Great. I had updated jvm version to 1.6.0_31 yesterday, and it works
well till now. Thanks a lot.
On 03/12/2013 12:07 AM, Brock Noland wrote:
You are using a known bad jvm version. I would upgrade:
http://wiki.apache.org/hadoop/HadoopJavaVersions
On Mon, Mar 11, 2013 at 5:49 AM, larryzhang <[email protected]
<mailto:[email protected]>> wrote:
Hi,
I want to collect and analyse user logs every 5 minutes. Now
we have origin log file which generated by nginx and divided by
hour, about 30,000,000 logs per hour. The log format is like this:
60.222.199.118 - - [11/Mar/2013:16:00:00 +0800] "GET ....
Because I want to firstly collect the logs into file. so I
wrote a FileEventSink, just did some modification based on
org.apache.flume.sink.hdfs.BucketWriter.java and
org.apache.flume.sink.hdfs.HDFSEventSink.java. Following is my
flume config file:
==================
a1.sources = r1
a1.channels = c1
a1.sinks = k1
a1.sources.r1.type = cn.larry.flume.source.MyExecSource //I
need to fetch time and other info into headers, so I add these
logics based on ExecSource
a1.sources.r1.command = tail -n +0 -F /data2/log/log_2013031117
<tel:2013031117>.log
a1.sources.r1.channels = c1
a1.sources.r1.batchSize = 1 //I set this to 1 because
otherwise it will lost data at the end of the log file, I apply
this patch https://issues.apache.org/jira/browse/FLUME-1819 but it
seems no help...
a1.channels.c1.type = memory
a1.channels.c1.capacity = 1000000
a1.channels.c1.transactionCapacity = 10000
a1.sinks.k1.type = cn.larry.flume.sink.FileEventSink
a1.sinks.k1.channel = c1
a1.sinks.k1.file.path = /opt/livedata/%Y%m%d/%H
a1.sinks.k1.file.filePrefix = log-%Y%m%d%H%M
a1.sinks.k1.file.round = true
a1.sinks.k1.file.roundValue = 5
a1.sinks.k1.file.roundUnit = minute
a1.sinks.k1.file.rollInterval=300
a1.sinks.k1.file.rollSize=0
a1.sinks.k1.file.rollCount=0
a1.sinks.k1.file.batchSize=100
And because I need to change the source log file name each
hour, so I wrote a script, which does 3 things:
1. At the 1st minute per hour:
->copy a new config file, which just change the source
log file name(a1.sources.r1.command = tail -n +0 -F
/data2/log/log_<new time>.log)
->start new flume process which use the new conifg. (I
did this because if flume process die, it won't affect next hour)
2. At the 30th minute per hour:
-> kill the flume process of last hour.
This project has been run more than 10 days, most of time it
works well, but sometimes flume process crashed due to JVM error,
about once every 2 days! I used jvm version 1.6.0_27-ea. Here's
the log of error which happened on 2013-03-11:
2013-03-11 05:45:11,302 (file-k1-roll-timer-0) [INFO -
cn.larry.flume.sink.FileBucketWriter.renameBucket(FileBucketWriter.java:408)]
Renaming /opt/livedata/20130311/05
<tel:20130311%2F05>/log-201303110540.1362951611255.tmp to
/opt/livedata/20130311/05
<tel:20130311%2F05>/log-201303110540.1362951611255
#
# A fatal error has been detected by the Java Runtime Environment:
#
# SIGSEGV (0xb) at pc=0x00002b4247be034e, pid=1463, tid=1098979648
#
# JRE version: 6.0_18-b07
# Java VM: Java HotSpot(TM) 64-Bit Server VM (16.0-b13 mixed mode
linux-amd64 )
# Problematic frame:
# V [libjvm.so+0x2de34e]
#
# An error report file with more information is saved as:
# /opt/scripts/tvhadoop/flume/flume-1.3.0/bin/hs_err_pid1463.log
#
# If you would like to submit a bug report, please visit:
# http://java.sun.com/webapps/bugreport/crash.jsp
#
+ exec /usr/local/jdk/bin/java -Xmx2048m
-Dflume.root.logger=INFO,console -cp
'/opt/tvhadoop/apache-flume-1.3.1-bin/conf:/opt/tvhadoop/apache-flume-1.3.1-bin/lib/*'
-Djava.library.path= org.apache.flume.node.Application -f
/opt/tvhadoop/apache-flume-1.3.1-bin/conf/flume_2013031106
<tel:2013031106>.conf -n a1
and the jvm dump file is in the attachments.
I wonder how to handle this problem.
And another question is about the execSource, I don't know why
it loss data if batchsize > 1. If I use file channel, I must make
a large batchsize to fulfill the throughputs...
Thanks & Best regards,
larry
--
Apache MRUnit - Unit testing MapReduce -
http://incubator.apache.org/mrunit/
