RegionServer not dying, and Master not removing RegionServer

[Matt Davies <ma...@tynt.com>]

All,

You may have seen my previous email regarding Master node crashing, but upon
further research we may have other issues.

I went out to the RegionServer (H3S3) and verified that the process was
running.  There was no meaningful log output, and the last output was from
48 hours ago indicating a log roll.

*Previous State:*
Master died (H3M1)

*Corrective Actions:*
I restarted the master out on H3M1, and it started by trying to recover logs
out on H3S3.  There were many entries like

2011-06-27 11:36:25,639 WARN org.apache.hadoop.hbase.util.FSUtils: Waited
1701352ms for lease recovery on
hdfs://H3M1:8020/hbase/.logs/H3S3,60020,1308946657608/H3S3%3A60020.1309015083129:org.apache.hadoop.hdfs.protocol.AlreadyBeingCreatedException:
failed to create file
/hbase/.logs/H3S3,60020,1308946657608/H3S3%3A60020.1309015083129 for
DFSClient_hb_m_H3M1:60000_1309194478697 on client 10.x.x.x, because this
file is already being created by
DFSClient_hb_rs_H3S3,60020,1308946657608_1308946657819 on 10.5.241.203
at
org.apache.hadoop.hdfs.server.namenode.FSNamesystem.startFileInternal(FSNamesystem.java:1196)
at
org.apache.hadoop.hdfs.server.namenode.FSNamesystem.appendFile(FSNamesystem.java:1284)
at org.apache.hadoop.hdfs.server.namenode.NameNode.append(NameNode.java:596)
at sun.reflect.GeneratedMethodAccessor48.invoke(Unknown Source)
at
sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:25)
at java.lang.reflect.Method.invoke(Method.java:597)
at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:557)
at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:1416)
at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:1412)
at java.security.AccessController.doPrivileged(Native Method)
at javax.security.auth.Subject.doAs(Subject.java:396)
at
org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1115)
at org.apache.hadoop.ipc.Server$Handler.run(Server.java:1410)


Then we decided to kill -9 the regionserver process out on H3S3, and
received the following in the master:

2011-06-27 11:36:26,006 INFO
org.apache.hadoop.hbase.zookeeper.RegionServerTracker: RegionServer
ephemeral node deleted, processing expiration [H3S3,60020,1308946657608]
2011-06-27 11:36:26,006 INFO
org.apache.hadoop.hbase.zookeeper.RegionServerTracker: No HServerInfo found
for H3S3,60020,1308946657608


Ultimately, the master came up and transitioned the regions to a different
regionserver.

Is there a situation where the regionserver may become unresponsive yet the
zookeeper client portion of the process can still check in and tickle
zookeeper so the master thinks it alive?  I think this may be why, in my
previous post, the master tried to assign to H3S3 and died in the attempt.


Thanks!
-Matt

Re: RegionServer not dying, and Master not removing RegionServer

[Jean-Daniel Cryans <jd...@apache.org>]

Did you jstack the region server before killing it? We could see what
was still living, although logs might also give us a clue. As long as
the client ZK thread is alive, the znode will stay up.

About the last message:

> org.apache.hadoop.hbase.zookeeper.RegionServerTracker: No HServerInfo found
> for H3S3,60020,1308946657608

That's usually given when you have a DNS issue, where the hostname and
FQDN are mixed up. See http://hbase.apache.org/book/os.html#dns

J-D

On Mon, Jun 27, 2011 at 10:45 AM, Matt Davies <ma...@tynt.com> wrote:
> All,
>
> You may have seen my previous email regarding Master node crashing, but upon
> further research we may have other issues.
>
> I went out to the RegionServer (H3S3) and verified that the process was
> running.  There was no meaningful log output, and the last output was from
> 48 hours ago indicating a log roll.
>
> *Previous State:*
> Master died (H3M1)
>
> *Corrective Actions:*
> I restarted the master out on H3M1, and it started by trying to recover logs
> out on H3S3.  There were many entries like
>
> 2011-06-27 11:36:25,639 WARN org.apache.hadoop.hbase.util.FSUtils: Waited
> 1701352ms for lease recovery on
> hdfs://H3M1:8020/hbase/.logs/H3S3,60020,1308946657608/H3S3%3A60020.1309015083129:org.apache.hadoop.hdfs.protocol.AlreadyBeingCreatedException:
> failed to create file
> /hbase/.logs/H3S3,60020,1308946657608/H3S3%3A60020.1309015083129 for
> DFSClient_hb_m_H3M1:60000_1309194478697 on client 10.x.x.x, because this
> file is already being created by
> DFSClient_hb_rs_H3S3,60020,1308946657608_1308946657819 on 10.5.241.203
> at
> org.apache.hadoop.hdfs.server.namenode.FSNamesystem.startFileInternal(FSNamesystem.java:1196)
> at
> org.apache.hadoop.hdfs.server.namenode.FSNamesystem.appendFile(FSNamesystem.java:1284)
> at org.apache.hadoop.hdfs.server.namenode.NameNode.append(NameNode.java:596)
> at sun.reflect.GeneratedMethodAccessor48.invoke(Unknown Source)
> at
> sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:25)
> at java.lang.reflect.Method.invoke(Method.java:597)
> at org.apache.hadoop.ipc.RPC$Server.call(RPC.java:557)
> at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:1416)
> at org.apache.hadoop.ipc.Server$Handler$1.run(Server.java:1412)
> at java.security.AccessController.doPrivileged(Native Method)
> at javax.security.auth.Subject.doAs(Subject.java:396)
> at
> org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1115)
> at org.apache.hadoop.ipc.Server$Handler.run(Server.java:1410)
>
>
> Then we decided to kill -9 the regionserver process out on H3S3, and
> received the following in the master:
>
> 2011-06-27 11:36:26,006 INFO
> org.apache.hadoop.hbase.zookeeper.RegionServerTracker: RegionServer
> ephemeral node deleted, processing expiration [H3S3,60020,1308946657608]
> 2011-06-27 11:36:26,006 INFO
> org.apache.hadoop.hbase.zookeeper.RegionServerTracker: No HServerInfo found
> for H3S3,60020,1308946657608
>
>
> Ultimately, the master came up and transitioned the regions to a different
> regionserver.
>
> Is there a situation where the regionserver may become unresponsive yet the
> zookeeper client portion of the process can still check in and tickle
> zookeeper so the master thinks it alive?  I think this may be why, in my
> previous post, the master tried to assign to H3S3 and died in the attempt.
>
>
> Thanks!
> -Matt
>