Thank you. Wellington. I ran the snapshotFormatter too one of the snapshot
files under zookeeper data directory ( there were a lot of them and each of
them 98MB in size).

The snapshot formatter reveals data the ZNode Count is 103,000
and out of which
90,000 of them correspond to /hbase/replication/  - I suspect the next step
is to find root cause for hbase replication lagging



Last login: Fri Mar 15 21:47:03 2019 from cmdev.c.autofocus-1.internal

[azafir@dn1 ~]$ java -cp
/opt/cloudera/parcels/CDH-5.14.2-1.cdh5.14.2.p0.3/jars/zookeeper-3.4.5-cdh5.14.2.jar:/opt/cloudera/parcels/CDH-5.14.2-1.cdh5.14.2.p0.3/lib/zookeeper/lib/*
org.apache.zookeeper.server.SnapshotFormatter
/home/azafir/snapshot.1007d594a7 | grep -E -i -w 'rs' | wc -l

90420

[azafir@dn1 ~]$ java -cp
/opt/cloudera/parcels/CDH-5.14.2-1.cdh5.14.2.p0.3/jars/zookeeper-3.4.5-cdh5.14.2.jar:/opt/cloudera/parcels/CDH-5.14.2-1.cdh5.14.2.p0.3/lib/zookeeper/lib/*
org.apache.zookeeper.server.SnapshotFormatter
/home/azafir/snapshot.1007d594a7  | grep -A3 -B3  count

ZNode Details (*count*=103974):

----

/

  cZxid = 0x00000000000000

--

  ephemeralOwner = 0x00000000000000

  dataLength = 31

----

/hbase/table/

  cZxid = 0x00000c000403ee

  ctime = Thu Aug 16 16:12:25 UTC 2018

  mZxid = 0x00000c0004f53e

--

  ephemeralOwner = 0x00000000000000

  dataLength = 31

----

/hbase/table/

  cZxid = 0x00000c0003f861

  ctime = Thu Aug 16 16:11:40 UTC 2018

  mZxid = 0x00000c0004dc14

--

  ephemeralOwner = 0x00000000000000

  dataLength = 31

----

/hbase/table/xxxx

  cZxid = 0x00000c00060d4d

  ctime = Sat Aug 18 00:19:22 UTC 2018

  mZxid = 0x00000c00067730

--

  ephemeralOwner = 0x00000000000000

  dataLength = 31

----

/hbase/table/xxxxx

  cZxid = 0x00000d266fb835

  ctime = Thu Dec 06 18:47:58 UTC 2018

  mZxid = 0x00000d266fcdf9

--

  ephemeralOwner = 0x00000000000000

  dataLength = 31

----

/hbase/table/xxxxxx

  cZxid = 0x00000c00060d42

  ctime = Sat Aug 18 00:19:09 UTC 2018

  mZxid = 0x00000c00067763

--

  ephemeralOwner = 0x00000000000000

  dataLength = 0

----

/hbase/table-lock/xxxx

  cZxid = 0x00000600004491

  ctime = Fri Jun 29 17:24:02 UTC 2018

  mZxid = 0x00000600004491

--

  ephemeralOwner = 0x00000000000000

  dataLength = 0

----

/hbase/table-lock/wf_artifact*count*

  cZxid = 0x0000060000c927

  ctime = Fri Jun 29 18:03:40 UTC 2018

  mZxid = 0x0000060000c927

--

  ephemeralOwner = 0x00000000000000

  dataLength = 0

----

/hbase/table-lock/wf_invalid_artifact_*count*

  cZxid = 0x00000c00060d4b

  ctime = Sat Aug 18 00:19:22 UTC 2018

  mZxid = 0x00000c00060d4b

--

  ephemeralOwner = 0x00000000000000

  dataLength = 0

----

/hbase/table-lock/wf_artifact*count*_2018_12_05

  cZxid = 0x00000d266fb531

  ctime = Thu Dec 06 18:47:56 UTC 2018

  mZxid = 0x00000d266fb531

--

  ephemeralOwner = 0x00000000000000

  dataLength = 0

----

/hbase/table-lock/xxxx

  cZxid = 0x00000c00060d40

  ctime = Sat Aug 18 00:19:09 UTC 2018

  mZxid = 0x00000c00060d40

--

  ephemeralOwner = 0x00000000000000

  dataLength = 0

----

/hbase/online-snapshot/abort/xxxxx

  cZxid = 0x00001003a90aad

  ctime = Fri Mar 08 22:36:36 UTC 2019

  mZxid = 0x00001003a90aad

--

  ephemeralOwner = 0x00000000000000

  dataLength = 509

----

/hbase/online-snapshot/abort/xxxxx

  cZxid = 0x000010046faf65

  ctime = Sat Mar 09 22:38:04 UTC 2019

  mZxid = 0x000010046faf65


On Tue, Mar 12, 2019 at 6:04 AM Wellington Chevreuil <
[email protected]> wrote:

> Slow replication can lead to too many znodes in ZK. These would not be
> direct children of "/hbase/replication/rs" znode, but of specific RSes
> replication queue. The "108" showed on your "stat" command output is only
> the number of RSes, and would not vary, but further znodes under each of
> those would. To have a better picture of how large is your znode tree, you
> should rather use ZK org.apache.zookeeper.server.SnapshotFormatter tool.
> This will print the whole structure, from a given snapshot file as input.
> Also, there may be some further hints about which znode is exceeding the
> buffer limits by looking at the full stack trace from the jute buffer
> error. Would you be able to share those?
>
> Em ter, 12 de mar de 2019 às 09:52, Asim Zafir <[email protected]>
> escreveu:
>
>>
>> Hi Weillington,
>>
>> Thanks for the response. I greatly appreciate. Yes we have replication
>> enabled and if I stat /hbase/replication/rs  I get the following
>>
>>
>> cZxid = 0x1000001c8
>>
>> ctime = Tue Jun 19 22:23:17 UTC 2018
>>
>> mZxid = 0x1000001c8
>>
>> mtime = Tue Jun 19 22:23:17 UTC 2018
>>
>> pZxid = 0x10000cd555
>>
>> cversion = 8050
>>
>> dataVersion = 0
>>
>> aclVersion = 0
>>
>> ephemeralOwner = 0x0
>>
>> dataLength = 0
>>
>> numChildren = 108
>>
>> how should I be able to see analyze the znode utilzation in this case and
>> more specifiically how is it impacting the jute-buffer size.. I can see
>> numChild nodes under /hbase/replication is 108 but how does it correspond
>> to zk jute buffer reaching max value?
>>
>> Also it is not clear the timestamp on this znodes isnt' incrementing. I
>> see the timestamp is still showing 2018 date.
>>
>> Thanks,
>> asim
>>
>>
>> -------------->>>>>
>>
>>
>>
>> This jute buffer len error generally means a given znode being watched/read
>> had grown too large to fit into the buffer. It's not specific to number of
>> watches attached, but amount of info stored in it, for example, too many
>> children znode under a given znode. In order to understand what's behind
>> the error, you should analyse your zookeeper znodes tree, you may have a
>> hint by looking at zookeeper snapashot files. Would you have replication
>> enabled on this cluster? A common cause for such errors in hbase is when
>> replication is slow/stuck, and source cluster is under heavy write load,
>> causing replication queue to grow much faster than it's ability to drain,
>> which will imply on many znodes created under "replication" znode.
>>
>>
  • Re: Wellington Chevreuil
    • Re: Asim Zafir

Reply via email to