hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Bharat Shetty <bharat.she...@gmail.com>
Subject Re: HMaster daemon is not starting up
Date Mon, 25 Nov 2013 05:39:24 GMT
Stack, Ted,

Thanks a lot for the pointers. I'm amazed at the quick responses that come
in this mailing lists from this community!

I had changed the directory from /tmp to some other directory for hbase.rootdir
long back only,  but had missed doing the same for Zookeeper in the
conf/hbase-site.xml, since I assumed that setting the HBASE_MANAGES_ZK=true
flag in the hbase-env.sh causes zookeeper to be managed internally by HBase
and as such control over the zookeeper data directory is not possible.
After reading the pointers you gave, I changed it to some other directory
other than /tmp for zookeeper.

So when the /tmp file system on the master node in our cluster went corrupt
somehow,  HQuorumPeer daemon went kaput forcing shut down of the HMaster
daemon.

Thanks once again.

Best,
- Bharat



-- B


On Sat, Nov 23, 2013 at 10:52 AM, Stack <stack@duboce.net> wrote:

> On Fri, Nov 22, 2013 at 6:11 PM, Bharat Shetty <bharat.shetty@gmail.com
> >wrote:
>
> > Yeah, error is related to HDFS.
> >
> > I'm using HBase 0.96.0 with Hadoop 2.1.0-beta.
> >
> > On more further digging, it appears that the /tmp directory on the master
> > node had some problems. Zookeeper for deployment is being managed
> > internally by the HBase, appears to store data in /tmp directory
> > (/tmp/hbase-iouser/zookeeper/). Since this region of the filesystem was
> > unstable owing filesystem problems in the /tmp folder of the master node,
> > zookeeper communication btw Master and slaves failed and as such HMaster
> > seems to have shut down. I was able to get up the master running on other
> > node with same configurations used previously.
> >
> > There was no changes to the lib directory of the HBase during deployment.
> > The setup was working fine and I was able to run map-reduce programs for
> > importing, exporting and filtering on millions of records in HBase prior
> to
> > HMaster failure.
> >
> > In a production level scenario, which is ideal ? Zookeeper managed
> > ourselves as opposed to zookeeper managed internally by the HBase ?
> >
> > Also, are there any documentation or links anywhere for production level
> > configurations for HBase running on top of the HDFS (Hadoop) ?
> >
>
> See the first section in the refguide, the quickstart section:
> http://hbase.apache.org/book.html#quickstart  The first thing it tells you
> do is edit hbase-site.xml and change where zk is writing its data.  "By
> default, hbase.rootdir is set to /tmp/hbase-${user.name} and similarly so
> for the default ZooKeeper data location which means you'll lose all your
> data whenever your server reboots unless you change it (Most operating
> systems clear /tmp on restart)."
>
> St.Ack
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message