[
https://issues.apache.org/jira/browse/HBASE-16812?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15838175#comment-15838175
]
stack commented on HBASE-16812:
-------------------------------
Thanks [~jingcheng.du] for working on this. How would you summarize the
behavioral change in how MOB works after this patch goes in? Would be good to
have as a release note. What you think mighty [~huaxiang]? Am interested in
your opinion sir. ([~appy]?)
Looking at the patch, I love the lock removals (smile) and clean up of
MobCompactionStoreScanner.
Seeing this new tag type makes me think of how MOB is spreading through the
code base which I think a little unfortunate.
Do you foresee much savings doing the below:
{code}
188 if (major && CellUtil.isDelete(c)) {
189 if (MobUtils.isMobOutputDeleteMarker(c)) {
190 // Directly write it to store file
191 writer.append(c);
192 } else {
193 // Add a tag to this cell and write it to store file.
194 writer.append(MobUtils.createMobOutputDeleteMarker(c));
195 // Write the cell to del file
196 delFileWriter.append(c);
197 deleteMarkersCount++;
198 }
{code}
i.e.MOB hfiles have few entries in them? Am I right (sorry, don't know this
area well)?
Is this right [~jingcheng.du] "104 // If a delete marker has this tag,
it is not flushed to a mob file in major compaction." Looking at code, it 'is'
flushed?
Otherwise, patch looks great.
> Clean up the locks in MOB
> -------------------------
>
> Key: HBASE-16812
> URL: https://issues.apache.org/jira/browse/HBASE-16812
> Project: HBase
> Issue Type: Task
> Reporter: Appy
> Assignee: Jingcheng Du
> Priority: Minor
> Attachments: HBASE-16812.master.001.patch, HBASE-16812.patch
>
>
> Clean up the locks in MOB.
> # Retain all the delete markers in mob-enabled columns until they are expired.
> # Remove the locks from major compaction of mob-enabled columns.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
