[
https://issues.apache.org/jira/browse/HBASE-2467?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12899426#action_12899426
]
stack commented on HBASE-2467:
------------------------------
What did you do to the patch J-D? In the hadoop we ship, we should include
hdfs-895? (Its a dfsclient-only fix, right?). What should we do about group
commit. Open a new issue? If hdfs-895 is not in place, we just run slower, is
that right?
On the patch:
If concurrency, should 'private boolean logRollRequested' be volatile? (Maybe
not necessary -- I see check is done in a synchronized method?)
I don't get this bit:
{code}
+ synchronized (this) {
+ this.syncTime += System.currentTimeMillis() - now;
+ this.syncOps++;
+ }
+
+ synchronized (this.updateLock) {
{code}
Why not do the syncTime and syncOps under the updateLock rather than do a
synchronize on 'this'? Wouldn't it have same effect but be less obnoxious?
(Haven't looked at rest of this class to check if syncops and synctime updates
are done under a synchronize(this)).
> Concurrent flushers in HLog sync using HDFS-895
> -----------------------------------------------
>
> Key: HBASE-2467
> URL: https://issues.apache.org/jira/browse/HBASE-2467
> Project: HBase
> Issue Type: Improvement
> Components: regionserver
> Reporter: Todd Lipcon
> Assignee: Todd Lipcon
> Fix For: 0.90.0
>
> Attachments: HBASE-2467-v2.patch, hbase-2467.txt
>
>
> HDFS-895 changes hflush() to be able to run concurrently from multiple
> threads, where flushes can be concurrent with further writes to the same file.
> We need to rip out/amend the group commit code a bit to take advantage of
> this.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
