[
https://issues.apache.org/jira/browse/KAFKA-7380?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16605089#comment-16605089
]
Guozhang Wang commented on KAFKA-7380:
--------------------------------------
Some initial thoughts about how to do it:
1. Whenever a thread transited to PARTITION_ASSIGNED state, check if the global
store's restoration has completed; if not, halt on further transiting to
RUNNING, and tries to start restoring global stores.
2. Only start a task after its corresponding stores changelog has been
restored, IN ADDITION, the global stores being restored as well.
> Global thread restore blocks KafkaStreams#start()
> -------------------------------------------------
>
> Key: KAFKA-7380
> URL: https://issues.apache.org/jira/browse/KAFKA-7380
> Project: Kafka
> Issue Type: Bug
> Components: streams
> Reporter: Matthias J. Sax
> Priority: Minor
>
> KafkaStreams#start() is documented to be non-blocking and should return
> immediately. However, if global stores are used, `start()` blocks until all
> global stores are restored.
> We should change this and start all threads at the same time while all
> StreamThread would to go `wait()` state on startup immediately, and only
> resume after global thread restore finishes and `wakeup()` all StreamThreads.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
