Adar Dembo has posted comments on this change.

Change subject: master: reduce timeout for master to tserver rpcs

Patch Set 2:

Filed KUDU-1526 for the test failure (which appears to be an unrelated flake).

 > I think we bumped this to be pretty high because we found that
 > creating a new tablet can actually take reasonably long (it
 > involves a bunch of fsyncs, preallocating log, etc) especially when
 > there are lots of them going on at the same time (as when you
 > create a table with 10s of tablets per server).

Note that this only changes the "generic" timeout. AsyncCreateReplica() uses 
its own timeout gflag which is still set to 30s. That one is more dangerous to 
dial down, as a timeout would cause the master to replace the tablet in 
question. But, a discrepancy between the two gflags is weird, so I am tempted 
to revert this.

I do recall seeing a once-in-a-blue-moon AlterTable() timeout that I traced 
back to this 30s timeout and a coinciding leader election, but I can't remember 
the details now. I'll retest without the timeout to see if I can retrigger it.

To view, visit
To unsubscribe, visit

Gerrit-MessageType: comment
Gerrit-Change-Id: I06b5f32c0295f0a9d24ff42695905858ae6101f6
Gerrit-PatchSet: 2
Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-Owner: Adar Dembo <>
Gerrit-Reviewer: Adar Dembo <>
Gerrit-Reviewer: Dan Burkert <>
Gerrit-Reviewer: David Ribeiro Alves <>
Gerrit-Reviewer: Kudu Jenkins
Gerrit-Reviewer: Todd Lipcon <>
Gerrit-HasComments: No

Reply via email to