Mailing-List: contact user-help@hbase.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@hbase.apache.org
Received-SPF: unknown ~alla (athena.apache.org: encountered unrecognized
 mechanism during SPF processing of domain of mcorgan@hotpads.com)
MIME-Version: 1.0
In-Reply-To: 
 <CALte62x0SJS_DDD4k9wm3L0BDs=OGrTGg03jnHKLhPQbPK-0=g@mail.gmail.com>
References: 
 <CAOKsKJURc8HMkG-RsHfkkaUAAFngHtNRzywHt3Nm-Yb+hiiQKg@mail.gmail.com>
	<CALte62x0SJS_DDD4k9wm3L0BDs=OGrTGg03jnHKLhPQbPK-0=g@mail.gmail.com>
Date: Sat, 3 Nov 2012 14:29:45 -0700
Message-ID: 
 <CAOKsKJWWqarxr9xDeTBj0=i22Kj2hbmp6E4vJmpT2pDPHjdAAQ@mail.gmail.com>
Subject: Re: infinite loop of RS_ZK_REGION_SPLIT on .94.2
From: Matt Corgan <mcorgan@hotpads.com>
To: user <user@hbase.apache.org>
Content-Type: multipart/alternative; boundary=20cf3071d01c829c0304cd9df333

--20cf3071d01c829c0304cd9df333
Content-Type: text/plain; charset=UTF-8

Here's a sample of the master's logs from yesterday.  It's not correlated
exactly with the other pastebin log, but there's 3GB of this from
yesterday: http://pastebin.com/wP2rNN1t

I'm am pushing the cluster a bit with importing data so testing the split
code harder than normal.  The regions are 500-1GB gzipped.  I can look into
it more but trying to figure out what to look for.

Thanks Ted,
Matt


On Sat, Nov 3, 2012 at 2:03 PM, Ted Yu <yuzhihong@gmail.com> wrote:

> Matt:
> This is the method which made the logging:
>   private static int tickleNodeSplit(ZooKeeperWatcher zkw,
>       HRegionInfo parent, HRegionInfo a, HRegionInfo b, ServerName
> serverName,
>       final int znodeVersion)
>   throws KeeperException, IOException {
>     byte [] payload = Writables.getBytes(a, b);
>     return ZKAssign.transitionNode(zkw, parent, serverName,
>       EventType.RS_ZK_REGION_SPLIT, EventType.RS_ZK_REGION_SPLIT,
>       znodeVersion, payload);
>   }
>
> transitionZKNode() calls tickleNodeSplit() when waiting for master to split
> the region. Obviously something caused the master not able to split.
>
> How large is the region ?
>
> Can you pastebin master log for that period of time ?
>
> Thanks
>
> On Sat, Nov 3, 2012 at 1:54 PM, Matt Corgan <mcorgan@hotpads.com> wrote:
>
> > We upgraded from .94.0 to .94.2 last week and have started to encounter
> > infinite loops of region-transition on splits.  I'm not sure yet if it's
> > all splits nor if it's related to load.  Solution so far has been to
> > restart the regionserver process.
> >
> > log snippet:
> > http://pastebin.com/LpienZ7B
> >
> > It's repeating these two lines:
> > 2012-11-02 01:35:33,312 DEBUG org.apache.hadoop.hbase.zookeeper.ZKAssign:
> > regionserver:60020-0x13ab46479832b76 Attempting to transition node
> > cf3e9bc069e1888983c06dc8e053ffcf from RS_ZK_REGION_SPLIT to
> > RS_ZK_REGION_SPLIT
> > 2012-11-02 01:35:33,364 DEBUG org.apache.hadoop.hbase.zookeeper.ZKAssign:
> > regionserver:60020-0x13ab46479832b76 Successfully transitioned node
> > cf3e9bc069e1888983c06dc8e053ffcf from RS_ZK_REGION_SPLIT to
> > RS_ZK_REGION_SPLIT
> >
> > with the occasional:
> > 2012-11-02 01:35:34,476 DEBUG
> > org.apache.hadoop.hbase.regionserver.SplitTransaction: Still waiting on
> the
> > master to process the split for cf3e9bc069e1888983c06dc8e053ffcf
> >
> > Should the region transition from RS_ZK_REGION_SPLIT to itself?  It looks
> > wrong, but I'm not familiar with the code at all.
> >
> > Thanks,
> > Matt
> >
>

--20cf3071d01c829c0304cd9df333--