[
https://issues.apache.org/jira/browse/HBASE-27263?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17574133#comment-17574133
]
Ujjawal Kumar commented on HBASE-27263:
---------------------------------------
I now see is that that the change proposed above is already there if we think
about it. Normally, HRegion control the RowTooBigException via the config
{{hbase.table.max.rowsize}} (default as 1gb) during scans. For live tables we
can't change this config, since it would require a RS restart for it to be
reflected (also this would defeat purpose of this change for OOME protection at
RS side). But since snapshot reads creates a new instance of HRegion within the
MR task any conf passed to the MR job would also be seen by this HRegion
instance. So we can just change this config to a high value for snapshot reads
and we wouldn't see this kind of issue anymore. Hence marking it as not
required for now.
> Make RowTooBigException configurable for snapshot reads
> --------------------------------------------------------
>
> Key: HBASE-27263
> URL: https://issues.apache.org/jira/browse/HBASE-27263
> Project: HBase
> Issue Type: Improvement
> Components: regionserver, snapshots
> Reporter: Ujjawal Kumar
> Assignee: Ujjawal Kumar
> Priority: Minor
> Attachments: Screenshot 2022-08-01 at 3.42.57 PM.png
>
>
> While trying to read snapshot data using TableSnapshotInputFormat in an MR
> job, we can skip OOME check (introduced by HBASE-10925 for region server side
> OOME protection ) since the allocated memory resources would be from MR tasks
> themselves (as compared to RS during live scans)
> !Screenshot 2022-08-01 at 3.42.57 PM.png!
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
