MINOR: fix Streams version-probing system test#6764
Conversation
|
Retest this please |
|
Call for review @guozhangwang @bbejeck @vvcephei @ableegoldman @cadonna @abbccdda |
|
System test passed: https://jenkins.confluent.io/job/system-test-kafka-branch-builder/2611/ |
There was a problem hiding this comment.
This is one fix. If rebalance happens quickly, we might see this message for the second rebalance early. Cf. #6763
There was a problem hiding this comment.
After #6779 is merged, I can rebase this PR -- we still want this fix I guess :)
There was a problem hiding this comment.
This is the actual fix (similar below).
|
Call for review @guozhangwang @bbejeck @vvcephei @ableegoldman @cadonna @abbccdda |
vvcephei
left a comment
vvcephei
left a comment
There was a problem hiding this comment.
Ah. Thanks for figuring this out!
|
I think we should consider whether streams/consumer should still maintain the |
|
cc @abbccdda |
|
@mjsax could you rebase this one given the other PR has been merged now? |
There was a problem hiding this comment.
nit: do we want to define 3 instead of leaving it as magic number?
There was a problem hiding this comment.
With the latest fixes, I think we actually don't need this any longer. We only needed if, if we get immediate rebalance. With disabling leave group request, this this should not be required any longer.
915a20a to
0080b24
Compare
|
Rebased and updated this PR. Version probing system test passed locally. Also triggered https://jenkins.confluent.io/job/system-test-kafka-branch-builder/2629/ If no objection, I'll merge this after Jenkins and system test build passed. |
guozhangwang
left a comment
guozhangwang
left a comment
There was a problem hiding this comment.
@mjsax feel free to merge after green builds.
|
System test passed. Jenkins failed with env issue. Retest this please. |
|
@mjsax seems ready to merge. |
Reviewers: John Roesler <john@confluent.io>, Bill Bejeck <bill@confluent.io>, Guozhang Wang <guozhang@confluent.io>, Boyang Chen <boyang@confluent.io>
|
Merged to |
Reviewers: John Roesler <john@confluent.io>, Bill Bejeck <bill@confluent.io>, Guozhang Wang <guozhang@confluent.io>, Boyang Chen <boyang@confluent.io>
Reviewers: John Roesler <john@confluent.io>, Bill Bejeck <bill@confluent.io>, Guozhang Wang <guozhang@confluent.io>, Boyang Chen <boyang@confluent.io>
6 participants
With KAFKA-8284, the
AbstractCoordinatordoes not allow to suppress sendingLeaveGroupRequestonConsumer#close()any longer. Hence, we get a different rebalance pattern.This implies, that the last rebalance, does not do any version probing any longer. When the last instance is bounced, an immediate rebalance "stabilized the group" and prepares it to upgrade to the new metadata version, because all members are on old version but support new version already. On restart of the last instance, we immediately switch to the new version instead of doing a version probing.