Gabriella Lotz has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/24476 )

Change subject: KUDU-3767 Add global leader awareness to auto leader rebalancer 
(Part 2)
......................................................................


Patch Set 5:

(2 comments)

http://gerrit.cloudera.org:8080/#/c/24476/3/src/kudu/master/auto_leader_rebalancer.cc
File src/kudu/master/auto_leader_rebalancer.cc:

http://gerrit.cloudera.org:8080/#/c/24476/3/src/kudu/master/auto_leader_rebalancer.cc@491
PS3, Line 491:   // A leader is only ever handed to a tserver that already 
hosts a follower
> For the example above, I'm not sure how those two factors can prevent endle
Yes, I traced it through. ceil_avg = ceil(12/4) = 3, so TS0/TS1/TS2 (at 4) do 
look overloaded. But a leader can only be handed to a tserver that already 
hosts a follower of that tablet, and the only candidates are the other two of 
TS0/TS1/TS2, which are also at 4 (not below the ceiling), so they're rejected. 
TS3 is below the ceiling but hosts no replica, so it's never a follower and 
can't receive leadership. So no tablet has a valid destination and the pass 
makes no moves. Here it's the replica rebalancer that schedules moves (to 
populate TS3), the leader rebalancer doesn't.


http://gerrit.cloudera.org:8080/#/c/24476/4/src/kudu/master/auto_leader_rebalancer.cc
File src/kudu/master/auto_leader_rebalancer.cc:

http://gerrit.cloudera.org:8080/#/c/24476/4/src/kudu/master/auto_leader_rebalancer.cc@807
PS4, Line 807: //
> Do you think it is possible that we end up with non-zero per-table-moves al
Good point, I think that's possible in a cluster with constant per-table churn. 
I've got a separate ticket for the leader rebalancer TODOs (metrics included), 
so I've added a TODO here to emit a metric when the global pass is skipped, and 
I'll cover it there.



--
To view, visit http://gerrit.cloudera.org:8080/24476
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I5d44761f6444ca1d1bf4fcee405d65d029e41980
Gerrit-Change-Number: 24476
Gerrit-PatchSet: 5
Gerrit-Owner: Gabriella Lotz <[email protected]>
Gerrit-Reviewer: Alexey Serbin <[email protected]>
Gerrit-Reviewer: Ashwani Raina <[email protected]>
Gerrit-Reviewer: Gabriella Lotz <[email protected]>
Gerrit-Reviewer: Kudu Jenkins (120)
Gerrit-Reviewer: Marton Greber <[email protected]>
Gerrit-Reviewer: Zoltan Chovan <[email protected]>
Gerrit-Comment-Date: Thu, 09 Jul 2026 11:22:15 +0000
Gerrit-HasComments: Yes

Reply via email to