[ https://issues.apache.org/jira/browse/KAFKA-3388?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15192535#comment-15192535 ]
Jiangjie Qin commented on KAFKA-3388: ------------------------------------- [~ijuma] I am currently blocked on this issue and working on this. Besides the issue in reported in this ticket, the batch expiration checking code seems not following the design in KIP-19. We are supposed to let each retry have a separate request timeout, but the code doesn't seem doing so. I will submit the patch shortly. > Producer should only timeout a batch in the accumulator when metadata is > missing. > --------------------------------------------------------------------------------- > > Key: KAFKA-3388 > URL: https://issues.apache.org/jira/browse/KAFKA-3388 > Project: Kafka > Issue Type: Bug > Affects Versions: 0.9.0.1 > Reporter: Jiangjie Qin > Priority: Blocker > Fix For: 0.10.0.0 > > > In KIP-19 we are reusing the request.timeout.ms to timeout the batches in the > accumulator. We were intended to avoid the case that the batches sitting in > the accumulator forever when topic metadata is missing. > Currently we are not checking if metadata is available or not when we timeout > the batches in the accumulator (although the comments says we will check the > metadata). This causes problem that once the previous batch hit a request > timeout and got retried, all the subsequent batches will fail with timeout > exception. We should only timeout the batches in the accumulator when the > metadata of the partition is missing. -- This message was sent by Atlassian JIRA (v6.3.4#6332)