Re: SO kills ns-ldapd because of memory usage

[Ludwig Krispenz <lkrispen@xxxxxxxxxx>]

Hi,

On 11/27/2012 09:04 AM, Moisés Barba Pérez wrote:

What do you think about change the OS overcommit_memory to 2 instead of 0???

I don't know what this would change.
But I think your graph looks like there is a memory leak, could you open a ticket a try to describe how to reproduce, it would be helpful to get the config and evtl logs, at least a description of the workload.
Does this problem occur  with the latest release ?

Regards,
Ludwig

Moses

2012/11/26 Moisés Barba Pérez <mbarperoi@xxxxxxxxx>

Hi,

   This is the end of graph I have get whit the command (while true; do ps -o 'vsz,rss' <PID>; sleep 60; done). Looks like there is not a big increase at any point. I had set swappiness to 10 and the ns-slapd lives for 2 days approx.

Any idea?

Regards, Moses.

2012/11/23 Moisés Barba Pérez <mbarperoi@xxxxxxxxx>

I'm monitoring it right now, sending it on monday.

Anyway, is there any tunning configuration for number of connections, memory, etc... that I can follow?? I have use https://access.redhat.com/knowledge/docs/en-US/Red_Hat_Directory_Server/8.2/html/Performance_Tuning_Guide/system-tuning.html and http://directory.fedoraproject.org/wiki/Performance_Tuning but I don't know if there is some specific for serveral databases, or multiple replication agreements, or very high number of searches (I have 27 database and 60 replication agreements and about 200 searches per second at rush hours)

Regards, Moses

2012/11/23 Ludwig Krispenz <lkrispen@xxxxxxxxxx>

Hi,

from the data you show, the server process should never reach 11GB, so It could be that you run into a memory leak. Could you monitor process size growth ?
Start the server, prime the caches for all backends and monitor process growth, eg running regular
ps -o 'vsz,rss' <pid>
See how fast the process grows, if it is steadily or if there is a pattern and you can relate it to some cliend load.

Regards,
Ludwig

On 11/22/2012 02:33 PM, Moisés Barba Pérez wrote:

Hi,

I have been searching for memory usage in the server. This are the results:

389-ds 1.2.5 in a CentOS 5.5 64bits 4GB ram and 6GB swap

* The ns-slapd proccess reaches 11GB of virtual memory. pmap shows multiple [anon] using the bigger part of that 11G virtual memory. I think the [anon] are memory reservation from malloc and mmap but I don't know what call this.

* Looking for cachememsize using this search for one of the database

ldapsearch -H ldaps://localhost -x -LLL -b "cn=monitor,cn=o_xxxx,cn=ldbm database,cn=plugins,cn=config" -D "cn=Directory Manager" -W "(objectclass=*)" | grep entrycache

Enter LDAP Password: 

entrycachehitratio: 99

currententrycachesize: 49973691

maxentrycachesize: 125829120

currententrycachecount: 6521

maxentrycachecount: -1

I have prime that database searching all entries with -> ldapsearch -H ldaps://localhost -x -LLL -b "o=cabu,dc=sacyl,dc=es" -D "cn=directory manager" -W "(objectclass=*)" 1.1 | grep dn: | wc -l 

The result is 7610 entries in that database, so looking the monitor again:

currententrycachesize: 59315175

maxentrycachesize: 125829120

currententrycachecount: 7611

The id2entry.db4 for that database is 59539456 so I guess I can reduce the cachememsize from 125829120 to about 60000000 Correct me if I am wrong.

And the same for all the another database.

* Now dbcachesize:

ldapsearch -H ldaps://localhost -x -LLL -b "cn=monitor, cn=ldbm database, cn=plugins,cn=config" -D "cn=Directory Manager" -W "(objectclass=*)"  | grep dbcache

Enter LDAP Password: 

dbcachehits: 1440910461

dbcachetries: 1440919648

dbcachehitratio: 99

dbcachepagein: 9187

dbcachepageout: 128041

dbcacheroevict: 9265

dbcacherwevict: 0

In some place I have read that dbcacheroevict and dbcachepageout should be 0 or increase the dbcachesize but if the ratio is 99 that should be ok, right?

The thing is, if i search with db_stat for cache statistics says ratio=99

db_stat -h /var/lib/dirsrv/slapd-xxx/db/ -m

0 Total cache size

1 Number of caches

800MB Pool individual cache size

0 Maximum memory-mapped file size

0 Maximum open file descriptors

0 Maximum sequential buffer writes

0 Sleep after writing maximum sequential buffers

0 Requested pages mapped into the process' address space

1448M Requested pages found in the cache (99%)

9588 Requested pages not found in the cache

112 Pages created in the cache

9588 Pages read into the cache

129932 Pages written from the cache to the backing file

9668 Clean pages forced from the cache

1 Dirty pages forced from the cache

0 Dirty pages written by trickle-sync thread

98066 Current total page count

98005 Current clean page count

61 Current dirty page count

131071 Number of hash buckets used for page location

1447M Total number of times hash chains searched for a page (1447895423)

5 The longest hash chain searched for a page

2819M Total number of hash buckets examined for page location (2819107178)

932 The number of hash bucket locks that required waiting (0%)

86 The maximum number of times any hash bucket lock was waited for

1 The number of region locks that required waiting (0%)

9728 The number of page allocations

60012 The number of hash buckets examined during allocations

1381 The maximum number of hash buckets examined for an allocation

9669 The number of pages examined during allocations

1 The max number of pages examined for an allocation

If I look for an index like inetuserstatus (pres and eq) I get "Requested pages found in the cache" less than 99% so I search for "inetuserstatus=*" (pres) and "inetuserstatus=active", "inetuserstatus=inactive" (eq) but the "requested pages" don't reaches the 99  or 100% and there is no more possibilities for that index.

The thing is, why ns-sldapd is growing to consume all the swap and all the ram memory the SO lets it. 

Any idea or suggestion???

2012/11/15 Ludwig Krispenz <lkrispen@xxxxxxxxxx>

you could use
ldapsearch ... -b  "cn=ldbm database,cn=plugins,cn=config" "cn=monitor" currententrycachesize

to monitor the usage of the entrycache.
But be aware that the process uses more memory than just the caches and the memory manager can also generate some overhead.

Regards,
Ludwig

On 11/15/2012 02:55 PM, Moisés Barba Pérez wrote:

yes, thats correct, but shouldn't use all that memory because don't need so much memory

2012/11/15 Ludwig Krispenz <lkrispen@xxxxxxxxxx>

Hi,

On 11/15/2012 01:54 PM, Moisés Barba Pérez wrote:

Hi,

I have a memory issue with 389-ds 1.2.5 in a CentOS 5.5 64bits 4GB ram. 

The server swaps when the server physical memory increase over 75% approx. When the swap is full the server reaches 100% of physical memory and the SO kills the ns-ldapd process.

Out of memory: Killed process 30383, UID 99, (ns-slapd).

The cache sizes are:

nsslapd-dbcachesize: 838860800

nsslapd-import-cachesize: 20000000

nsslapd-cachememsize: 125829120 (for each 26 db)

do you mean you have 26 db backends with 125MB entrycache each ? So you would reach 3.2GB for entrycache and 800MB dbcache.

Regards,
Ludwig

Which can be the problem?

--
389 users mailing list
389-users@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/389-users

--
389 users mailing list
389-users@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/389-users

--
389 users mailing list
389-users@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/389-users

--
389 users mailing list
389-users@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/389-users

--
389 users mailing list
389-users@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/389-users

--
389 users mailing list
389-users@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/389-users

--
389 users mailing list
389-users@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/389-users

--
389 users mailing list
389-users@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/389-users