GitHub user apurvam opened a pull request:
https://github.com/apache/kafka/pull/4020
KAFKA-6003: Accept appends on replicas and when rebuilding the log
unconditionally
This is a port of #4004 for the 0.11.0 branch.
With this patch so that we _only_ validate appends which originate
from the client. In general, once the append is validated and written to
the leader the first time, revalidating it is undesirable since we can't
do anything if validation fails, and also because it is hard to maintain
the correct assumptions during validation, leading to spurious
validation failures.
For example, when we have compacted topics, it is possible for batches
to be compacted on the follower but not on the leader. This case would
also lead to an OutOfOrderSequencException during replication. The same
applies to when we rebuild state from compacted topics: we would get
gaps in the sequence numbers, causing the OutOfOrderSequence.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/apurvam/kafka
KAKFA-6003-0.11.0-handle-unknown-producer-on-replica
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/kafka/pull/4020.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #4020
----
commit 0a6a0213c091c8e6b6a9c5ce7655b7e0d06c9db0
Author: Apurva Mehta <apu...@confluent.io>
Date: 2017-10-04T20:42:17Z
KAFKA-6003: Accept appends on replicas and when rebuilding state from
the log unconditionally.
With this patch so that we _only_ validate appends which originate
from the client. In general, once the append is validated and written to
the leader the first time, revalidating it is undesirable since we can't
do anything if validation fails, and also because it is hard to maintain
the correct assumptions during validation, leading to spurious
validation failures.
For example, when we have compacted topics, it is possible for batches
to be compacted on the follower but not on the leader. This case would
also lead to an OutOfOrderSequencException during replication. The same
applies to when we rebuild state from compacted topics: we would get
gaps in the sequence numbers, causing the OutOfOrderSequence.
----
---
