[
https://issues.apache.org/jira/browse/HBASE-8580?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13662384#comment-13662384
]
Enis Soztutar commented on HBASE-8580:
--------------------------------------
We do log rolling if we detect that the replica count has gone below
"hbase.regionserver.hlog.tolerable.lowreplication" (which is 3 on hdfs) for
every sync in WAL. For flushes, and compactions, we do not care about the
replication count for the blocks that are being written, but we might care for
the min replication for all the blocks in the file after finalizing the file
and before committing the results. For compaction for example, we probably do
not want to replace files with replication 3 with compacted file with
replication 2. But as Sergey said, datanodes can still fail just after we
checked etc.
> Ensure that there are 3 replicas when Flushing or Compacting
> ------------------------------------------------------------
>
> Key: HBASE-8580
> URL: https://issues.apache.org/jira/browse/HBASE-8580
> Project: HBase
> Issue Type: Improvement
> Reporter: Elliott Clark
>
> It is possible that datanodes in a DFS pipeline can be dropped. When this
> happens it possible for there to be fewer than the number of replicas
> expected. After a close on the file the namenode tries to re-replicate the
> file. On flush and compaction we should block waiting for the number of
> datanodes reporting that they have the needed blocks to be 3.
--
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
