[
https://issues.apache.org/jira/browse/HBASE-17407?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15798278#comment-15798278
]
Duo Zhang commented on HBASE-17407:
-----------------------------------
And I have claimed many times, you need to think about the future of the
project. We should not plant bombs in the code...
The lowestUnflushedSequenceIds, as its name, is used to store the lowest
unflushed sequence id of a store. In the old time, we will always flush all the
contents of a memstore for a store, so we will just remove the entries in it
and when new edits come it will put its value into it. And in general, we will
abort the RS if flush fails as we can not put the snapshot back because the
memstore may already have some new data. The exception is when the RS is
assigned a new region which need to be recovered from a crash. Before the
recovery done, we will not accept any writes so it is OK to just put back the
snapshot when flush fails. And IMO, the flush for recovery should be much
simpler so maybe we should not let it share code with the normal flush. So in
the normal path, lowestUnflushedSequenceIds will never go backward and
actually, it is not something transactional. After we remove the entry from
lowestUnflushedSequenceIds it will be filled by new edits soon. So if you do
not flush all the contents in memstore for a store, I think the right way is to
get the correct lowest unflushed sequence id at the first place and store it
into lowestUnflushedSequenceIds. I do not know why you guys would like to
remove the value in it and let others store an incorrect value in it and at
last override it with the correct value. What is the benefits? The
{{onlyIfGreater}} parameter makes me really nervous...
Thanks.
> Correct update of maxFlushedSeqId in HRegion
> --------------------------------------------
>
> Key: HBASE-17407
> URL: https://issues.apache.org/jira/browse/HBASE-17407
> Project: HBase
> Issue Type: Bug
> Reporter: Eshcar Hillel
>
> The attribute maxFlushedSeqId in HRegion is used to track the max sequence id
> in the store files and is reported to HMaster. When flushing only part of the
> memstore content this value might be incorrect and may cause data loss.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
