[
https://issues.apache.org/jira/browse/HBASE-7671?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13564052#comment-13564052
]
chunhui shen commented on HBASE-7671:
-------------------------------------
Each flushed file has a sequence id, and all hlog entries which smaller than
the max sequence id will be skipped when replaying edit logs.
It means all data in memstore smaller than the sequence id has been flushed to
storefile when flushing, but broken by the case that snapshot called again
without clearing previous because of last failure.
bq.Could somehow store the correct sequenceId with the snapshot?
I have considered this solution, but it seems not convenient for multi stores.
In that way we should maintain sequence id for each memstore rather than one
region.
Snapshot means creating a snapshot of the current memstore. If data belongs to
memstore before successfully flushed, IMO, copies any new KVs is reasonable.
> Flushing memstore again after last failure could cause data loss
> ----------------------------------------------------------------
>
> Key: HBASE-7671
> URL: https://issues.apache.org/jira/browse/HBASE-7671
> Project: HBase
> Issue Type: Bug
> Affects Versions: 0.94.4
> Reporter: chunhui shen
> Assignee: chunhui shen
> Priority: Critical
> Fix For: 0.96.0, 0.94.5
>
> Attachments: HBASE-7671.patch, HBASE-7671v2.patch, HBASE-7671v3.patch
>
>
> See the following logs first:
> {code}
> 2013-01-23 18:58:38,801 INFO org.apache.hadoop.hbase.regionserver.Store:
> Flushed , sequenceid=9746535080, memsize=101.8m, into tmp file
> hdfs://dw77.kgb.sqa.cm4:9900/hbase-test3/writetest1/8dc14e35b4d7c0e481e0bb30849cff7d/.tmp/bebeeecc56364b6c8126cf1dc6782a25
> 2013-01-23 18:58:41,982 WARN org.apache.hadoop.hbase.regionserver.MemStore:
> Snapshot called again without clearing previous. Doing nothing. Another
> ongoing flush or did we fail last attempt?
> 2013-01-23 18:58:43,274 INFO org.apache.hadoop.hbase.regionserver.Store:
> Flushed , sequenceid=9746599334, memsize=101.8m, into tmp file
> hdfs://dw77.kgb.sqa.cm4:9900/hbase-test3/writetest1/8dc14e35b4d7c0e481e0bb30849cff7d/.tmp/4eede32dc469480bb3d469aaff332313
> {code}
> The first time memstore flush is failed when commitFile()(Logged the first
> edit above), then trigger server abort, but another flush is coming
> immediately(could caused by move/split,Logged the third edit above) and
> successful.
> For the same memstore's snapshot, we get different sequenceid, it causes data
> loss when replaying log edits
> See details from the unit test case in the patch
--
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
