CASSANDRA-20741: Add rebootstrap to Accord to enable a lagging node or a node that has been down for a longer period to rejoin the cluster #4227
Conversation
… been down for a longer period to rejoin the cluster
| replayJournal(as); | ||
| // If we hit an error during journal replay, we need to mark ourselves unsafe to read, perform full data repair, | ||
| // and trigger RX before we can continue serving traffic | ||
| if (replayJournal(as) && ClusterMetadata.current().directory.allJoinedEndpoints().size() > 1) |
There was a problem hiding this comment.
is allJoinedEndpoints.size() > 1 enough? Don't we need a quorum (or some stronger user command insisting we proceed with a minority)?
There was a problem hiding this comment.
Oh, we definitely need that in order to succeed. Here, I am just checking for whether or not we are in the single node cluster, in which case rebootstrap will simply not work.
| } | ||
|
|
||
| @VisibleForTesting | ||
| public static void replayJournal(AccordService as) | ||
| public static boolean replayJournal(AccordService as) |
There was a problem hiding this comment.
I think it is surprising for true to mean replay failed. I think we should either use false, or introduce an enum
There was a problem hiding this comment.
sure good point, inverted this and called the variable success instead
|
Do we want to add a "clean shutdown" marker here too, for periodic mode, so we know we may need to rebootstrap via simple RX? Or leave it for a later patch? |
belliottsmith
force-pushed
the
trunk
branch
2 times, most recently
from
df3eb40 to
54e39a9
Compare
No description provided.