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 <[email protected] > 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 < > [email protected]> 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
