Fix cluster-wide restart issue. #2283
Conversation
| if len(joinURLs) > 0 { | ||
| joinServer(s, cmd.config.ClusterURL(), joinURLs) | ||
| return s | ||
| } | ||
|
|
||
| if err := s.Initialize(cmd.config.ClusterURL()); err != nil { | ||
| log.Fatalf("server initialization error: %s", err) | ||
| log.Fatalf("server initialization error(0): %s", err) |
There was a problem hiding this comment.
You can make better log messages than this. :-) E.g.
server initialization error (server ID 0) ?
There was a problem hiding this comment.
I think the 0 is to delineate the two places where Initialize could have returned an error. Here and L606. Don't think it's the server ID but could be wrong.
There was a problem hiding this comment.
OK, you could be right. It would still be better to see clearer log messages to help the next guy. E.g. server initialization error during ..... -- I presume the software is in different states depending at the different points.
There was a problem hiding this comment.
This whole file needs some rework so I just wanted to differentiate between this error message and another with the same text below (as @jwilder mentioned).
|
+1 from me. CHANGELOG needs to be updated too. Issue opened regarding possible re-work of delay: #2285 |
|
CHANGELOG updated, merging. |
Fix cluster-wide restart issue.
|
Adding 1 final note here for posterity. This issue is specifically about multi-node restart, because at restart-time the cluster already exists as a multi-node cluster. Therefore leader election requires more than 1 node to vote. This is not the case when a multi-node cluster is being created for the first time because it always starts as a single-node, and leader-election takes place synchronously with the start of the first broker. Single-node clusters would not suffer from this restart problem either. |
Overview
This pull request fixes some error checking and adds a delay when starting to allow the brokers to elect a leader when starting in the case where the whole cluster is restarted.