You are viewing a plain text version of this content. The canonical link for it is here.
Posted to solr-user@lucene.apache.org by "Vauthrin, Laurent" <La...@disney.com> on 2009/11/09 20:22:26 UTC
Solr on OOM
Hello,
One of our deployed Solr (1.3) setup is having out of memory issues and
I'm not sure how to troubleshoot it. I've read a few posts (including
http://old.nabble.com/Debugging-Solr-memory-usage-heap-problems-ts883279
4.html#a8832794) but I think this situation is slightly different.
Here's the setup:
1 master and 1 slave are located on a the same VM (using a 64-bit JVM)
1 slave running on its own VM (using a 64-bit JVM)
>From what I've been told, nothing else is running on those VMs.
Index size is about 100-200 MB.
Solrconfig.xml cache settings:
<filterCache class="solr.LRUCache" size="15000" initialSize="5000"
autowarmCount="5000"/>
<queryResultCache class="solr.LRUCache" size="15000"
initialSize="5000" autowarmCount="5000"/>
<documentCache class="solr.LRUCache" size="15000"
initialSize="5000"/>
Both slaves at some point have gone out of memory (though not both at
the same time) when receiving a moderate load of queries (a few queries
per second - don't have an exact stat here). We started with a heap
size of 1GB and ended up having to bump it up to 3.5GB. It seems really
odd that we'd have to have a heap size that large when the index itself
is not really big. Any thoughts on what could be really off here? Is
there a way to determine the cache sizes in bytes? I noticed that there
was a thread about other issues running Solr on VMs, has anyone else had
problems using VMWare? From what I'm told, it seems like moving to
physical servers won't be a fast/easy change so I'm looking for any help
I can get for this configuration.
Thanks,
Laurent Vauthrin
RE: Solr on OOM
Posted by "Vauthrin, Laurent" <La...@disney.com>.
Ok, I'll see if I can get their heap dump. Thanks.
-----Original Message-----
From:
solr-user-return-29005-Laurent.Vauthrin=disney.com@lucene.apache.org
[mailto:solr-user-return-29005-Laurent.Vauthrin=disney.com@lucene.apache
.org] On Behalf Of Otis Gospodnetic
Sent: Monday, November 09, 2009 2:17 PM
To: solr-user@lucene.apache.org
Subject: Re: Solr on OOM
Laurent,
The autowarmCounts look biggest, but they are probably not causing OOMs.
Maybe you can see how big the caches are right before you OOM.
Or you can also start the JVM with -XX:+HeapDumpOnOutOfMemoryError and
even specify the file where the heap should be dumped. You can then
analyze it and see what's eating the memory.
Otis
--
Sematext is hiring -- http://sematext.com/about/jobs.html?mls
Lucene, Solr, Nutch, Katta, Hadoop, HBase, UIMA, NLP, NER, IR
----- Original Message ----
> From: "Vauthrin, Laurent" <La...@disney.com>
> To: solr-user@lucene.apache.org
> Sent: Mon, November 9, 2009 2:22:26 PM
> Subject: Solr on OOM
>
> Hello,
>
>
>
> One of our deployed Solr (1.3) setup is having out of memory issues
and
> I'm not sure how to troubleshoot it. I've read a few posts (including
>
http://old.nabble.com/Debugging-Solr-memory-usage-heap-problems-ts883279
> 4.html#a8832794) but I think this situation is slightly different.
>
>
>
> Here's the setup:
>
> 1 master and 1 slave are located on a the same VM (using a 64-bit JVM)
>
> 1 slave running on its own VM (using a 64-bit JVM)
>
> From what I've been told, nothing else is running on those VMs.
>
> Index size is about 100-200 MB.
>
>
>
> Solrconfig.xml cache settings:
>
>
>
>
> autowarmCount="5000"/>
>
>
> initialSize="5000" autowarmCount="5000"/>
>
>
> initialSize="5000"/>
>
>
>
> Both slaves at some point have gone out of memory (though not both at
> the same time) when receiving a moderate load of queries (a few
queries
> per second - don't have an exact stat here). We started with a heap
> size of 1GB and ended up having to bump it up to 3.5GB. It seems
really
> odd that we'd have to have a heap size that large when the index
itself
> is not really big. Any thoughts on what could be really off here?
Is
> there a way to determine the cache sizes in bytes? I noticed that
there
> was a thread about other issues running Solr on VMs, has anyone else
had
> problems using VMWare? From what I'm told, it seems like moving to
> physical servers won't be a fast/easy change so I'm looking for any
help
> I can get for this configuration.
>
>
>
> Thanks,
> Laurent Vauthrin
Re: Solr on OOM
Posted by Otis Gospodnetic <ot...@yahoo.com>.
Laurent,
The autowarmCounts look biggest, but they are probably not causing OOMs. Maybe you can see how big the caches are right before you OOM.
Or you can also start the JVM with -XX:+HeapDumpOnOutOfMemoryError and even specify the file where the heap should be dumped. You can then analyze it and see what's eating the memory.
Otis
--
Sematext is hiring -- http://sematext.com/about/jobs.html?mls
Lucene, Solr, Nutch, Katta, Hadoop, HBase, UIMA, NLP, NER, IR
----- Original Message ----
> From: "Vauthrin, Laurent" <La...@disney.com>
> To: solr-user@lucene.apache.org
> Sent: Mon, November 9, 2009 2:22:26 PM
> Subject: Solr on OOM
>
> Hello,
>
>
>
> One of our deployed Solr (1.3) setup is having out of memory issues and
> I'm not sure how to troubleshoot it. I've read a few posts (including
> http://old.nabble.com/Debugging-Solr-memory-usage-heap-problems-ts883279
> 4.html#a8832794) but I think this situation is slightly different.
>
>
>
> Here's the setup:
>
> 1 master and 1 slave are located on a the same VM (using a 64-bit JVM)
>
> 1 slave running on its own VM (using a 64-bit JVM)
>
> From what I've been told, nothing else is running on those VMs.
>
> Index size is about 100-200 MB.
>
>
>
> Solrconfig.xml cache settings:
>
>
>
>
> autowarmCount="5000"/>
>
>
> initialSize="5000" autowarmCount="5000"/>
>
>
> initialSize="5000"/>
>
>
>
> Both slaves at some point have gone out of memory (though not both at
> the same time) when receiving a moderate load of queries (a few queries
> per second - don't have an exact stat here). We started with a heap
> size of 1GB and ended up having to bump it up to 3.5GB. It seems really
> odd that we'd have to have a heap size that large when the index itself
> is not really big. Any thoughts on what could be really off here? Is
> there a way to determine the cache sizes in bytes? I noticed that there
> was a thread about other issues running Solr on VMs, has anyone else had
> problems using VMWare? From what I'm told, it seems like moving to
> physical servers won't be a fast/easy change so I'm looking for any help
> I can get for this configuration.
>
>
>
> Thanks,
> Laurent Vauthrin