is not currently assigned (through rd_kafka_*assign()). See Upgrade considerations above for more information.
  • rd_kafka_*assign() will now reset/clear the stored offset. See Upgrade considerations above for more information.
  • seek() followed by pause() would overwrite the seeked offset when later calling resume(). This is now fixed. (#3471). Note: Avoid storing offsets (offsets_store()) after calling seek() as this may later interfere with resuming a paused partition, instead store offsets prior to calling seek.
  • A ERR_MSG_SIZE_TOO_LARGE consumer error would previously be raised if the consumer received a maximum sized FetchResponse only containing (transaction) aborted messages with no control messages. The fetching did not stop, but some applications would terminate upon receiving this error. No error is now raised in this case. (#2993) Thanks to @jacobmikesell for providing an application to reproduce the issue.
  • The consumer no longer backs off the next fetch request (default 500ms) when the parsed fetch response is truncated (which is a valid case). This should speed up the message fetch rate in case of maximum sized fetch responses.
  • Fix consumer crash (assert: rkbuf->rkbuf_rkb) when parsing malformed JoinGroupResponse consumer group metadata state.
  • Fix crash (cant handle op type) when using consume_batch_queue() (et.al) and an OAUTHBEARER refresh callback was set. The callback is now triggered by the consume call. (#3263)
  • Fix partition.assignment.strategy ordering when multiple strategies are configured. If there is more than one eligible strategy, preference is determined by the configured order of strategies. The partitions are assigned to group members according to the strategy order preference now. (#3818)
  • Any form of unassign*() (absolute or incremental) is now allowed during consumer close rebalancing and they're all treated as absolute unassigns. (@kevinconaway)

Transactional producer fixes

  • Fix message loss in idempotent/transactional producer. A corner case has been identified that may cause idempotent/transactional messages to be lost despite being reported as successfully delivered: During cluster instability a restarting broker may report existing topics as non-existent for some time before it is able to acquire up to date cluster and topic metadata. If an idempotent/transactional producer updates its topic metadata cache from such a broker the producer will consider the topic to be removed from the cluster and thus remove its local partition objects for the given topic. This also removes the internal message sequence number counter for the given partitions. If the producer later receives proper topic metadata for the cluster the previously "removed" topics will be rediscovered and new partition objects will be created in the producer. These new partition objects, with no knowledge of previous incarnations, would start counting partition messages at zero again. If new messages were produced for these partitions by the same producer instance, the same message sequence numbers would be sent to the broker. If the broker still maintains state for the producer's PID and Epoch it could deem that these messages with reused sequence numbers had already been written to the log and treat them as legit duplicates. This would seem to the producer that these new messages were successfully written to the partition log by the broker when they were in fact discarded as duplicates, leading to silent message loss. The fix included in this release is to save the per-partition idempotency state when a partition is removed, and then recover and use that saved state if the partition comes back at a later time.
  • The transactional producer would retry (re)initializing its PID if a PRODUCER_FENCED error was returned from the broker (added in Apache Kafka 2.8), which could cause the producer to seemingly hang. This error code is now correctly handled by raising a fatal error.
  • If the given group coordinator connection was not up by the time send_offsets_to_transactions() was called, and the first connection attempt failed then no further connection attempts were performed, ulimately leading to send_offsets_to_transactions() timing out, and possibly also the transaction timing out on the transaction coordinator. This is now fixed by keep retrying to connect to the group coordinator until the connection is successful or the call times out. Additionally, the coordinator will be now re-queried once per second until the coordinator comes up or the call times out, to detect change in coordinators.

Producer fixes

  • Improved producer queue wakeup scheduling. This should significantly decrease the number of wakeups and thus syscalls for high message rate producers. (#3538, #2912)
  • The logic for enforcing that message.timeout.ms is greather than an explicitly configured linger.ms was incorrect and instead of erroring out early the lingering time was automatically adjusted to the message timeout, ignoring the configured linger.ms. This has now been fixed so that an error is returned when instantiating the producer. Thanks to @larry-cdn77 for analysis and test-cases. (#3709)

librdkafka v1.8.2

librdkafka v1.8.2 is a maintenance release.

Enhancements

  • Added ssl.ca.pem to add CA certificate by PEM string. (#2380)
  • Prebuilt binaries for Mac OSX now contain statically linked OpenSSL v1.1.1l. Previously the OpenSSL version was either v1.1.1 or v1.0.2 depending on build type.

Fixes

  • The librdkafka.redist 1.8.0 package had two flaws:
    • the linux-arm64 .so build was a linux-x64 build.
    • the included Windows MSVC 140 runtimes for x64 were infact x86. The release script has been updated to verify the architectures of provided artifacts to avoid this happening in the future.
  • Prebuilt binaries for Mac OSX Sierra (10.12) and older are no longer provided. This affects confluent-kafka-go.
  • Some of the prebuilt binaries for Linux were built on Ubuntu 14.04, these builds are now performed on Ubuntu 16.04 instead. This may affect users on ancient Linux distributions.
  • It was not possible to configure ssl.ca.location on OSX, the property would automatically revert back to probe (default value). This regression was introduced in v1.8.0. (#3566)
  • librdkafka's internal timers would not start if the timeout was set to 0, which would result in some timeout operations not being enforced correctly, e.g., the transactional producer API timeouts. These timers are now started with a timeout of 1 microsecond.

Transactional producer fixes

  • Upon quick repeated leader changes the transactional producer could receive an OUT_OF_ORDER_SEQUENCE error from the broker, which triggered an Epoch bump on the producer resulting in an InitProducerIdRequest being sent to the transaction coordinator in the middle of a transaction. This request would start a new transaction on the coordinator, but the producer would still think (erroneously) it was in current transaction. Any messages produced in the current transaction prior to this event would be silently lost when the application committed the transaction, leading to message loss. This has been fixed by setting the Abortable transaction error state in the producer. #3575.
  • The transactional producer could stall during a transaction if the transaction coordinator changed while adding offsets to the transaction (send_offsets_to_transaction()). This stall lasted until the coordinator connection went down, the transaction timed out, transaction was aborted, or messages were produced to a new partition, whichever came first. #3571.

Note: there was no v1.8.1 librdkafka release

librdkafka v1.8.0

librdkafka v1.8.0 is a security release:

  • Upgrade bundled zlib version from 1.2.8 to 1.2.11 in the librdkafka.redist NuGet package. The updated zlib version fixes CVEs: CVE-2016-9840, CVE-2016-9841, CVE-2016-9842, CVE-2016-9843 See confluentinc#2934 for more information.
  • librdkafka now uses vcpkg for up-to-date Windows dependencies in the librdkafka.redist NuGet package: OpenSSL 1.1.1l, zlib 1.2.11, zstd 1.5.0.
  • The upstream dependency (OpenSSL, zstd, zlib) source archive checksums are now verified when building with ./configure --install-deps. These builds are used by the librdkafka builds bundled with confluent-kafka-go, confluent-kafka-python and confluent-kafka-dotnet.

Enhancements

  • Producer flush() now overrides the linger.ms setting for the duration of the flush() call, effectively triggering immediate transmission of queued messages. (#3489)
  • Added AWS_MSK_IAM to supported sasl.mechanisms. This feature provides support for using IAM authentication on AWS MSK clusters. (@garrett528, #3402)

Fixes

General fixes

  • Correctly detect presence of zlib via compilation check. (Chris Novakovic)
  • ERR__ALL_BROKERS_DOWN is no longer emitted when the coordinator connection goes down, only when all standard named brokers have been tried. This fixes the issue with ERR__ALL_BROKERS_DOWN being triggered on consumer_close(). It is also now only emitted if the connection was fully up (past handshake), and not just connected.
  • rd_kafka_query_watermark_offsets(), rd_kafka_offsets_for_times(), consumer_lag metric, and auto.offset.reset now honour isolation.level and will return the Last Stable Offset (LSO) when isolation.level is set to read_committed (default), rather than the uncommitted high-watermark when it is set to read_uncommitted. (#3423)
  • SASL GSSAPI is now usable when sasl.kerberos.min.time.before.relogin is set to 0 - which disables ticket refreshes (by @mpekalski, #3431).
  • Rename internal crc32c() symbol to rd_crc32c() to avoid conflict with other static libraries (#3421).
  • txidle and rxidle in the statistics object was emitted as 18446744073709551615 when no idle was known. -1 is now emitted instead. (#3519)

Consumer fixes

  • Automatically retry offset commits on ERR_REQUEST_TIMED_OUT, ERR_COORDINATOR_NOT_AVAILABLE, and ERR_NOT_COORDINATOR (#3398). Offset commits will be retried twice.
  • Timed auto commits did not work when only using assign() and not subscribe(). This regression was introduced in v1.7.0.
  • If the topics matching the current subscription changed (or the application updated the subscription) while there was an outstanding JoinGroup or SyncGroup request, an additional request would sometimes be sent before handling the response of the first. This in turn lead to internal state issues that could cause a crash or malbehaviour. The consumer will now wait for any outstanding JoinGroup or SyncGroup responses before re-joining the group.
  • auto.offset.reset could previously be triggered by temporary errors, such as disconnects and timeouts (after the two retries are exhausted). This is now fixed so that the auto offset reset policy is only triggered for permanent errors.
  • The error that triggers auto.offset.reset is now logged to help the application owner identify the reason of the reset.
  • If a rebalance takes longer than a consumer's session.timeout.ms, the consumer will remain in the group as long as it receives heartbeat responses from the broker.

Admin fixes

  • DeleteRecords() could crash if one of the underlying requests (for a given partition leader) failed at the transport level (e.g., timeout). (#3476).

librdkafka v1.7.0

librdkafka v1.7.0 is feature release:

  • KIP-360 - Improve reliability of transactional producer. Requires Apache Kafka 2.5 or later.
  • OpenSSL Engine support (ssl.engine.location) by @adinigam and @ajbarb.

Enhancements

  • Added connections.max.idle.ms to automatically close idle broker connections. This feature is disabled by default unless bootstrap.servers contains the string azure in which case the default is set to <4 minutes to improve connection reliability and circumvent limitations with the Azure load balancers (see #3109 for more information).
  • Bumped to OpenSSL 1.1.1k in binary librdkafka artifacts.
  • The binary librdkafka artifacts for Alpine are now using Alpine 3.12. OpenSSL 1.1.1k.
  • Improved static librdkafka Windows builds using MinGW (@neptoess, #3130).
  • The librdkafka.redist NuGet package now has updated zlib, zstd and OpenSSL versions (from vcpkg).

Security considerations

  • The zlib version bundled with the librdkafka.redist NuGet package has now been upgraded from zlib 1.2.8 to 1.2.11, fixing the following CVEs:
    • CVE-2016-9840: undefined behaviour (compiler dependent) in inflate (decompression) code: this is used by the librdkafka consumer. Risk of successfully exploitation through consumed messages is eastimated very low.
    • CVE-2016-9841: undefined behaviour (compiler dependent) in inflate code: this is used by the librdkafka consumer. Risk of successfully exploitation through consumed messages is eastimated very low.
    • CVE-2016-9842: undefined behaviour in inflateMark(): this API is not used by librdkafka.
    • CVE-2016-9843: issue in crc32_big() which is called from crc32_z(): this API is not used by librdkafka.

Upgrade considerations

  • The C++ oauthbearer_token_refresh_cb() was missing a Handle * argument that has now been added. This is a breaking change but the original function signature is considered a bug. This change only affects C++ OAuth developers.
  • KIP-735 The consumer session.timeout.ms default was changed from 10 to 45 seconds to make consumer groups more robust and less sensitive to temporary network and cluster issues.
  • Statistics: consumer_lag is now using the committed_offset, while the new consumer_lag_stored is using stored_offset (offset to be committed). This is more correct than the previous consumer_lag which was using either committed_offset or app_offset (last message passed to application).
  • The librdkafka.redist NuGet package is now built with MSVC runtime v140 (VS 2015). Previous versions were built with MSVC runtime v120 (VS 2013).

Fixes

General fixes

  • Fix accesses to freed metadata cache mutexes on client termination (#3279)
  • There was a race condition on receiving updated metadata where a broker id update (such as bootstrap to proper broker transformation) could finish after the topic metadata cache was updated, leading to existing brokers seemingly being not available. One occurrence of this issue was query_watermark_offsets() that could return ERR__UNKNOWN_PARTITION for existing partitions shortly after the client instance was created.
  • The OpenSSL context is now initialized with TLS_client_method() (on OpenSSL >= 1.1.0) instead of the deprecated and outdated SSLv23_client_method().
  • The initial cluster connection on client instance creation could sometimes be delayed up to 1 second if a group.id or transactional.id was configured (#3305).
  • Speed up triggering of new broker connections in certain cases by exiting the broker thread io/op poll loop when a wakeup op is received.
  • SASL GSSAPI: The Kerberos kinit refresh command was triggered from rd_kafka_new() which made this call blocking if the refresh command was taking long. The refresh is now performed by the background rdkafka main thread.
  • Fix busy-loop (100% CPU on the broker threads) during the handshake phase of an SSL connection.
  • Disconnects during SSL handshake are now propagated as transport errors rather than SSL errors, since these disconnects are at the transport level (e.g., incorrect listener, flaky load balancer, etc) and not due to SSL issues.
  • Increment metadata fast refresh interval backoff exponentially (@ajbarb, #3237).
  • Unthrottled requests are no longer counted in the brokers[].throttle statistics object.
  • Log CONFWARN warning when global topic configuration properties are overwritten by explicitly setting a default_topic_conf.

Consumer fixes

  • If a rebalance happened during a consume_batch..() call the already accumulated messages for revoked partitions were not purged, which would pass messages to the application for partitions that were no longer owned by the consumer. Fixed by @jliunyu. #3340.
  • Fix balancing and reassignment issues with the cooperative-sticky assignor. #3306.
  • Fix incorrect detection of first rebalance in sticky assignor (@hallfox).
  • Aborted transactions with no messages produced to a partition could cause further successfully committed messages in the same Fetch response to be ignored, resulting in consumer-side message loss. A log message along the lines Abort txn ctrl msg bad order at offset 7501: expected before or at 7702: messages in aborted transactions may be delivered to the application would be seen. This is a rare occurrence where a transactional producer would register with the partition but not produce any messages before aborting the transaction.
  • The consumer group deemed cached metadata up to date by checking topic.metadata.refresh.interval.ms: if this property was set too low it would cause cached metadata to be unusable and new metadata to be fetched, which could delay the time it took for a rebalance to settle. It now correctly uses metadata.max.age.ms instead.
  • The consumer group timed auto commit would attempt commits during rebalances, which could result in "Illegal generation" errors. This is now fixed, the timed auto committer is only employed in the steady state when no rebalances are taking places. Offsets are still auto committed when partitions are revoked.
  • Retriable FindCoordinatorRequest errors are no longer propagated to the application as they are retried automatically.
  • Fix rare crash (assert rktp_started) on consumer termination (introduced in v1.6.0).
  • Fix unaligned access and possibly corrupted snappy decompression when building with MSVC (@azat)
  • A consumer configured with the cooperative-sticky assignor did not actively Leave the group on unsubscribe(). This delayed the rebalance for the remaining group members by up to session.timeout.ms.
  • The current subscription list was sometimes leaked when unsubscribing.

Producer fixes

  • The timeout value of flush() was not respected when delivery reports were scheduled as events (such as for confluent-kafka-go) rather than callbacks.
  • There was a race conditition in purge() which could cause newly created partition objects, or partitions that were changing leaders, to not have their message queues purged. This could cause abort_transaction() to time out. This issue is now fixed.
  • In certain high-thruput produce rate patterns producing could stall for 1 second, regardless of linger.ms, due to rate-limiting of internal queue wakeups. This is now fixed by not rate-limiting queue wakeups but instead limiting them to one wakeup per queue reader poll. #2912.

Transactional Producer fixes

  • KIP-360: Fatal Idempotent producer errors are now recoverable by the transactional producer and will raise a txn_requires_abort() error.
  • If the cluster went down between produce() and commit_transaction() and before any partitions had been registered with the coordinator, the messages would time out but the commit would succeed because nothing had been sent to the coordinator. This is now fixed.
  • If the current transaction failed while commit_transaction() was checking the current transaction state an invalid state transaction could occur which in turn would trigger a assertion crash. This issue showed up as "Invalid txn state transition: .." crashes, and is now fixed by properly synchronizing both checking and transition of state.

librdkafka v1.6.1

librdkafka v1.6.1 is a maintenance release.

Upgrade considerations

  • Fatal idempotent producer errors are now also fatal to the transactional producer. This is a necessary step to maintain data integrity prior to librdkafka supporting KIP-360. Applications should check any transactional API errors for the is_fatal flag and decommission the transactional producer if the flag is set.
  • The consumer error raised by auto.offset.reset=error now has error-code set to ERR__AUTO_OFFSET_RESET to allow an application to differentiate between auto offset resets and other consumer errors.

Fixes

General fixes

  • Admin API and transactional send_offsets_to_transaction() coordinator requests, such as TxnOffsetCommitRequest, could in rare cases be sent multiple times which could cause a crash.
  • ssl.ca.location=probe is now enabled by default on Mac OSX since the librdkafka-bundled OpenSSL might not have the same default CA search paths as the system or brew installed OpenSSL. Probing scans all known locations.

Transactional Producer fixes

  • Fatal idempotent producer errors are now also fatal to the transactional producer.
  • The transactional producer could crash if the transaction failed while send_offsets_to_transaction() was called.
  • Group coordinator requests for transactional send_offsets_to_transaction() calls would leak memory if the underlying request was attempted to be sent after the transaction had failed.
  • When gradually producing to multiple partitions (resulting in multiple underlying AddPartitionsToTxnRequests) subsequent partitions could get stuck in pending state under certain conditions. These pending partitions would not send queued messages to the broker and eventually trigger message timeouts, failing the current transaction. This is now fixed.
  • Committing an empty transaction (no messages were produced and no offsets were sent) would previously raise a fatal error due to invalid state on the transaction coordinator. We now allow empty/no-op transactions to be committed.

Consumer fixes

  • The consumer will now retry indefinitely (or until the assignment is changed) to retrieve committed offsets. This fixes the issue where only two retries were attempted when outstanding transactions were blocking OffsetFetch requests with ERR_UNSTABLE_OFFSET_COMMIT. #3265

librdkafka v1.6.0

librdkafka v1.6.0 is feature release:

  • KIP-429 Incremental rebalancing with sticky consumer group partition assignor (KIP-54) (by @mhowlett).
  • KIP-480 Sticky producer partitioning (sticky.partitioning.linger.ms) - achieves higher throughput and lower latency through sticky selection of random partition (by @abbycriswell).
  • AdminAPI: Add support for DeleteRecords(), DeleteGroups() and DeleteConsumerGroupOffsets() (by @gridaphobe)
  • KIP-447 Producer scalability for exactly once semantics - allows a single transactional producer to be used for multiple input partitions. Requires Apache Kafka 2.5 or later.
  • Transactional producer fixes and improvements, see Transactional Producer fixes below.
  • The librdkafka.redist NuGet package now supports Linux ARM64/Aarch64.

Upgrade considerations

  • Sticky producer partitioning (sticky.partitioning.linger.ms) is enabled by default (10 milliseconds) which affects the distribution of randomly partitioned messages, where previously these messages would be evenly distributed over the available partitions they are now partitioned to a single partition for the duration of the sticky time (10 milliseconds by default) before a new random sticky partition is selected.
  • The new KIP-447 transactional producer scalability guarantees are only supported on Apache Kafka 2.5 or later, on earlier releases you will need to use one producer per input partition for EOS. This limitation is not enforced by the producer or broker.
  • Error handling for the transactional producer has been improved, see the Transactional Producer fixes below for more information.

Known issues

  • The Transactional Producer's API timeout handling is inconsistent with the underlying protocol requests, it is therefore strongly recommended that applications call rd_kafka_commit_transaction() and rd_kafka_abort_transaction() with the timeout_ms parameter set to -1, which will use the remaining transaction timeout.

Enhancements

  • KIP-107, KIP-204: AdminAPI: Added DeleteRecords() (by @gridaphobe).
  • KIP-229: AdminAPI: Added DeleteGroups() (by @gridaphobe).
  • KIP-496: AdminAPI: Added DeleteConsumerGroupOffsets().
  • KIP-464: AdminAPI: Added support for broker-side default partition count and replication factor for CreateTopics().
  • Windows: Added ssl.ca.certificate.stores to specify a list of Windows Certificate Stores to read CA certificates from, e.g., CA,Root. Root remains the default store.
  • Use reentrant rand_r() on supporting platforms which decreases lock contention (@azat).
  • Added assignor debug context for troubleshooting consumer partition assignments.
  • Updated to OpenSSL v1.1.1i when building dependencies.
  • Update bundled lz4 (used when ./configure --disable-lz4-ext) to v1.9.3 which has vast performance improvements.
  • Added rd_kafka_conf_get_default_topic_conf() to retrieve the default topic configuration object from a global configuration object.
  • Added conf debugging context to debug - shows set configuration properties on client and topic instantiation. Sensitive properties are redacted.
  • Added rd_kafka_queue_yield() to cancel a blocking queue call.
  • Will now log a warning when multiple ClusterIds are seen, which is an indication that the client might be erroneously configured to connect to multiple clusters which is not supported.
  • Added rd_kafka_seek_partitions() to seek multiple partitions to per-partition specific offsets.

Fixes

General fixes

  • Fix a use-after-free crash when certain coordinator requests were retried.
  • The C++ oauthbearer_set_token() function would call free() on a new-created pointer, possibly leading to crashes or heap corruption (#3194)

Consumer fixes

  • The consumer assignment and consumer group implementations have been decoupled, simplified and made more strict and robust. This will sort out a number of edge cases for the consumer where the behaviour was previously undefined.
  • Partition fetch state was not set to STOPPED if OffsetCommit failed.
  • The session timeout is now enforced locally also when the coordinator connection is down, which was not previously the case.

Transactional Producer fixes

  • Transaction commit or abort failures on the broker, such as when the producer was fenced by a newer instance, were not propagated to the application resulting in failed commits seeming successful. This was a critical race condition for applications that had a delay after producing messages (or sendings offsets) before committing or aborting the transaction. This issue has now been fixed and test coverage improved.
  • The transactional producer API would return RD_KAFKA_RESP_ERR__STATE when API calls were attempted after the transaction had failed, we now try to return the error that caused the transaction to fail in the first place, such as RD_KAFKA_RESP_ERR__FENCED when the producer has been fenced, or RD_KAFKA_RESP_ERR__TIMED_OUT when the transaction has timed out.
  • Transactional producer retry count for transactional control protocol requests has been increased from 3 to infinite, retriable errors are now automatically retried by the producer until success or the transaction timeout is exceeded. This fixes the case where rd_kafka_send_offsets_to_transaction() would fail the current transaction into an abortable state when CONCURRENT_TRANSACTIONS was returned by the broker (which is a transient error) and the 3 retries were exhausted.

Producer fixes

  • Calling rd_kafka_topic_new() with a topic config object with message.timeout.ms set could sometimes adjust the global linger.ms property (if not explicitly configured) which was not desired, this is now fixed and the auto adjustment is only done based on the default_topic_conf at producer creation.
  • rd_kafka_flush() could previously return RD_KAFKA_RESP_ERR__TIMED_OUT just as the timeout was reached if the messages had been flushed but there were now no more messages. This has been fixed.

librdkafka v1.5.3

librdkafka v1.5.3 is a maintenance release.

Upgrade considerations

  • CentOS 6 is now EOL and is no longer included in binary librdkafka packages, such as NuGet.

Fixes

General fixes

  • Fix a use-after-free crash when certain coordinator requests were retried.
  • Coordinator requests could be left uncollected on instance destroy which could lead to hang.
  • Fix rare 1 second stalls by forcing rdkafka main thread wakeup when a new next-timer-to-be-fired is scheduled.
  • Fix additional cases where broker-side automatic topic creation might be triggered unexpectedly.
  • AdminAPI: The operation_timeout (on-broker timeout) previously defaulted to 0, but now defaults to socket.timeout.ms (60s).
  • Fix possible crash for Admin API protocol requests that fail at the transport layer or prior to sending.

Consumer fixes

  • Consumer would not filter out messages for aborted transactions if the messages were compressed (#3020).
  • Consumer destroy without prior close() could hang in certain cgrp states (@gridaphobe, #3127).
  • Fix possible null dereference in Message::errstr() (#3140).
  • The roundrobin partition assignment strategy could get stuck in an endless loop or generate uneven assignments in case the group members had asymmetric subscriptions (e.g., c1 subscribes to t1,t2 while c2 subscribes to t2,t3). (#3159)
  • Mixing committed and logical or absolute offsets in the partitions passed to rd_kafka_assign() would in previous released ignore the logical or absolute offsets and use the committed offsets for all partitions. This is now fixed. (#2938)

librdkafka v1.5.2

librdkafka v1.5.2 is a maintenance release.

Upgrade considerations

  • The default value for the producer configuration property retries has been increased from 2 to infinity, effectively limiting Produce retries to only message.timeout.ms. As the reasons for the automatic internal retries vary (various broker error codes as well as transport layer issues), it doesn't make much sense to limit the number of retries for retriable errors, but instead only limit the retries based on the allowed time to produce a message.
  • The default value for the producer configuration property request.timeout.ms has been increased from 5 to 30 seconds to match the Apache Kafka Java producer default. This change yields increased robustness for broker-side congestion.

Enhancements

  • The generated CONFIGURATION.md (through rd_kafka_conf_properties_show())) now include all properties and values, regardless if they were included in the build, and setting a disabled property or value through rd_kafka_conf_set() now returns RD_KAFKA_CONF_INVALID and provides a more useful error string saying why the property can't be set.
  • Consumer configs on producers and vice versa will now be logged with warning messages on client instantiation.

Fixes

Security fixes

  • There was an incorrect call to zlib's inflateGetHeader() with unitialized memory pointers that could lead to the GZIP header of a fetched message batch to be copied to arbitrary memory. This function call has now been completely removed since the result was not used. Reported by Ilja van Sprundel.

General fixes

  • rd_kafka_topic_opaque() (used by the C++ API) would cause object refcounting issues when used on light-weight (error-only) topic objects such as consumer errors (#2693).
  • Handle name resolution failures when formatting IP addresses in error logs, and increase printed hostname limit to ~256 bytes (was ~60).
  • Broker sockets would be closed twice (thus leading to potential race condition with fd-reuse in other threads) if a custom socket_cb would return error.

Consumer fixes

  • The roundrobin partition.assignment.strategy could crash (assert) for certain combinations of members and partitions. This is a regression in v1.5.0. (#3024)
  • The C++ KafkaConsumer destructor did not destroy the underlying C rd_kafka_t instance, causing a leak if close() was not used.
  • Expose rich error strings for C++ Consumer Message->errstr().
  • The consumer could get stuck if an outstanding commit failed during rebalancing (#2933).
  • Topic authorization errors during fetching are now reported only once (#3072).

Producer fixes

  • Topic authorization errors are now properly propagated for produced messages, both through delivery reports and as ERR_TOPIC_AUTHORIZATION_FAILED return value from produce*() (#2215)
  • Treat cluster authentication failures as fatal in the transactional producer (#2994).
  • The transactional producer code did not properly reference-count partition objects which could in very rare circumstances lead to a use-after-free bug if a topic was deleted from the cluster when a transaction was using it.
  • ERR_KAFKA_STORAGE_ERROR is now correctly treated as a retriable produce error (#3026).
  • Messages that timed out locally would not fail the ongoing transaction. If the application did not take action on failed messages in its delivery report callback and went on to commit the transaction, the transaction would be successfully committed, simply omitting the failed messages.
  • EndTxnRequests (sent on commit/abort) are only retried in allowed states (#3041). Previously the transaction could hang on commit_transaction() if an abortable error was hit and the EndTxnRequest was to be retried.

Note: there was no v1.5.1 librdkafka release

librdkafka v1.5.0

The v1.5.0 release brings usability improvements, enhancements and fixes to librdkafka.

Enhancements

  • Improved broker connection error reporting with more useful information and hints on the cause of the problem.
  • Consumer: Propagate errors when subscribing to unavailable topics (#1540)
  • Producer: Add batch.size producer configuration property (#638)
  • Add topic.metadata.propagation.max.ms to allow newly manually created topics to be propagated throughout the cluster before reporting them as non-existent. This fixes race issues where CreateTopics() is quickly followed by produce().
  • Prefer least idle connection for periodic metadata refreshes, et.al., to allow truly idle connections to time out and to avoid load-balancer-killed idle connection errors (#2845)
  • Added rd_kafka_event_debug_contexts() to get the debug contexts for a debug log line (by @wolfchimneyrock).
  • Added Test scenarios which define the cluster configuration.
  • Added MinGW-w64 builds (@ed-alertedh, #2553)
  • ./configure --enable-XYZ now requires the XYZ check to pass, and --disable-XYZ disables the feature altogether (@benesch)
  • Added rd_kafka_produceva() which takes an array of produce arguments for situations where the existing rd_kafka_producev() va-arg approach can't be used.
  • Added rd_kafka_message_broker_id() to see the broker that a message was produced or fetched from, or an error was associated with.
  • Added RTT/delay simulation to mock brokers.

Upgrade considerations

  • Subscribing to non-existent and unauthorized topics will now propagate errors RD_KAFKA_RESP_ERR_UNKNOWN_TOPIC_OR_PART and RD_KAFKA_RESP_ERR_TOPIC_AUTHORIZATION_FAILED to the application through the standard consumer error (the err field in the message object).
  • Consumer will no longer trigger auto creation of topics, allow.auto.create.topics=true may be used to re-enable the old deprecated functionality.
  • The default consumer pre-fetch queue threshold queued.max.messages.kbytes has been decreased from 1GB to 64MB to avoid excessive network usage for low and medium throughput consumer applications. High throughput consumer applications may need to manually set this property to a higher value.
  • The default consumer Fetch wait time has been increased from 100ms to 500ms to avoid excessive network usage for low throughput topics.
  • If OpenSSL is linked statically, or ssl.ca.location=probe is configured, librdkafka will probe known CA certificate paths and automatically use the first one found. This should alleviate the need to configure ssl.ca.location when the statically linked OpenSSL's OPENSSLDIR differs from the system's CA certificate path.
  • The heuristics for handling Apache Kafka < 0.10 brokers has been removed to improve connection error handling for modern Kafka versions. Users on Brokers 0.9.x or older should already be configuring api.version.request=false and broker.version.fallback=... so there should be no functional change.
  • The default producer batch accumulation time, linger.ms, has been changed from 0.5ms to 5ms to improve batch sizes and throughput while reducing the per-message protocol overhead. Applications that require lower produce latency than 5ms will need to manually set linger.ms to a lower value.
  • librdkafka's build tooling now requires Python 3.x (python3 interpreter).

Fixes

General fixes

  • The client could crash in rare circumstances on ApiVersion or SaslHandshake request timeouts (#2326)
  • ./configure --LDFLAGS='a=b, c=d' with arguments containing = are now supported (by @sky92zwq).
  • ./configure arguments now take precedence over cached configure variables from previous invocation.
  • Fix theoretical crash on coord request failure.
  • Unknown partition error could be triggered for existing partitions when additional partitions were added to a topic (@benesch, #2915)
  • Quickly refresh topic metadata for desired but non-existent partitions. This will speed up the initial discovery delay when new partitions are added to an existing topic (#2917).

Consumer fixes

  • The roundrobin partition assignor could crash if subscriptions where asymmetrical (different sets from different members of the group). Thanks to @ankon and @wilmai for identifying the root cause (#2121).
  • The consumer assignors could ignore some topics if there were more subscribed topics than consumers in taking part in the assignment.
  • The consumer would connect to all partition leaders of a topic even for partitions that were not being consumed (#2826).
  • Initial consumer group joins should now be a couple of seconds quicker thanks expedited query intervals (@benesch).
  • Fix crash and/or inconsistent subscriptions when using multiple consumers (in the same process) with wildcard topics on Windows.
  • Don't propagate temporary offset lookup errors to application.
  • Immediately refresh topic metadata when partitions are reassigned to other brokers, avoiding a fetch stall of up to topic.metadata.refresh.interval.ms. (#2955)
  • Memory for batches containing control messages would not be freed when using the batch consume APIs (@pf-qiu, #2990).

Producer fixes

  • Proper locking for transaction state in EndTxn handler.

librdkafka v1.4.4

v1.4.4 is a maintenance release with the following fixes and enhancements:

  • Transactional producer could crash on request timeout due to dereferencing NULL pointer of non-existent response object.
  • Mark rd_kafka_send_offsets_to_transaction() CONCURRENT_TRANSACTION (et.al) errors as retriable.
  • Fix crash on transactional coordinator FindCoordinator request failure.
  • Minimize broker re-connect delay when broker's connection is needed to send requests.
  • Proper locking for transaction state in EndTxn handler.
  • socket.timeout.ms was ignored when transactional.id was set.
  • Added RTT/delay simulation to mock brokers.

Note: there was no v1.4.3 librdkafka release

librdkafka v1.4.2

v1.4.2 is a maintenance release with the following fixes and enhancements:

  • Fix produce/consume hang after partition goes away and comes back, such as when a topic is deleted and re-created.
  • Consumer: Reset the stored offset when partitions are un-assign()ed (fixes #2782). This fixes the case where a manual offset-less commit() or the auto-committer would commit a stored offset from a previous assignment before a new message was consumed by the application.
  • Probe known CA cert paths and set default ssl.ca.location accordingly if OpenSSL is statically linked or ssl.ca.location is set to probe.
  • Per-partition OffsetCommit errors were unhandled (fixes #2791)
  • Seed the PRNG (random number generator) by default, allow application to override with enable.random.seed=false (#2795)
  • Fix stack overwrite (of 1 byte) when SaslHandshake MechCnt is zero
  • Align bundled c11 threads (tinycthreads) constants to glibc and musl (#2681)
  • Fix return value of rd_kafka_test_fatal_error() (by @ckb42)
  • Ensure CMake sets disabled defines to zero on Windows (@benesch)

Note: there was no v1.4.1 librdkafka release

Older releases

See https://github.com/edenhill/librdkafka/releases