[ 
https://issues.apache.org/jira/browse/HBASE-7213?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13510829#comment-13510829
 ] 

Devaraj Das commented on HBASE-7213:
------------------------------------

bq. Do other logs have a '.log' suffix? You add .hlog with the below?

Currently (in the trunk codebase), the log files don't have extension like .log 
or something (the files look like 
192.168.1.8%2C60020%2C1354695930505.1354695931168). They are in directory 
called ".logs/<region-server>". So I am wondering whether I should just have 
".meta" as the suffix for the meta log files, and be done with it (as I had it 
in the previous patch). In the current patch, the meta logs have .meta.hlog as 
the extension. Or, would it make sense to change the log files to have ".hlog" 
extension as well? Please let me know..

bq. The SSH changes look good. There is a hole though at the moment around ROOT 
handling? We need to wait on ROOT removal before this can go in, right? (The 
RootServerShutdownHandler would go....or should RSSH be calling MSSH when it is 
done w/ its process?)

It can be done either way in theory (HBASE-3171 first or this first). On the 
hole, yes, if this patch goes in first, and if there is a case where the same 
RS hosts both the root and meta regions, and that RS goes down, and the root's 
edits weren't already persisted, there will be a problem. But given that the 
root doesn't have edits other than the one row to do with .META., it is 
unlikely that we will run into the hole in practice. But am okay to wait till 
HBASE-3171 goes in (and that's a better way to line the commits, but would like 
to close on on the other comments).

bq. This below is given to an executor? Should have a better name than handler?
bq. + private UncaughtExceptionHandler handler;

Will take a look and update.

bq. Will we always create the metahlog just because getMetaHLog was called? 
Even if we are NOT carrying .META.:

No. It'll be created on the first call to getMetaWAL (and getMetaWAL gets 
indirectly called only when we are asked to open the meta region in 
OpenRegionHandler)

bq. If .META. moves away from a RS and then comes back, we'll just use the 
already made meta log roller, etc.?

Yes.

bq. Replication will skip this .META. log?

Yes. The MetaLogRoller doesn't "register" with the Replication folks when 
instantiated.

bq. It is intentional that MetaServices is still in the patch?

Mistake.. Will update.

bq. In SplitLogManager, we have added splitMetaLogDistributed. A more generic 
method might have been splitLogDistributed that took a file path filter 
instead...

Hmm.. Let me see..

bq. There is a define for .META. HRegionInfo in 
HRegionInfo#FIRST_META_REGIONINFO so you don't have to make it each time, FYI.

I had seen this one but had forgotten to update code to use it.. will update.

bq. Don't have to deprecate the below I'd say. This is for 0.96 the singularity 
and this is on a class with annotation @InterfaceAudience.Private
bq. /** @return the HLog */
bq. + @Deprecated
bq. public HLog getWAL();

Okay..

{quote}
Here....
return new Path(dir, prefix + "." + filenum);
+ String child = prefix + "." + filenum;
+ return new Path(dir, child);
the 'normal' WALs do not seem to pick up the suffix in here. Is it possible 
that the '.log' is appended elsewhere? Are your .META. logs getting double 
'.log'? Just wondering.
{quote}

No.. as I said before, the log files don't have the string extensions.. So what 
do you think - should I just remove the .hlog from the .meta.hlog extension 
that I did for meta hlogs? Or, leave .meta.hlog as it is and add .hlog to the 
non-meta hlog files, or, leave the non-meta hlog names as they are now?
                
> Have HLog files for .META. edits only
> -------------------------------------
>
>                 Key: HBASE-7213
>                 URL: https://issues.apache.org/jira/browse/HBASE-7213
>             Project: HBase
>          Issue Type: Improvement
>          Components: master, regionserver
>            Reporter: Devaraj Das
>            Assignee: Devaraj Das
>             Fix For: 0.96.0
>
>         Attachments: 7213-2.4.patch, 7213-in-progress.2.2.patch, 
> 7213-in-progress.2.patch, 7213-in-progress.patch
>
>
> Over on HBASE-6774, there is a discussion on separating out the edits for 
> .META. regions from the other regions' edits w.r.t where the edits are 
> written. This jira is to track an implementation of that.

--
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

Reply via email to