Alan Boudreault created CASSANDRA-10285:
-------------------------------------------
Summary: Compaction running indefinitely on system.hints
Key: CASSANDRA-10285
URL: https://issues.apache.org/jira/browse/CASSANDRA-10285
Project: Cassandra
Issue Type: Bug
Reporter: Alan Boudreault
During my hints storage benchmarks, I've experienced an issue using C* 2.2. The
hints was never replayed. After a while (more than 24H ...), I noticed that
there were still compactions running on system.hints and that some new ones was
triggered every 10-20 minutes.
To reproduce, we create a cluster of 2 nodes. RF=2 and we generate hints by
shutting down the node2.
{code}
ccm create --install-dir=`pwd` -n 2 local && ccm start
ccm node1 stress -- write n=50M -rate threads=300 -port jmx=7198 -errors ignore
-schema replication\(factor=2\)
# wait 5-6 seconds to get the schema creation propagated on all nodes, then in
another window, stop node 2
ccm node2 stop
# wait the initial 50M writes are finished, bring back node2 up and write
another 50M keys.
ccm node2 start
ccm node1 stress -- write n=50M -rate threads=300 -port jmx=7198 -errors ignore
-schema replication\(factor=2\)
# You should get the initial compaction finished after 15-20 minutes. You can
set the mb throughput to 0 to get that done faster.
# Monitor the node1. the hints will never be replayed and you should see
compactions happening indefinitely.
{code}
//cc [~krummas] [~yukim]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
