Ernst,
Can you share the logs just before the crash. Specially the
GCInspector logs.Check the last reported used heap space and whether
it was close to the threshold for full GC.
Also how frequent are your OOM crashes?
The cassandra default for kicking in full GC is 75% (
-XX:CMSInitiatingOccupancyFraction=75). You can try lowering that to
maybe 65 or some lower number and see if that helps. Basically you
will have a little more frequent GCs which will take a little less
time than before. And possibly that unused heap buffer will provide
for the operation which is causing OOM.
I saw an OOM on one node after 2 weeks. The heap used was close to the
GC threshold and full GC takes around 80 seconds.
-Adi
2011/8/24 Ernst D Schoen-René <er...@peoplebrowsr.com>:
> So, we're on 8, so I don't think there's a key cache setting. Am I wrong?
>
> here's my newest crash log:
>
> ERROR [Thread-210] 2011-08-24 06:29:53,247 AbstractCassandraDaemon.java
> (line 113) Fatal exception in thread Thread[Thread-210,5,main]
> java.util.concurrent.RejectedExecutionException: ThreadPoolExecutor has shut
> down
> at
> org.apache.cassandra.concurrent.DebuggableThreadPoolExecutor$1.rejectedExecution(DebuggableThreadPoolExecutor.java:73)
> at
> java.util.concurrent.ThreadPoolExecutor.reject(ThreadPoolExecutor.java:767)
> at
> java.util.concurrent.ThreadPoolExecutor.execute(ThreadPoolExecutor.java:658)
> at
> org.apache.cassandra.net.MessagingService.receive(MessagingService.java:444)
> at
> org.apache.cassandra.net.IncomingTcpConnection.run(IncomingTcpConnection.java:117)
> INFO [ScheduledTasks:1] 2011-08-24 06:29:53,249 StatusLogger.java (line 65)
> InternalResponseStage 0 0
> INFO [ScheduledTasks:1] 2011-08-24 06:29:53,250 StatusLogger.java (line 65)
> HintedHandoff 1 16
> INFO [ScheduledTasks:1] 2011-08-24 06:29:53,250 StatusLogger.java (line 69)
> CompactionManager n/a 1298
> INFO [ScheduledTasks:1] 2011-08-24 06:29:53,250 StatusLogger.java (line 81)
> MessagingService n/a 1,0
> INFO [ScheduledTasks:1] 2011-08-24 06:29:53,250 StatusLogger.java (line 85)
> ColumnFamily Memtable ops,data Row cache size/cap Key cache
> size/cap
>
>
>
> On 8/24/11 2:38 AM, aaron morton wrote:
>
> First thing is are you on 0.8 ? It has some automagical memory management
> that is both automatic and
> magical http://thelastpickle.com/2011/05/04/How-are-Memtables-measured/
> Secondly if you are OOM'ing you need to look at how much memory your schema
> is taking. See the link above, or just use 0.8
> Thirdly, try to reduce the cache as Adi noted.
> Finally try to work out when it's OOM'ing. Is it during a compaction, a
> repair, during peak load or just regular operations. With 1TB of data i'd be
> taking a look at the compaction or repair processes to see whats happening
> GC wise. Also check the GC settings as described in cassandra-env.sh
> Hope that helps.
>
> -----------------
> Aaron Morton
> Freelance Cassandra Developer
> @aaronmorton
> http://www.thelastpickle.com
> On 24/08/2011, at 12:38 PM, Ernst D Schoen-René wrote:
>
> Thanks,
> We had already been running cassandra with a larger heap size, but it
> meant that java took way too long between garbage collections. The advice
> I'd found was to set the heap size at the 8 we're running at. It was ok for
> a while, but now some nodes crash. It's definitely our experience that
> adding more memory per node actually makes things worse eventually, as java
> starts eating up too many resources for it to handle.
>
>
> On 8/23/11 5:28 PM, Adi wrote:
>
> 2011/8/23 Ernst D Schoen-René<er...@peoplebrowsr.com>:
>
> Hi,
>
> I'm running a 16-node cassandra cluster, with a reasonably large amount of
>
> data per node (~1TB). Nodes have 16G ram, but heap is set to 8G.
>
> The nodes keep stopping with this output in the log. Any ideas?
>
> ERROR [Thread-85] 2011-08-23 21:00:38,723 AbstractCassandraDaemon.java (line
>
> 113) Fatal exception in thread Thread[Thread-85,5,main]
>
> java.lang.OutOfMemoryError: Java heap space
>
> ERROR [ReadStage:568] 2011-08-23 21:00:38,723 AbstractCassandraDaemon.java
>
> (line 113) Fatal exception in thread Thread[ReadStage:568,5,main]
>
> java.lang.OutOfMemoryError: Java heap space
>
> INFO [HintedHandoff:1] 2011-08-23 21:00:38,720 HintedHandOffManager.java
>
> (line 320) Started hinted handoff for endpoint /10.28.0.184
>
> INFO [GossipStage:2] 2011-08-23 21:00:50,751 Gossiper.java (line 606)
>
> InetAddress /10.29.20.67 is now UP
>
> ERROR [Thread-34] 2011-08-23 21:00:50,525 AbstractCassandraDaemon.java (line
>
> 113) Fatal exception in thread Thread[Thread-34,5,main]
>
> java.util.concurrent.RejectedExecutionException: ThreadPoolExecutor has shut
>
> down
>
> at
>
> org.apache.cassandra.concurrent.DebuggableThreadPoolExecutor$1.rejectedExecution(DebuggableThreadPoolExecutor.java:73)
>
> at
>
> java.util.concurrent.ThreadPoolExecutor.reject(ThreadPoolExecutor.java:767)
>
> at
>
> java.util.concurrent.ThreadPoolExecutor.execute(ThreadPoolExecutor.java:658)
>
> at
>
> org.apache.cassandra.net.MessagingService.receive(MessagingService.java:444)
>
> at
>
> org.apache.cassandra.net.IncomingTcpConnection.run(IncomingTcpConnection.java:117)
>
> ERROR [Thread-36] 2011-08-23 21:00:50,518 AbstractCassandraDaemon.java (line
>
> 113) Fatal exception in thread Thread[Thread-36,5,main]
>
> java.util.concurrent.RejectedExecutionException: ThreadPoolExecutor has shut
>
> down
>
> at
>
> org.apache.cassandra.concurrent.DebuggableThreadPoolExecutor$1.rejectedExecution(DebuggableThreadPoolExecutor.java:73)
>
> at
>
> java.util.concurrent.ThreadPoolExecutor.reject(ThreadPoolExecutor.java:767)
>
> at
>
> java.util.concurrent.ThreadPoolExecutor.execute(ThreadPoolExecutor.java:658)
>
> at
>
> org.apache.cassandra.net.MessagingService.receive(MessagingService.java:444)
>
> at
>
> org.apache.cassandra.net.IncomingTcpConnection.run(IncomingTcpConnection.java:117)
>
> INFO [GossipTasks:1] 2011-08-23 21:00:50,466 Gossiper.java (line 620)
>
> InetAddress /10.29.20.67 is now dead.
>
> INFO [HintedHandoff:1] 2011-08-23 21:00:50,751 HintedHandOffManager.java
>
> (line 376) Finished hinted handoff of 0 rows to endpoint /10.28.0.184
>
> ERROR [Thread-33] 2011-08-23 21:01:05,048 AbstractCassandraDaemon.java (line
>
> 113) Fatal exception in thread Thread[Thread-33,5,main]
>
> java.util.concurrent.RejectedExecutionException: ThreadPoolExecutor has shut
>
> down
>
> at
>
> org.apache.cassandra.concurrent.DebuggableThreadPoolExecutor$1.rejectedExecution(DebuggableThreadPoolExecutor.java:73)
>
> at
>
> java.util.concurrent.ThreadPoolExecutor.reject(ThreadPoolExecutor.java:767)
>
> at
>
> java.util.concurrent.ThreadPoolExecutor.execute(ThreadPoolExecutor.java:658)
>
> at
>
> org.apache.cassandra.net.MessagingService.receive(MessagingService.java:444)
>
> at
>
> org.apache.cassandra.net.IncomingTcpConnection.run(IncomingTcpConnection.java:117)
>
> ERROR [Thread-128] 2011-08-23 21:01:05,048 AbstractCassandraDaemon.java
>
> (line 113) Fatal exception in thread Thread[Thread-128,5,main]
>
> java.util.concurrent.RejectedExecutionException: ThreadPoolExecutor has shut
>
> down
>
> at
>
> org.apache.cassandra.concurrent.DebuggableThreadPoolExecutor$1.rejectedExecution(DebuggableThreadPoolExecutor.java:73)
>
> at
>
> java.util.concurrent.ThreadPoolExecutor.reject(ThreadPoolExecutor.java:767)
>
> at
>
> java.util.concurrent.ThreadPoolExecutor.execute(ThreadPoolExecutor.java:658)
>
> at
>
> org.apache.cassandra.net.MessagingService.receive(MessagingService.java:444)
>
> at
>
> org.apache.cassandra.net.IncomingTcpConnection.run(IncomingTcpConnection.java:117)
>
> root@cass1:~#
>
>
>
> You can try the cargo cult solution of upping the heap to 12GB and
>
> see if the nodes stabilize. We have a 4-node cluster with 2-3 TB data
>
> per node and that was the heap at which it the nodes were managing to
>
> serve requests without running out of memory. Ultimately we ordered
>
> more memory and are running it with 24 GB heap and the cluster has
>
> been stable without complains.
>
> Other things you can do for reducing memory usage if they are
>
> appropriate for your read/write profile:
>
> a) reduce memtable throughput(most reduction in mem footprint)
>
> b) disable row caching
>
> c) reduce/disable key caching(least reduction)
>
> Ultimately you will have to tune based on your
>
> 1) row sizes
>
> 2) read/write load
>
> -Adi
>
>
>
>
