Rich Megginson wrote:
> Tamas Bagyal wrote:
>> Rich Megginson wrote:
>>> Bagyal Tamas wrote:
>>>> Rich Megginson wrote:
>>>>> Tamas Bagyal wrote:
>>>>>> hello Ryan,
>>>>>>
>>>>>> you tried this version? i have two fedora-ds 1.0.4 in mmr
>>>>>> configuration. i migrate one of those to 1.1 (builded by your and
>>>>>> Rich's instrutctions). but i have a problem with memory usage
of
>>>>>> ns-slapd process. initially mem usage is 18.5% but after 2 hours
>>>>>> this changed to 23.1% and growed until killed by kernel. (i
>>>>>> think...)
>>>>>>
>>>>>> mostly read transactions happen (dns) with a few write (cups).
>>>>>> this is a debian etch, mem size is 512 mbyte (i know this is too
>>>>>> low, but this is a test environment). cache size of slapd is
>>>>>> 67108864.
>>>>> Are you using SSL? Anything interesting in your server error log?
>>>>
>>>> I running the setupssl2.sh but not use any ssl connection. error
>>>> log shows nothing, only the server start.
>>> The reason I ask is that older versions of the NSS crypto/SSL
>>> libraries had a memory leak. NSS 3.11.7 does not have this
>>> problem. But you would only see the problem if you were using SSL
>>> connections.
>>
>> ok. I tried again from begining. fresh install, no ssl, no migration,
>> used the setup-ds-admi.pl and setup the mmr with a fedora-ds 1.0.4.
>> but nothing changed, memory usage growing...
>> All setting is default except the mmr/changelog and access.log is off.
>>
>> errors:
>>
>> Fedora-Directory/1.1.0 B2008.059.1017
>> tower.fmintra.hu:389 (/opt/dirsrv/etc/dirsrv/slapd-tower)
>>
>>
>> [05/Mar/2008:10:19:20 +0100] - dblayer_instance_start: pagesize:
>> 4096, pages: 128798, procpages: 5983
>> [05/Mar/2008:10:19:20 +0100] - cache autosizing: import cache: 204800k
>> [05/Mar/2008:10:19:21 +0100] - li_import_cache_autosize: 50,
>> import_pages: 51200, pagesize: 4096
>> [05/Mar/2008:10:19:21 +0100] - WARNING: Import is running with
>> nsslapd-db-private-import-mem on; No other process is allowed to
>> access the database
>> [05/Mar/2008:10:19:21 +0100] - dblayer_instance_start: pagesize:
>> 4096, pages: 128798, procpages: 5983
>> [05/Mar/2008:10:19:21 +0100] - cache autosizing: import cache: 204800k
>> [05/Mar/2008:10:19:21 +0100] - li_import_cache_autosize: 50,
>> import_pages: 51200, pagesize: 4096
>> [05/Mar/2008:10:19:21 +0100] - import userRoot: Beginning import job...
>> [05/Mar/2008:10:19:21 +0100] - import userRoot: Index buffering
>> enabled with bucket size 100
>> [05/Mar/2008:10:19:21 +0100] - import userRoot: Processing file
>> "/tmp/ldifZHth0D.ldif"
>> [05/Mar/2008:10:19:21 +0100] - import userRoot: Finished scanning
>> file "/tmp/ldifZHth0D.ldif" (9 entries)
>> [05/Mar/2008:10:19:21 +0100] - import userRoot: Workers finished;
>> cleaning up...
>> [05/Mar/2008:10:19:21 +0100] - import userRoot: Workers cleaned up.
>> [05/Mar/2008:10:19:21 +0100] - import userRoot: Cleaning up producer
>> thread...
>> [05/Mar/2008:10:19:21 +0100] - import userRoot: Indexing complete.
>> Post-processing...
>> [05/Mar/2008:10:19:21 +0100] - import userRoot: Flushing caches...
>> [05/Mar/2008:10:19:21 +0100] - import userRoot: Closing files...
>> [05/Mar/2008:10:19:21 +0100] - All database threads now stopped
>> [05/Mar/2008:10:19:21 +0100] - import userRoot: Import complete.
>> Processed 9 entries in 0 seconds. (inf entries/sec)
>> [05/Mar/2008:10:19:22 +0100] - Fedora-Directory/1.1.0 B2008.059.1017
>> starting up
>> [05/Mar/2008:10:19:22 +0100] - I'm resizing my cache now...cache was
>> 209715200 and is now 8000000
>> [05/Mar/2008:10:19:22 +0100] - slapd started. Listening on All
>> Interfaces port 389 for LDAP requests
>> [05/Mar/2008:10:22:23 +0100] NSMMReplicationPlugin - changelog
>> program - cl5Open: failed to open changelog
>> [05/Mar/2008:10:22:24 +0100] NSMMReplicationPlugin - changelog
>> program - changelog5_config_add: failed to start changelog
>> [05/Mar/2008:10:26:49 +0100] NSMMReplicationPlugin - agmt="cn=replica
>> to backup" (backup:389): Replica has a different generation ID than
>> the local data.
>> [05/Mar/2008:10:32:00 +0100] NSMMReplicationPlugin -
>> repl_set_mtn_referrals: could not set referrals for replica
>> dc=fmintra,dc=hu: 32
>> [05/Mar/2008:10:32:00 +0100] NSMMReplicationPlugin -
>> multimaster_be_state_change: replica dc=fmintra,dc=hu is going
>> offline; disabling replication
>> [05/Mar/2008:10:32:00 +0100] - WARNING: Import is running with
>> nsslapd-db-private-import-mem on; No other process is allowed to
>> access the database
>> [05/Mar/2008:10:32:13 +0100] - import userRoot: Workers finished;
>> cleaning up...
>> [05/Mar/2008:10:32:13 +0100] - import userRoot: Workers cleaned up.
>> [05/Mar/2008:10:32:13 +0100] - import userRoot: Indexing complete.
>> Post-processing...
>> [05/Mar/2008:10:32:13 +0100] - import userRoot: Flushing caches...
>> [05/Mar/2008:10:32:13 +0100] - import userRoot: Closing files...
>> [05/Mar/2008:10:32:14 +0100] - import userRoot: Import complete.
>> Processed 12242 entries in 13 seconds. (941.69 entries/sec)
>> [05/Mar/2008:10:32:14 +0100] NSMMReplicationPlugin -
>> multimaster_be_state_change: replica dc=fmintra,dc=hu is coming
>> online; enabling replication
>>
>> memory usage by top:
>>
>> top - 10:58:21 up 25 days, 22:36, 2 users, load average: 0.01,
>> 0.13, 0.22
>> Tasks: 61 total, 2 running, 59 sleeping, 0 stopped, 0 zombie
>> Cpu(s): 0.0%us, 0.0%sy, 0.0%ni,100.0%id, 0.0%wa, 0.0%hi,
>> 0.0%si, 0.0%st
>> Mem: 515192k total, 189600k used, 325592k free, 36472k buffers
>> Swap: 489848k total, 18292k used, 471556k free, 106188k cached
>>
>> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
>> 27647 fds 15 0 464m 47m 25m S 0.0 9.4 1:34.57 ns-slapd
>>
>>
>> top - 11:23:12 up 25 days, 23:01, 2 users, load average: 0.36,
>> 0.27, 0.20
>> Tasks: 61 total, 2 running, 59 sleeping, 0 stopped, 0 zombie
>> Cpu(s): 3.0%us, 0.0%sy, 0.0%ni, 96.0%id, 1.0%wa, 0.0%hi,
>> 0.0%si, 0.0%st
>> Mem: 515192k total, 210700k used, 304492k free, 36488k buffers
>> Swap: 489848k total, 18288k used, 471560k free, 117204k cached
>>
>> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
>> 27647 fds 15 0 473m 59m 28m S 3.0 11.9 2:52.77 ns-slapd
>>
>>
>> top - 11:48:26 up 25 days, 23:26, 2 users, load average: 0.02,
>> 0.08, 0.10
>> Tasks: 61 total, 1 running, 60 sleeping, 0 stopped, 0 zombie
>> Cpu(s): 3.0%us, 0.0%sy, 0.0%ni, 97.0%id, 0.0%wa, 0.0%hi,
>> 0.0%si, 0.0%st
>> Mem: 515192k total, 222756k used, 292436k free, 36520k buffers
>> Swap: 489848k total, 18288k used, 471560k free, 118932k cached
>>
>> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
>> 27647 fds 15 0 483m 72m 30m S 0.0 14.4 4:12.04 ns-slapd
>>
>>
>> top - 13:31:42 up 26 days, 1:09, 2 users, load average: 0.28,
>> 0.17, 0.15
>> Tasks: 61 total, 2 running, 59 sleeping, 0 stopped, 0 zombie
>> Cpu(s): 1.1%us, 0.0%sy, 0.0%ni, 98.9%id, 0.0%wa, 0.0%hi,
>> 0.0%si, 0.0%st
>> Mem: 515192k total, 285572k used, 229620k free, 36540k buffers
>> Swap: 489848k total, 18288k used, 471560k free, 140412k cached
>>
>> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
>> 27647 fds 15 0 523m 116m 34m S 0.0 23.3 9:35.65 ns-slapd
> Can you post your dse.ldif to pastebin.com? Be sure to omit or
> obscure any sensitive data first. I'd like to see what all of your
> cache settings are. Normally the server will increase in memory usage
> until the caches are full, then memory usage should level off. The
> speed at which this occurs depends on usage.
>
http://www.pastebin.org/22477
i forget a thing. i use some custom schema (ldapdns, ibm... etc.) if
this is changed anything. (but i think this is not relevant info)
> When the kernel kills your server, how much memory is it using? Is
> there anything in the server error log at around the time the kernel
> kills it?
>
i'm not sure, but at the time use the maximum as possible (512ram + 512
swap available) i think around 940mb, the kernel first kill some other
processes, like mc, and after these the ns-slapd. I can't see anything
in the log file, just the server start.
> Finally, if you are convinced that there is a real memory leak in the
> server, would it be possible for you to run it under valgrind? Just
> running it under valgrind for 30 minutes or so should reveal any
> memory leaks in normal usage.
http://www.pastebin.org/22484
I can't understand this output, I never used valgrind before. I hope
used the right options for valgrind.