Memory management

Elastic Stack Elasticsearch
1

Hi,

Heap-Memory in my ES nodes still increases while filter and field cache is
bounded. Filter cache is max 800MB, field cache is max 1.4GB (I check it in
bigdesk plugin). Heap-Memory grows up to 7.9GB (max) after few hours and I
get OutOfMemory exceptions (because field cache is resident). What else can
take memory in ES and how to control it ?

Thank you.
Best regards.
Marcin Dojwa.

--

2

Hi,

That may help ElasticSearch Cache Usage - Sematext

Best,

Ferhat
www.searchbox.io

On Wednesday, December 12, 2012 8:56:16 PM UTC+2, Marcin Dojwa wrote:

Hi,

Heap-Memory in my ES nodes still increases while filter and field cache is
bounded. Filter cache is max 800MB, field cache is max 1.4GB (I check it in
bigdesk plugin). Heap-Memory grows up to 7.9GB (max) after few hours and I
get OutOfMemory exceptions (because field cache is resident). What else can
take memory in ES and how to control it ?

Thank you.
Best regards.
Marcin Dojwa.

--

3

Hi, thanks, unfortunatelly this does not help :frowning: all the caches are bounded
and monitored. I think the problem is in GC that frees memory too late and
field cache does not have enough memory to allocate. I will experiment with
G1, maybe this would help combaining with soft field cache.

Best regards

W dniu czwartek, 13 grudnia 2012 użytkownik ferhatsb napisał:

Hi,

That may help
ElasticSearch Cache Usage - Sematext

Best,

Ferhat
www.searchbox.io

On Wednesday, December 12, 2012 8:56:16 PM UTC+2, Marcin Dojwa wrote:

Hi,

Heap-Memory in my ES nodes still increases while filter and field cache
is bounded. Filter cache is max 800MB, field cache is max 1.4GB (I check it
in bigdesk plugin). Heap-Memory grows up to 7.9GB (max) after few hours and
I get OutOfMemory exceptions (because field cache is resident). What else
can take memory in ES and how to control it ?

Thank you.
Best regards.
Marcin Dojwa.

--

--

4

Hi,

Last time I tried G1 while helping a client with performance was maybe a
year ago. It did worse than what we were able to achieve with other
garbage collectors and various JVM params to control when GC kicks in and
such. If memory serves me well, SurvivorRatio helped us a lot.

Have a look
at http://jprante.github.com/2012/11/28/Elasticsearch-Java-Virtual-Machine-settings-explained.html

Otis

ELASTICSEARCH Performance Monitoring - Sematext Monitoring | Infrastructure Monitoring Service
Search Analytics - Cloud Monitoring Tools & Services | Sematext

On Friday, December 14, 2012 1:23:55 AM UTC-5, Marcin Dojwa wrote:

Hi, thanks, unfortunatelly this does not help :frowning: all the caches are
bounded and monitored. I think the problem is in GC that frees memory too
late and field cache does not have enough memory to allocate. I will
experiment with G1, maybe this would help combaining with soft field cache.

Best regards

W dniu czwartek, 13 grudnia 2012 użytkownik ferhatsb napisał:

Hi,

That may help
ElasticSearch Cache Usage - Sematext

Best,

Ferhat
www.searchbox.io

On Wednesday, December 12, 2012 8:56:16 PM UTC+2, Marcin Dojwa wrote:

Hi,

Heap-Memory in my ES nodes still increases while filter and field cache
is bounded. Filter cache is max 800MB, field cache is max 1.4GB (I check it
in bigdesk plugin). Heap-Memory grows up to 7.9GB (max) after few hours and
I get OutOfMemory exceptions (because field cache is resident). What else
can take memory in ES and how to control it ?

Thank you.
Best regards.
Marcin Dojwa.

--

--

5

Hi,

Thank you. I set the following parameters:
-XX:SurvivorRatio=3
-XX:+UseConcMarkSweepGC
-XX:+UseParNewGC
-XX:+CMSIncrementalMode

and it seems it works fine now. I continue testing to check if it works, if
not I will continue playing with GC settings :slight_smile:

Best regards.
Marcin Dojwa

2012/12/15 Otis Gospodnetic otis.gospodnetic@gmail.com

Hi,

Last time I tried G1 while helping a client with performance was maybe a
year ago. It did worse than what we were able to achieve with other
garbage collectors and various JVM params to control when GC kicks in and
such. If memory serves me well, SurvivorRatio helped us a lot.

Have a look at
http://jprante.github.com/2012/11/28/Elasticsearch-Java-Virtual-Machine-settings-explained.html

Otis

ELASTICSEARCH Performance Monitoring - Sematext Monitoring | Infrastructure Monitoring Service
Search Analytics - Cloud Monitoring Tools & Services | Sematext

On Friday, December 14, 2012 1:23:55 AM UTC-5, Marcin Dojwa wrote:

Hi, thanks, unfortunatelly this does not help :frowning: all the caches are
bounded and monitored. I think the problem is in GC that frees memory too
late and field cache does not have enough memory to allocate. I will
experiment with G1, maybe this would help combaining with soft field cache.

Best regards

W dniu czwartek, 13 grudnia 2012 użytkownik ferhatsb napisał:

Hi,

That may help http://blog.sematext.com/2012/**05/17/elasticsearch-cache-
**usage/http://blog.sematext.com/2012/05/17/elasticsearch-cache-usage/

Best,

Ferhat
www.searchbox.io

On Wednesday, December 12, 2012 8:56:16 PM UTC+2, Marcin Dojwa wrote:

Hi,

Heap-Memory in my ES nodes still increases while filter and field cache
is bounded. Filter cache is max 800MB, field cache is max 1.4GB (I check it
in bigdesk plugin). Heap-Memory grows up to 7.9GB (max) after few hours and
I get OutOfMemory exceptions (because field cache is resident). What else
can take memory in ES and how to control it ?

Thank you.
Best regards.
Marcin Dojwa.

--

--

--

6

I've just read that CMSIncrementalMode is good for low processors number. I
removed that and check how it works.

2012/12/15 Marcin Dojwa m.dojwa@livechatinc.com

CMSIncrementalMode

--

7

From
here Elasticsearch Platform — Find real-time answers at scale | Elastic

index.cache.field.max_sizeThe max size (count, not byte size) of the cache
(per search segment in a shard). Defaults to not set (-1).It's a count
which is the number of documents per segment and not total size in bytes.
Hope that helps.

On Saturday, December 15, 2012 2:06:31 AM UTC-8, Marcin Dojwa wrote:

I've just read that CMSIncrementalMode is good for low processors number.
I removed that and check how it works.

2012/12/15 Marcin Dojwa <m.d...@livechatinc.com <javascript:>>

CMSIncrementalMode

--