I have sharded cluster of 8xPSA and 3xRS conf servers. My issue is when one of slaves in PSA goes down for maintenance or build index, the local sessions fail to cleanup
Failed to refresh session cache...timeout.
What happens here is that
config database is sharded which means there is chunk of data on each node in cluster. The config server tries to update session on PSA which slave is down and fails to refresh sessions which results in stacking local sessions until it exceeds max sessions limit and whole cluster fails due to not being able to open new session for next operation.
Is there a way to disable
w: majority for config session updates or move
config database from PSA shards?
Or do I need PSS instead of PSA? Also it would be nice to know why session update need to have
w: majority on update.