The
secondary node (passive replica) collects changelogs in ".processing"
as the primary node (first replica) performs the synchronization. On a
replica failover, the passive replica (now active) starts where the
primary left. This _overload_ of changelog backlog in the passive node
is the cause of inode fill up (In primary, changelog are moved to
".processed" which would still fill up inodes, but I guess you purge
them periodically now).
I
guess you'd need to purge changlogs from ".processing" in the secondary
node, but the correct thing to do is to perform this automatically in
geo-replication. Could you raise a bug for this?
Venky
Hi, Incurred in this
problem again. After purging the .processed everything went ok for a
while. But now the .processing directory is filling up which is going to
bring to the same issue.
After some further investigation I found that this is happening only on
the secondary node, the primary one cleans up everything from the
.processing as soon as the file gets synced.
Thanks,
It's
safe to purge everything under .processed. That what geo-rep had
already replicated, so it's OK to delete it.
Also,
consider purging these entries periodically as geo-rep doesn't purge
them on it's own (at least for now).
Venky