hadoop-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Philippe Signoret <philippe.signo...@gmail.com>
Subject Re: Please help me with heartbeat storm
Date Wed, 29 May 2013 14:00:40 GMT
This might be relevant: https://issues.apache.org/jira/browse/MAPREDUCE-4478

"There are two configuration items to control the TaskTracker's heartbeat
interval. One is *mapreduce.tasktracker.outofband.heartbeat*. The other is*
mapreduce.tasktracker.outofband.heartbeat.damper*. If we set *
mapreduce.tasktracker.outofband.heartbeat* with true and set*
mapreduce.tasktracker.outofband.heartbeat.damper* with default value
(1000000), TaskTracker may send heartbeat without any interval."


Philippe

-------------------------------
*Philippe Signoret*


On Tue, May 28, 2013 at 4:44 AM, Rajesh Balamohan <
rajesh.balamohan@gmail.com> wrote:

> Default value of CLUSTER_INCREMENT is 100. Math.max(1000* 29/100, 3000) =
> 3000 always. This is the reason why you are seeing so many heartbeats. *You
> might want to set it to 1 or 5.* This would increase the time taken to
> send the heartbeat from TT to JT.
>
>
> ~Rajesh.B
>
>
> On Mon, May 27, 2013 at 2:12 PM, Eremikhin Alexey <
> a.eremihin@corp.badoo.com> wrote:
>
>>  Hi!
>>
>> Tried 5 seconds. Less number of nodes get into storm, but still they do.
>> Additionaly update of ntp service helped a little.
>>
>> Initially almost 50% got into storming each MR job. But after ntp update
>> and and increasing heart-beatto 5 seconds level is around 10%.
>>
>>
>> On 26/05/13 10:43, murali adireddy wrote:
>>
>> Hi ,
>>
>>  Just try this one.
>>
>>  in the file "hdfs-site.xml" try to add the below property
>> "dfs.heartbeat.interval" and value  in seconds.
>>
>>  Default value is '3' seconds. In your case increase value.
>>
>>  <property>
>>  <name>dfs.heartbeat.interval</name>
>>  <value>3</value>
>> </property>
>>
>>  You can find more properties and default values in the below link.
>>
>>
>> http://hadoop.apache.org/docs/current/hadoop-project-dist/hadoop-hdfs/hdfs-default.xml
>>
>>
>>  Please let me know is the above solution worked for you ..?
>>
>>
>>
>>
>>  On Fri, May 24, 2013 at 6:40 PM, Eremikhin Alexey <
>> a.eremihin@corp.badoo.com> wrote:
>>
>>> Hi all,
>>> I have 29 servers hadoop cluster in almost default configuration.
>>> After installing Hadoop 1.0.4 I've noticed that JT and some TT waste CPU.
>>> I started stracing its behaviour and found that some TT send heartbeats
>>> in an unlimited ways.
>>> It means hundreds in a second.
>>>
>>> Daemon restart solves the issue, but even easiest Hive MR returns issue
>>> back.
>>>
>>> Here is the filtered strace of heartbeating process
>>>
>>> hadoop9.mlan:~$ sudo strace -tt -f -s 10000 -p 6032 2>&1  | grep 6065
|
>>> grep write
>>>
>>>
>>> [pid  6065] 13:07:34.801106 write(70,
>>> "\0\0\1\30\0:\316N\0\theartbeat\0\0\0\5\0*org.apache.hadoop.mapred.TaskTrackerStatus\0*org.apache.hadoop.mapred.TaskTrackerStatus.tracker_hadoop9.mlan:localhost/
>>> 127.0.0.1:52355\fhadoop9.mlan\0\0\303\214\0\0\0\0\0\0\0\2\0\0\0\2\213\1\367\373\200\0\214\367\223\220\0\213\1\341p\220\0\214\341\351\200\0\377\377\213\6\243\253\200\0\214q\r\33\300\215$\205\266\4B\16\333n\0\0\0\0\1\0\0\0\0\0\0\0\0\0\0\7boolean\0\0\7boolean\0\0\7boolean\1\0\5short\316\30",
>>> 284) = 284
>>> [pid  6065] 13:07:34.807968 write(70,
>>> "\0\0\1\30\0:\316O\0\theartbeat\0\0\0\5\0*org.apache.hadoop.mapred.TaskTrackerStatus\0*org.apache.hadoop.mapred.TaskTrackerStatus.tracker_hadoop9.mlan:localhost/
>>> 127.0.0.1:52355\fhadoop9.mlan\0\0\303\214\0\0\0\0\0\0\0\2\0\0\0\2\213\1\367\373\200\0\214\367\223\220\0\213\1\341p\220\0\214\341\351\200\0\377\377\213\6\243\253\200\0\214q\r\33\312\215$\205\266\4B\16\333n\0\0\0\0\1\0\0\0\0\0\0\0\0\0\0\7boolean\0\0\7boolean\0\0\7boolean\1\0\5short\316\31",
>>> 284 <unfinished ...>
>>> [pid  6065] 13:07:34.808080 <... write resumed> ) = 284
>>> [pid  6065] 13:07:34.814473 write(70,
>>> "\0\0\1\30\0:\316P\0\theartbeat\0\0\0\5\0*org.apache.hadoop.mapred.TaskTrackerStatus\0*org.apache.hadoop.mapred.TaskTrackerStatus.tracker_hadoop9.mlan:localhost/
>>> 127.0.0.1:52355\fhadoop9.mlan\0\0\303\214\0\0\0\0\0\0\0\2\0\0\0\2\213\1\367\373\200\0\214\367\223\220\0\213\1\341p\220\0\214\341\351\200\0\377\377\213\6\243\253\200\0\214q\r\33\336\215$\205\266\4B\16\333n\0\0\0\0\1\0\0\0\0\0\0\0\0\0\0\7boolean\0\0\7boolean\0\0\7boolean\1\0\5short\316\32",
>>> 284 <unfinished ...>
>>> [pid  6065] 13:07:34.814595 <... write resumed> ) = 284
>>> [pid  6065] 13:07:34.820960 write(70,
>>> "\0\0\1\30\0:\316Q\0\theartbeat\0\0\0\5\0*org.apache.hadoop.mapred.TaskTrackerStatus\0*org.apache.hadoop.mapred.TaskTrackerStatus.tracker_hadoop9.mlan:localhost/
>>> 127.0.0.1:52355\fhadoop9.mlan\0\0\303\214\0\0\0\0\0\0\0\2\0\0\0\2\213\1\367\373\200\0\214\367\223\220\0\213\1\341p\220\0\214\341\351\200\0\377\377\213\6\243\253\200\0\214q\r\33\336\215$\205\266\4B\16\333n\0\0\0\0\1\0\0\0\0\0\0\0\0\0\0\7boolean\0\0\7boolean\0\0\7boolean\1\0\5short\316\33",
>>> 284 <unfinished ...>
>>>
>>>
>>> Please help me to stop this storming 8(
>>>
>>>
>>
>>
>
>
> --
> ~Rajesh.B
>

Mime
View raw message