fix(kafka-python): record actual producer partition from send() future#4858
Open
bourbonkk wants to merge 4 commits into
Open
fix(kafka-python): record actual producer partition from send() future#4858bourbonkk wants to merge 4 commits into
bourbonkk wants to merge 4 commits into
Conversation
|
|
…r spans When neither a key nor an explicit partition is provided, the DefaultPartitioner selects a partition at random. The instrumentation called `_partition()` to record it in the span, but the producer's `send()` calls `_partition()` again independently and may pick a different partition, so `messaging.kafka.partition` frequently did not match where the message was actually delivered. Record the partition only when it can be determined without changing where the message lands (explicit partition, or key-based hash which is deterministic) and omit the attribute for the random keyless case. Fixes open-telemetry#4625 Assisted-by: Claude Opus 4.8 Signed-off-by: allen <allen.k1m@kakaocorp.com>
bourbonkk
force-pushed
the
fix/kafka-python-partition-4625
branch
from
July 23, 2026 06:28
10cc6d8 to
8823299
Compare
Instead of estimating the destination partition before send() (which, with the default partitioner and no key, runs a random choice that send() then redoes and diverges from), read the partition back from the FutureRecordMetadata returned by send() via _produce_future.topic_partition. This is accurate for explicit, key-based and random-partition cases alike, and removes the speculative _wait_on_metadata()/_serialize()/_partition() calls the instrumentation previously made. The producer span now wraps the send() call so the partition can be read from its result before the span ends. Fixes open-telemetry#4625 Assisted-by: Claude Opus 4.8 Signed-off-by: allen <allen.k1m@kakaocorp.com>
bourbonkk
changed the title
Contributor
Author
|
Verified against a live Kafka broker (single-node, topic with 12 partitions, 40 messages sent with no key and no explicit partition, so the DefaultPartitioner assigns randomly):
Also worth noting: the old code's _wait_on_metadata(topic, max_block_ms / 1000.0) call raises KafkaTimeoutError on kafka-python 2.3.2, since that argument is now milliseconds — so the pre-send estimation path was already broken on newer versions. This PR removes that call entirely. |
herin049
approved these changes
Jul 24, 2026
Address review nits: use `list[str]` and `int | None` for the annotations touched in this change. Assisted-by: Claude Opus 4.8 Signed-off-by: allen <allen.k1m@kakaocorp.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Fixes #4625
The kafka-python producer instrumentation recorded
messaging.kafka.partitionfrom a value it computed beforesend(), by callinginstance._partition()itself. With the default partitioner and no key,_partition()selects a partition at random;send()then calls_partition()again independently and picks a different partition, so the span attribute frequently did not match where the message was actually delivered.This reads the partition back from the
FutureRecordMetadatareturned bysend()(_produce_future.topic_partition), which is the partitionsend()actually assigned. It is accurate for all cases (explicit partition, key-based, and random keyless), and removes the speculative_wait_on_metadata()/_serialize()/_partition()calls the instrumentation previously made. If the attribute can't be read it is omitted rather than recording a wrong value.The producer span now wraps the
send()call so the partition can be read from its result before the span ends.Type of change
How Has This Been Tested?
tests/test_utils.py: partition read from the future (incl.0), a check against a real kafka-pythonFutureRecordMetadata, gracefulNonewhen internals are unavailable, and_enrich_spanrecording/omitting the attribute._produce_future.topic_partitionexists across the supported range (kafka-python 2.0.3 … 3.0.8).Does This PR Require a Core Repo Change?
Checklist: