phoenix-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Geoffrey Jacoby (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (PHOENIX-5048) Index Rebuilder does not handle INDEX_STATE timestamp check for all index
Date Mon, 03 Dec 2018 18:01:00 GMT

    [ https://issues.apache.org/jira/browse/PHOENIX-5048?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16707597#comment-16707597
] 

Geoffrey Jacoby commented on PHOENIX-5048:
------------------------------------------

[~mihir6692], I think I agree with [~vincentpoon]. Why wouldn't you want to keep going in
the for loop even if the else block had an exception?

I think either an exception there needs to be caught, or there needs to be a comment explaining
why it's not, because at the very least it _looks_ like there should be a catch block there,
even if you're right that there's a subtlety that makes it not needed. 

> Index Rebuilder does not handle INDEX_STATE timestamp check for all index
> -------------------------------------------------------------------------
>
>                 Key: PHOENIX-5048
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-5048
>             Project: Phoenix
>          Issue Type: Improvement
>    Affects Versions: 4.14.0, 4.14.1
>            Reporter: Monani Mihir
>            Assignee: Monani Mihir
>            Priority: Major
>         Attachments: PHOENIX-5048.patch, PHOENIX-5048.v2.patch
>
>
> After rebuilder is finished for Partial Index Rebuild, It will check if Index state has
been updated after Upper bound of the scan we use in partial index Rebuild. If that happens
then it will fail Index Rebuild as Index write failure occured while we were rebuilding Index.
> {code:java}
> MetaDataEndpointImpl.java#updateIndexState()
> public void updateIndexState(RpcController controller, UpdateIndexStateRequest request,
> RpcCallback<MetaDataResponse> done) {
> ...
> // If the index status has been updated after the upper bound of the scan we use
> // to partially rebuild the index, then we need to fail the rebuild because an
> // index write failed before the rebuild was complete.
> if (actualTimestamp > expectedTimestamp) {
> builder.setReturnCode(MetaDataProtos.MutationCode.UNALLOWED_TABLE_MUTATION);
> builder.setMutationTime(EnvironmentEdgeManager.currentTimeMillis());
> done.run(builder.build());
> return;
> }
> ...
> }{code}
> After Introduction of TrackingParallelWriterIndexCommitter [PHOENIX-3815|https://issues.apache.org/jira/browse/PHOENIX-3815],
we only disable Index which get failure . Before that , in ParallelWriterIndexCommitter we
were disabling all index even if Index failure happens for one Index only. 
> Suppose Data Table has 3 index and above condition becomes true for first index , then
we won't even check for remain two Index.
> {code:java}
> MetaDataRegionObserver.java#BuildIndexScheduleTask.java#run()
> for (PTable indexPTable : indexesToPartiallyRebuild) {
> String indexTableFullName = SchemaUtil.getTableName(
> indexPTable.getSchemaName().getString(),
> indexPTable.getTableName().getString());
> if (scanEndTime == latestUpperBoundTimestamp) {
> IndexUtil.updateIndexState(conn, indexTableFullName, PIndexState.ACTIVE, 0L, latestUpperBoundTimestamp);
> batchExecutedPerTableMap.remove(dataPTable.getName());
> LOG.info("Making Index:" + indexPTable.getTableName() + " active after rebuilding");
> } else {
> // Increment timestamp so that client sees updated disable timestamp
> IndexUtil.updateIndexState(conn, indexTableFullName, indexPTable.getIndexState(), scanEndTime
* signOfDisableTimeStamp, latestUpperBoundTimestamp);
> Long noOfBatches = batchExecutedPerTableMap.get(dataPTable.getName());
> if (noOfBatches == null) {
> noOfBatches = 0l;
> }
> batchExecutedPerTableMap.put(dataPTable.getName(), ++noOfBatches);
> LOG.info("During Round-robin build: Successfully updated index disabled timestamp for
"
> + indexTableFullName + " to " + scanEndTime);
> }
> }
> {code}
>  
>  
>  



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Mime
View raw message