Adar Dembo has posted comments on this change.

Change subject: KUDU-1358 (part 1): master should accept heartbeat even if 

Patch Set 12:

Commit Message:

Line 20: possible; the only way a master can "lose" a cached TSDescriptor is if 
> what about when a tserver restarts? is it still going to do a proper full T
If the tserver restarts, it always issues a full TR as its first heartbeat. And 
if that heartbeat fails, it'll get retried until the server acknowledges it.

That's always been the case; the subsequent patch just adds "if you see your 
master go from not_leader to leader, send a full TR".
File src/kudu/integration-tests/

Line 289:             auto ts = mini_tablet_server->server();
> should this verify that the TS process is running? (there's a getter for th
I'm inclined to keep it the way it is. This is old code; my goal was to add the 
DO_NOT_MATCH_TSERVERS mode so I could use it in my integration test.

I'll document that the tservers may not be running, though.

Line 299:       }
> else LOG(FATAL) perhaps? or use a switch() so that you get a compilation er
I'll use a switch. Note that the downside of using enum classes is that you 
can't use operator<< on the value (it doesn't implicit cast to an int).
File src/kudu/integration-tests/mini_cluster.h:

Line 145:     // Match the tservers retrieved from each master against the 
tservers in
> not sure what 'Match' means here. Is this a verification? (i.e that the cou
OK, will elaborate in the comment.

To view, visit
To unsubscribe, visit

Gerrit-MessageType: comment
Gerrit-Change-Id: I578674927b65b4171e8437de8515130e4a0ed139
Gerrit-PatchSet: 12
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: Yes

Reply via email to