Details
-
Bug
-
Status: Resolved
-
Normal
-
Resolution: Not A Problem
-
None
-
None
-
None
-
centos 6.6 server, java version "1.8.0_45", cassandra 2.1.7,
3 machines, keyspace with replication factor 3
-
Normal
Description
Application does 'insert ... if not exists' with consistency level quorum into table and expects to receive 'applied' only once for the same key.
Under normal conditions that expectation holds but it fails when cassandra cluster is being restarted one by one.
Restart scenario is: kill -9 one node, wait 6 seconds, start it back, wait till it becomes available, wait 40 seconds (during that time all 3 nodes are up), repeat with the next node.
The problem was observed on 2.1.4, upgrading to 2.1.7 didn't change the behaviour