hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ankit Singhal (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-19826) Provide a option to see rows behind a delete in a time range queries
Date Tue, 30 Jan 2018 10:43:00 GMT

    [ https://issues.apache.org/jira/browse/HBASE-19826?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16344850#comment-16344850
] 

Ankit Singhal commented on HBASE-19826:
---------------------------------------

{quote}Could phoenix hide the HBase table to user? So that you are free to set KeepDeletedCells.TRUE
as user do not know it...
{quote}
Not really, we let the user have full access to the HBase table. We just don't want to hardcode KeepDeletedCells.TRUE
in the descriptor because of performance reason. As all user queries would read the deleted
data unnecessary(even after compaction) and increase the latency of the scan.

Current use-cases of Phoenix requires it to be set dynamically in user scans or in pre-hook. 
 # During compaction, we want to keep deleted cells only if there are lagging indexes, which will
eventually get build by reading the cells of the data table.
 # While doing Index scrutiny on a live table, time range scan wants to see PUTs not eclipsed
by newer DELETE markers.(raw scan cannot be utilized here as it will give all cells even
if we have delete markers within the time range)

 

 

> Provide a option to see rows behind a delete in a time range queries
> --------------------------------------------------------------------
>
>                 Key: HBASE-19826
>                 URL: https://issues.apache.org/jira/browse/HBASE-19826
>             Project: HBase
>          Issue Type: Improvement
>            Reporter: Ankit Singhal
>            Assignee: Ankit Singhal
>            Priority: Major
>             Fix For: 2.0.0
>
>
> We can provide an option (something like seePastDeleteMarkers) in a scan to let the user
see the versions behind the delete marker even if keepDeletedCells is set to false in the
descriptor.
> With the previous version, we workaround the same in preStoreScannerOpen hook. For reference
PHOENIX-4277
> {code}
> 	    @Override
> 	    public KeyValueScanner preStoreScannerOpen(final ObserverContext<RegionCoprocessorEnvironment>
c,
> 	        final Store store, final Scan scan, final NavigableSet<byte[]> targetCols,
> 	        final KeyValueScanner s) throws IOException {
> 	
> 	      if (scan.isRaw() || ScanInfoUtil.isKeepDeletedCells(store.getScanInfo()) || scan.getTimeRange().getMax()
== HConstants.LATEST_TIMESTAMP || TransactionUtil.isTransactionalTimestamp(scan.getTimeRange().getMax()))
{
> 	        return s;
> 	      }
> 	
> 	      ScanInfo scanInfo = ScanInfoUtil.cloneScanInfoWithKeepDeletedCells(store.getScanInfo());
> 	      return new StoreScanner(store, scanInfo, scan, targetCols,
> 	          c.getEnvironment().getRegion().getReadpoint(scan.getIsolationLevel()));
> 	    }
> {code}
> Another way is to provide a way to set KEEP_DELETED_CELLS to true in ScanOptions of preStoreScannerOpen.



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

Mime
View raw message