cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Stefania (JIRA)" <>
Subject [jira] [Commented] (CASSANDRA-12397) Altering a column's type breaks commitlog replay
Date Wed, 28 Sep 2016 08:16:20 GMT


Stefania commented on CASSANDRA-12397:

The problem is that {{PartitionUpdateSerializer}} does not serialize the column types because
it calls {{UnfilteredRowIteratorSerializer.serializer.serialize(iter, null, out, version,
update.rowCount())}}, which calls {{SerializationHeader.serializer.serializeForMessaging(header,
selection, out, hasStatic)}}, which only serializes the column names. 

CASSANDRA-12461 fixes the exception that I've reproduced above because it recycles commit
log segments in the shutdown hook, although there is still one CL file left.  However, this
wouldn't cover the case of a crash just after a column is modified, in which case we would
not be able to replay the mutation in the CL after startup. When a table is modified, a flush
is scheduled in {{cfs.reload()}}, so there is a short window where the process may crash and
be unable to restart. I'm not so sure what we can do about this other than changing CL format
to include column types like we do for sstables.

> Altering a column's type breaks commitlog replay
> ------------------------------------------------
>                 Key: CASSANDRA-12397
>                 URL:
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Carl Yeksigian
>            Assignee: Stefania
> When switching from a fixed-length column to a variable-length column, replaying the
commitlog on restart will have the same issue as CASSANDRA-11820. Seems like it is related
to the schema being flushed and used when restarted, but commitlogs having been written in
the old format.
> {noformat}
> org.apache.cassandra.db.commitlog.CommitLogReadHandler$CommitLogReadException: Unexpected
error deserializing mutation; saved to /tmp/mutation4816372620457789996dat.  This may be caused
by replaying a mutation against a table with the same name but incompatible schema.  Exception
follows: EOF after 259 bytes out of 3336
>         at org.apache.cassandra.db.commitlog.CommitLogReader.readMutation(
>         at org.apache.cassandra.db.commitlog.CommitLogReader.readSection(
>         at org.apache.cassandra.db.commitlog.CommitLogReader.readCommitLogSegment(
>         at org.apache.cassandra.db.commitlog.CommitLogReader.readAllFiles(
>         at org.apache.cassandra.db.commitlog.CommitLogReplayer.replayFiles(
>         at org.apache.cassandra.db.commitlog.CommitLog.recoverFiles(
>         at org.apache.cassandra.db.commitlog.CommitLog.recoverSegmentsOnDisk(
>         at org.apache.cassandra.service.CassandraDaemon.setup(
>         at org.apache.cassandra.service.CassandraDaemon.activate(
>         at org.apache.cassandra.service.CassandraDaemon.main(
> {noformat}

This message was sent by Atlassian JIRA

View raw message