hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Tao Xiao <xiaotao.cs....@gmail.com>
Subject Re: All regions stay on two nodes out of 18 nodes
Date Thu, 17 Apr 2014 05:29:59 GMT
BTW, the region server reported:

2014-04-16 11:30:31,890 INFO  [RS_OPEN_REGION-b05:60020-0]
handler.OpenRegionHandler: Opening of region {ENCODED =>
6886ac98a71a47dc78a9e0ab5b3f07cd, NAME =>
'E_MP_DAY_READ_20140315,,1396363260513.6886ac98a71a47dc78a9e0ab5b3f07cd.',
STARTKEY => '', ENDKEY => '170000346762_20140315'} failed, transitioning
from OPENING to FAILED_OPEN in ZK, expecting version 1

Here what does "expecting version 1" indicate?


2014-04-17 13:27 GMT+08:00 Tao Xiao <xiaotao.cs.nju@gmail.com>:

> Take the region
> E_MP_DAY_READ_20140315,,1396363260513.6886ac98a71a47dc78a9e0ab5b3f07cdfor example.
>
> I checked the master's log and the region server (*b05.jsepc.com
> <http://b05.jsepc.com>*) log, and found that in the master log there are
> just 4 logging lines about that region and the logging time was as early as
> 2014-04-02.
>
> In the region server's log, there are more logging lines about that
> region, but the logging time is quite recent, say 2014-04-16. It seems that
> the master has lost control of that region for a long time, but the region
> server is still managing that region although it cannot open it.
>
> The master log is here <http://pastebin.com/6J6v9tSg>, and the region
> server log is here <http://pastebin.com/fbuu0RpC>.
>
>
> 2014-04-17 9:34 GMT+08:00 Ted Yu <yuzhihong@gmail.com>:
>
> You can pick a region which is stuck in transition, find which region
>> server is hosting it and search region server log on that server.
>>
>> By correlating events from master and region server logs, you should see
>> what is happening.
>>
>>
>> On Wed, Apr 16, 2014 at 6:24 PM, Tao Xiao <xiaotao.cs.nju@gmail.com>
>> wrote:
>>
>> > Actually, open that link and then click on the picture, it will zoom in
>> and
>> > become quite clear.
>> >
>> > I checked the HMaster UI just now and I am sure that these regions are
>> > always in transition,  I suppose there would be some exceptions
>> happening.
>> > How to prevent regions from being in transition for a long time ?
>> >
>> >
>> > 2014-04-17 9:00 GMT+08:00 Ted Yu <yuzhihong@gmail.com>:
>> >
>> > > The picture is not very clear.
>> > > I don't see E_MP_DAY_READ having regions in transition.
>> > >
>> > > Anyway, as long as there is region in transition, balancer would not
>> run.
>> > >
>> > > Cheers
>> > >
>> > >
>> > > On Wed, Apr 16, 2014 at 5:52 PM, Tao Xiao <xiaotao.cs.nju@gmail.com>
>> > > wrote:
>> > >
>> > > > Ted,
>> > > >
>> > > > I can see some regions of other tables in transition now , but I'm
>> not
>> > > sure
>> > > > how long have them been in transition and I will check the HBase
>> master
>> > > UI
>> > > > later. Here is the
>> > > > screenshot<
>> > > >
>> > >
>> >
>> http://picpaste.com/Regions_in_Transition_-_2014-04-17_08-38-qyf5anz8.png
>> > > > >.
>> > > > From the screenshot, there is a region with state of FAILED_OPEN,
>> which
>> > > is
>> > > > in red, and there are 9 regions in transition for more than 60
>> seconds.
>> > > >
>> > > > Note that the table whose regions all stay in 2 nodes is
>> E_MP_DAY_READ,
>> > > > while the other tables shown in the screenshot are named as
>> > > > E_MP_DAY_READ_20140315, E_MP_DAY_READ_20140322,
>> E_MP_DAY_READ_20140324,
>> > > and
>> > > > so on.
>> > > >
>> > > > Thanks.
>> > > >
>> > > >
>> > > > 2014-04-16 23:10 GMT+08:00 Ted Yu <yuzhihong@gmail.com>:
>> > > >
>> > > > > bq. found some regions of other tables in transition, not of
this
>> > > table.
>> > > > >
>> > > > > That can explain why "balancer" command returned false.
>> > > > > Are those regions stuck in transition ?
>> > > > >
>> > > > > Cheers
>> > > > >
>> > > > >
>> > > > > On Tue, Apr 15, 2014 at 10:47 PM, Tao Xiao <
>> xiaotao.cs.nju@gmail.com
>> > >
>> > > > > wrote:
>> > > > >
>> > > > > > The command "balance_switch true" returns true, but the
command
>> > > > > "balancer"
>> > > > > > returns false. I checked the HMaster UI and found some regions
>> of
>> > > other
>> > > > > > tables in transition, not of this table.
>> > > > > >
>> > > > > > This table's name is E_MP_DAY_READ, I did grep it in the
master
>> log
>> > > and
>> > > > > > found only the following lines:
>> > > > > >
>> > > > > > 2014-04-15 15:50:59,925 INFO
>> >  [MASTER_SERVER_OPERATIONS-b03:60000-1]
>> > > > > > handler.ServerShutdownHandler: Skip assigning region
>> > > > > >
>> > > > > >
>> > > > >
>> > > >
>> > >
>> >
>> E_MP_DAY_READ,160001123745_2014-01-25:00:00:00,1395753408476.ba5c8291f8dad37d5b9621b7334c17b0.
>> > > > > > because it has been opened in a04.jsepc.com,60020,1397548219084
>> > > > > > 2014-04-15 15:50:59,926 INFO
>> >  [MASTER_SERVER_OPERATIONS-b03:60000-1]
>> > > > > > handler.ServerShutdownHandler: Skip assigning region
>> > > > > >
>> > > > > >
>> > > > >
>> > > >
>> > >
>> >
>> E_MP_DAY_READ,300007915618_2014-03-13:00:00:00,1395994146202.ec4e397baffd1cc40bdc18ce0ab2f28a.
>> > > > > > because it has been opened in a04.jsepc.com,60020,1397548219084
>> > > > > > 2014-04-15 15:50:59,926 INFO
>> >  [MASTER_SERVER_OPERATIONS-b03:60000-1]
>> > > > > > handler.ServerShutdownHandler: Skip assigning region
>> > > > > >
>> > > > > >
>> > > > >
>> > > >
>> > >
>> >
>> E_MP_DAY_READ,300013608840_2014-02-21:00:00:00,1395749573711.744bab52befec279a7ee97497801e10f.
>> > > > > > because it has been opened in a04.jsepc.com,60020,1397548219084
>> > > > > > 2014-04-15 15:50:59,937 INFO
>> >  [MASTER_SERVER_OPERATIONS-b03:60000-2]
>> > > > > > handler.ServerShutdownHandler: Skip assigning region
>> > > > > >
>> > > > > >
>> > > > >
>> > > >
>> > >
>> >
>> E_MP_DAY_READ,300000497780_2014-01-23:00:00:00,1395746363941.79b831e698053b1005f7a97c9f2a6ddc.
>> > > > > > because it has been opened in a04.jsepc.com,60020,1397548219084
>> > > > > > 2014-04-15 15:50:59,938 INFO
>> >  [MASTER_SERVER_OPERATIONS-b03:60000-2]
>> > > > > > handler.ServerShutdownHandler: Skip assigning region
>> > > > > >
>> > > > > >
>> > > > >
>> > > >
>> > >
>> >
>> E_MP_DAY_READ,300008188567_2014-03-04:00:00:00,1395756104426.eb1806c2dc5833152b6b5e7b5e4a88b8.
>> > > > > > because it has been opened in a04.jsepc.com,60020,1397548219084
>> > > > > > 2014-04-15 15:50:59,940 INFO
>> >  [MASTER_SERVER_OPERATIONS-b03:60000-2]
>> > > > > > handler.ServerShutdownHandler: Skip assigning region
>> > > > > >
>> > > > > >
>> > > > >
>> > > >
>> > >
>> >
>> E_MP_DAY_READ,300016987143_2014-01-21:00:00:00,1395986789897.e4d143865d354bdc2a427c1f00df6ad7.
>> > > > > > because it has been opened in a04.jsepc.com,60020,1397548219084
>> > > > > >
>> > > > > > so few logging lines about it, looks strange ?
>> > > > > >
>> > > > > >
>> > > > > > BTW, I can spread the regions of this table evenly across
the
>> whole
>> > > > > cluster
>> > > > > > after I shutdown the two region servers where the regions
of
>> this
>> > > table
>> > > > > > resided originally.
>> > > > > >
>> > > > > >
>> > > > > > 2014-04-15 19:47 GMT+08:00 Ted Yu <yuzhihong@gmail.com>:
>> > > > > >
>> > > > > > > Is load balancer enabled ?
>> > > > > > >
>> > > > > > > Can you grep this table in master log and pastebin
what you
>> > found ?
>> > > > > > >
>> > > > > > > Cheers
>> > > > > > >
>> > > > > > > On Apr 15, 2014, at 4:40 AM, Tao Xiao <
>> xiaotao.cs.nju@gmail.com>
>> > > > > wrote:
>> > > > > > >
>> > > > > > > > I am using HDP 2.0.6, which has 18 nodes(region
servers).
>> One
>> > of
>> > > my
>> > > > > > HBase
>> > > > > > > > tables has 50 regions but I found that the 50
regions all
>> stay
>> > in
>> > > > > just
>> > > > > > > two
>> > > > > > > > nodes, not spread evenly in the 18 nodes. I did
not
>> pre-create
>> > > > splits
>> > > > > > so
>> > > > > > > > this table was gradually split into 50 regions
itself.
>> > > > > > > >
>> > > > > > > > I'd like to know why all the regions stay in just
two nodes,
>> > not
>> > > > the
>> > > > > 18
>> > > > > > > > nodes of the cluster, and how to spread the regions
evenly
>> > across
>> > > > all
>> > > > > > the
>> > > > > > > > region servers. Thanks.
>> > > > > > >
>> > > > > >
>> > > > >
>> > > >
>> > >
>> >
>>
>
>

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