Details
-
Bug
-
Resolution: Cannot Reproduce
-
Major
-
1.6.4.1
-
Security Level: Public
Description
This is a higher level tracking bug. I am attempting to gather logs but the cluster has been wiped so it may be hard. The anectdotal scenario is as follows:
-10 nodes cluster with ~1 billion keys (+1 replica) running fine for an extended period of time
-5 nodes crashed due to bug MB-3443
-After the memcached processes finished warming up, replication started
-This is where things get a little hazier...upon looking at the cluster sometime after replication started, it was clear that almost ALL items (replica and active) were ejected to disk.
We need to attempt to reproduce this behavior