Andrew Wong has posted comments on this change. Change subject: docs: clarify steps for changing master from multi-master deployment ......................................................................
Patch Set 3: (1 comment) http://gerrit.cloudera.org:8080/#/c/8032/3/docs/administration.adoc File docs/administration.adoc: PS3, Line 399: * Establish a maintenance window (one hour should be sufficient). During this time the Kudu cluster : will be unavailable. > So this works around KUDU-1620, right? What if we were to restart the remai Original nodes A (dead), B*, C, and attempted to replace A with D. Tried going through this process and a few things to note: * When I brought up D, both B and C's /masters pages successfully updated A's address to D's. * Looking at B's logs, this was not the case; it was still trying to contact A, as expected. * Looking at D's logs, I could see it losing a bunch of pre-elections since the remaining two masters already had a quorum (also, the D's web UI showed four masters, its UUID duplicated, both showing D's address). * After updating the DNS aliases, I restarted C. Once it came up, B continue being leader, and D still was not allowed in. * After restarting B, a C was elected, and the logs appeared normal across B, C*, D. * Interestingly, at the end of this all, B's, C's, and D's web UIs all showed an exact duplicate for D (rpc address and all). So it seems like nothing "goes wrong" with this approach, but I think while C was restarting, wewe were unavailable: single leader but no voters, and an effectively bricked replacement node, resulting in an extremely familiar window of unavailability of size <length of master restart>. If, after I updated the DNS aliases, I'd restarted B* instead, would things have been different? With no leader, would we have been forced into an election? No; things would be the pretty much the same--D and C would not have been able to accept ops individually, and would not have elected a leader for the same unfortunate DNS alias reasons. TL;DR: Doesn't seem like it. -- To view, visit http://gerrit.cloudera.org:8080/8032 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-MessageType: comment Gerrit-Change-Id: I4196dbb2f8a185e868a6906c7cf917d79c404c0d Gerrit-PatchSet: 3 Gerrit-Project: kudu Gerrit-Branch: master Gerrit-Owner: Andrew Wong <aw...@cloudera.com> Gerrit-Reviewer: Adar Dembo <a...@cloudera.com> Gerrit-Reviewer: Alexey Serbin <aser...@cloudera.com> Gerrit-Reviewer: Andrew Wong <aw...@cloudera.com> Gerrit-Reviewer: Kudu Jenkins Gerrit-Reviewer: Mike Percy <mpe...@apache.org> Gerrit-HasComments: Yes