[
https://issues.apache.org/jira/browse/CASSANDRA-5116?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13546565#comment-13546565
]
Yuki Morishita commented on CASSANDRA-5116:
-------------------------------------------
Couple of things I noticed during investigation:
* As Michael commented above, there are some sstables that are not referenced
by CF. That happens when those sstables are ancestors of other sstables(see
https://github.com/apache/cassandra/blob/cassandra-1.2.0/src/java/org/apache/cassandra/db/ColumnFamilyStore.java#L267).
Those sstables are supposed to be deleted through
SSTableReader#releaseReference() from the comment below that line. But sstables
would not be deleted since releaseReferece() checks isCompacted flag which, as
the name suggests, only becomes true when compacted. There is no log about this
behavior, so when that happens after upgradesstables, I think it is confusing.
So I propose fix for actually deleting file + logging when that happens.
* Maybe we can raise priority of updatesstables task among other compaction
tasks to be processed as first as possible.
* nodetool upgradesstables just submits task for upgrade but never knows when
it finishes. It may be useful if we track the progress of upgrade like
CASSANDRA-4767. (same can be said to cleanup, scrub, etc).
> upgradesstables does not upgrade all sstables on a node
> -------------------------------------------------------
>
> Key: CASSANDRA-5116
> URL: https://issues.apache.org/jira/browse/CASSANDRA-5116
> Project: Cassandra
> Issue Type: Bug
> Affects Versions: 1.2.0
> Environment: Ubuntu 12.04
> Reporter: Michael Kjellman
> Assignee: Yuki Morishita
>
> upgradesstables appears to be skipping sstables randomly.
> finding a sstable with an mtime < the upgrade time and grepping through the
> logs for a corresponding compaction log line, i find nothing. I have
> reproduced this on all of my nodes across the cluster.
> is performAllSSTableOperation somehow skipping sstables?
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
