[
https://issues.apache.org/jira/browse/KAFKA-5317?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16043541#comment-16043541
]
Gwen Shapira commented on KAFKA-5317:
-------------------------------------
I understand the importance, but is it really a release blocker?
> Update KIP-98 to reflect changes during implementation.
> -------------------------------------------------------
>
> Key: KAFKA-5317
> URL: https://issues.apache.org/jira/browse/KAFKA-5317
> Project: Kafka
> Issue Type: Sub-task
> Components: clients, core, producer
> Reporter: Apurva Mehta
> Priority: Blocker
> Labels: documentation, exactly-once
> Fix For: 0.11.0.0
>
>
> While implementing the EOS design, there are some minor (or major?) tweaks we
> made as we hands-on the code bases. We will compile all these changes in a
> single run at the end of the code-ready and this JIRA is for keeping track of
> all the changes we made.
> 02/27/2017: collapse the two types of transactional log messages into a
> single type with key as transactional id, and value as [pid, epoch,
> transaction_timeout, transaction_state, [topic partition ] ]. Also using a
> single memory map in cache instead of two on the TC.
> 03/01/2017: for pid expiration, we decided to use min(transactional id
> expiration timeout, topic retention). For topics enabled for compaction only,
> we just use the transactional timeout. If the retention setting is larger
> than the transactional id expiration timeout, then the pid will be
> "logically" expired (i.e. we will remove it from the cached pid mapping and
> ignore it when rebuilding the cache from the log)
> 03/20/2017: add a new exception type in `o.a.k.common.errors` for invalid
> transaction timeout values.
> 03/25/2017: extend WriteTxnMarkerRequest to contain multiple markers for
> multiple PIDs with a single request.
> 04/20/2017: add transactionStartTime to TransactionMetadata
> 04/26/2017: added a new retriable error: Errors.CONCURRENT_TRANSACTIONS
> 04/01/2017: We also enforce acks=all on the client when idempotence is
> enabled. Without this, we cannot again guarantee idemptoence.
> 04/10/2017: WE also don't pass the underlying exception to
> `RetriableOffsetCommitFailedException` anymore:
> https://issues.apache.org/jira/browse/KAFKA-5052
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
