Wednesday, August 11, 2021

[389-users] Re: Lock table is out of available lock entries


On 8/11/21 9:09 AM, Pierre Rogier wrote:
Hi,
I suspect that Kees discovered a new bug ...

As the retro changelog is persistent across reboot, it is normal that their timestamp are absolute, so IMHO there is no way relative time should be used/seen in trimming computation.
(in other words cur_time is wrong. )
My suspicion is that we got a regression when the slapi_eq timers where  rewritten to use relative time instead of absolute time) (cur_time is provided by slapi_eq API and is now relative ( it should rather be initialized to time(0) )
Looking at the code we are mixing the use of the same monotonic time for the interval (correct) and the age of an entry (not correct).
The funny thing is that nobody have seen that retrocl trimming was broken for so long ...

That's because it silently just builds up and no one notices :-(

I opened a ticket for this:  https://github.com/389ds/389-ds-base/issues/4869

Mark

Regards Pierre,

On Wed, Aug 11, 2021 at 11:50 AM Kees Bakker <keesb@ghs.com> wrote:
On 09-08-2021 20:23, Mark Reynolds wrote:
> On 8/9/21 1:47 PM, Kees Bakker wrote:
>> On 09-08-2021 19:25, Mark Reynolds wrote:
>>> On 8/9/21 1:16 PM, Kees Bakker wrote:
>>>> On 09-08-2021 18:43, Mark Reynolds wrote:
>>>>> On 8/9/21 11:20 AM, Kees Bakker wrote:
>>>>>> On 09-08-2021 16:00, Mark Reynolds wrote:
>>>>>>> On 8/9/21 8:09 AM, Kees Bakker wrote:
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> When my dirsrv was trying to compact the databases I was getting
>>>>>>>> this
>>>>>>>> error
>>>>>>>>
>>>>>>>> [07/Aug/2021:23:59:02.715984489 +0200] - NOTICE - bdb_compact -
>>>>>>>> Compacting databases ...
>>>>>>>> [07/Aug/2021:23:59:02.765932397 +0200] - NOTICE - bdb_compact -
>>>>>>>> Compacting DB start: userRoot
>>>>>>>> [07/Aug/2021:23:59:03.518175414 +0200] - NOTICE - bdb_compact -
>>>>>>>> compactdb: compact userRoot - 417 pages freed
>>>>>>>> [07/Aug/2021:23:59:03.576427786 +0200] - NOTICE - bdb_compact -
>>>>>>>> Compacting DB start: ipaca
>>>>>>>> [07/Aug/2021:23:59:03.659941533 +0200] - NOTICE - bdb_compact -
>>>>>>>> compactdb: compact ipaca - 419 pages freed
>>>>>>>> [07/Aug/2021:23:59:03.718445310 +0200] - NOTICE - bdb_compact -
>>>>>>>> Compacting DB start: changelog
>>>>>>>> [08/Aug/2021:00:00:40.807571334 +0200] - NOTICE -
>>>>>>>> NSMMReplicationPlugin - changelog program - cl5CompactDBs -
>>>>>>>> compacting
>>>>>>>> replication changelogs...
>>>>>>>> [08/Aug/2021:00:00:54.309357211 +0200] - ERR - libdb - BDB2055 Lock
>>>>>>>> table is out of available lock entries
>>>>>>>> [08/Aug/2021:00:00:54.726504736 +0200] - ERR - bdb_compact -
>>>>>>>> compactdb: failed to compact changelog; db error - 12 Cannot
>>>>>>>> allocate
>>>>>>>> memory
>>>>>>>> [08/Aug/2021:00:00:54.801571421 +0200] - ERR - libdb - BDB2055 Lock
>>>>>>>> table is out of available lock entries
>>>>>>>> [08/Aug/2021:00:00:54.876618702 +0200] - ERR -
>>>>>>>> NSMMReplicationPlugin -
>>>>>>>> changelog program - cl5CompactDBs - Failed to compact
>>>>>>>> a797bb0b-be1d11eb-88c0b677-613aa2ad; db error - 12 Cannot allocate
>>>>>>>> memory
>>>>>>>> [08/Aug/2021:00:00:57.253006449 +0200] - NOTICE - bdb_compact -
>>>>>>>> Compacting databases finished.
>>>>>>>>
>>>>>>>> There are about 402k entries in cn=changelog.
>>>>>>>>
>>>>>>>> I have a few questions
>>>>>>>> 1) is it normal to have so many entries in cn=changelog? On another
>>>>>>>> replica I have almost 3M entries Isn't this cleaned up?
>>>>>>>> 2) the number of locks is 50000 (there are two config items).
>>>>>>>> Should I
>>>>>>>> increase that number? If so, increase to what?
>>>>>>>> 3) is there maybe something else going on, causing the
>>>>>>>> exhaustion of
>>>>>>>> the locks?
>>>>>>>
>>>>>>> Ok, so by default there is no changelog trimming enabled. So the
>>>>>>> changelog will grow without bounds, which is bad.
>>>>>>
>>>>>> How much of this [1] applies? Indeed it says "By default the
>>>>>> Replication Changelog does not use any trimming by default."
>>>>>> So, how is the trimming actually enabled? I have these entries
>>>>>>
>>>>>> dn: cn=changelog5,cn=config
>>>>>> cn: changelog5
>>>>>> nsslapd-changelogdir: /var/lib/dirsrv/slapd-EXAMPLE-COM/cldb
>>>>>> nsslapd-changelogmaxage: 30d
>>>>>> objectClass: top
>>>>>> objectClass: extensibleobject
>>>>>
>>>>> So trimming is set, but it's set to 30 days (30d), that
>>>>> could/should be
>>>>> shortened to "7d".
>>>>
>>>> Somehow I doubt that this will do anything.
>>>
>>> Ahhh, I thought this was the Replication Changelog, but it is the
>>> "Retro" Changelog.  Ok so check this out:
>>>
>>> https://access.redhat.com/documentation/en-us/red_hat_directory_server/11/html/administration_guide/managing_replication-using_the_retro_changelog_plug_in#Using_the_Retro_Changelog_Plug_in-Trimming_the_Retro_Changelog
>>>
>>>
>>> Or update config directly using ldapmodify:
>>>
>>> # ldapmodify -D ...
>>> dn: cn=Retro Changelog Plugin,cn=plugins,cn=config
>>> changetype: modify
>>> replace: nsslapd-changelogmaxage: 7d
>>> nsslapd-changelogmaxage: 7d
>>
>> (( Typo? The replace line is without the value ))
> yes, sorry...
>>
>> Hmm, that maxage was already at 2 days
>
> Ok, perhaps you are running into a known bug where indexes created by
> plugins are not enabled.  Or might be a config issue?
>
> Try this first.  Set in the retro changelog entry:
>
> nsslapd-changelog-trim-interval: 300   <--- 5 minutes
(Sorry to be picky, but it is nsslapd-changelogtrim-interval. It's default is 300.)
>
> Restart the server to make sure it was applied.  You can enable "plugin"
> logging to see what it does.
>
> nsslapd-errorlog-level: 65536
>
> wait over 5 minutes, then grep the errors log for "DSRetroclPlugin".
> This will tell us what the plugin is doing when it goes to trim the
> database.
>
> IMPORTANT, disable plugin logging immediately after you get your data.
> Set the errorlog level to 0 to go back to default logging.
>
> If the logging shows it's not finding any candidates then it might be
> the indexing issue I mentioned above.  In that case you just need to
> reindex the retro changelog:

Before changing anything I enabled Plug-in debugging which revealed something interesting

[root@linge ~]# grep trim /var/log/dirsrv/slapd-GHS-NL/errors-plugin-debug-20210811-1108
[11/Aug/2021:11:01:49.792450548 +0200] - DEBUG - DSRetroclPlugin - cltrim: ldrc=0, first_time=1622032141, cur_time=3702600
[11/Aug/2021:11:01:49.842339467 +0200] - DEBUG - DSRetroclPlugin - retrocl_housekeeping - changelog does not need to be trimmed
[11/Aug/2021:11:06:49.953401407 +0200] - DEBUG - DSRetroclPlugin - cltrim: ldrc=0, first_time=1622032141, cur_time=3702900
[11/Aug/2021:11:06:50.012249703 +0200] - DEBUG - DSRetroclPlugin - retrocl_housekeeping - changelog does not need to be trimmed

This shows that the interval is 300 (even if nsslapd-changelogtrim-interval is not explicitly set in cn=changelog5,cn=config

Look at cur_time. That is the "uptime" of the system, not UTC.

[root@linge ~]# uptime
  11:21:42 up 42 days, 20:49,  2 users,  load average: 0.15, 0.14, 0.16

Next, assuming this code from retrocl_trim.c is making the decision to trim

             first_time = retrocl_getchangetime(SLAPI_SEQ_FIRST, &ldrc);
             slapi_log_err(SLAPI_LOG_PLUGIN, RETROCL_PLUGIN_NAME,
                           "cltrim: ldrc=%d, first_time=%ld, cur_time=%ld\n",
                           ldrc, first_time, cur_time);
             if (LDAP_SUCCESS == ldrc && first_time > (time_t)0L &&
                 first_time + ts.ts_c_max_age < cur_time) {
                 must_trim = 1;
             }

There is a mixup of absolute (first_time) and relative (cur_time) time. Which of the two is wrong?

>
> # dsctl slapd-YOUR_INSTANCE stop
>
> # dsctl slapd-YOUR_INSTANCE db2index changelog
>
> # dsctl slapd-YOUR_INSTANCE start
>
> Then wait 5 minutes and see if anything was trimmed.
>
> Mark
>
>>
>> dn: cn=Retro Changelog Plugin,cn=plugins,cn=config
>> cn: Retro Changelog Plugin
>> nsslapd-attribute: nsuniqueid:targetUniqueId
>> nsslapd-changelogmaxage: 2d
>> nsslapd-include-suffix: cn=dns,dc=example,dc=com
>> nsslapd-plugin-depends-on-named: Class of Service
>> nsslapd-plugin-depends-on-type: database
>> nsslapd-pluginDescription: Retrocl Plugin
>> nsslapd-pluginEnabled: on
>> nsslapd-pluginId: retrocl
>> nsslapd-pluginInitfunc: retrocl_plugin_init
>> nsslapd-pluginPath: libretrocl-plugin
>> nsslapd-pluginType: object
>> nsslapd-pluginVendor: 389 Project
>> nsslapd-pluginVersion: 1.4.3.231
>> nsslapd-pluginbetxn: on
>> nsslapd-pluginprecedence: 25
>> objectClass: top
>> objectClass: nsSlapdPlugin
>> objectClass: extensibleObject
>>
>>>
>>>
>>> HTH,
>>>
>>> Mark
>>>
>>>>
>>>> The first entry in cn=changelog is more than two months old. (May 26th
>>>> is about
>>>> the time I installed this server as a FreeIPA replica.)
>>>>
>>>> dn: changenumber=1,cn=changelog
>>>> objectClass: top
>>>> objectClass: changelogentry
>>>> objectClass: extensibleObject
>>>> targetuniqueid: 1e89ebcb-e20111e8-8820f96e-fc0c60a4
>>>> changeNumber: 1
>>>> targetDn:
>>>> idnsname=_ldap._tcp,idnsname=example.com.,cn=dns,dc=example,dc=com
>>>> changeTime: 20210526122901Z
>>>> changeType: modify
>>>> changes:: YWRkOiBzUlZSZWNvc...gplbnRyeXVzbjogMTEyCi0KAA==
>>>>
>>>> Is there a way to see trimming in action? I'm afraid it is not doing
>>>> anything.
>>
>> What diagnostic can I enable to see the trimming in action?
>>
>>>>
>>>>>
>>>>>>
>>>>>>
>>>>>>>
>>>>>>> I recommend setting up the changelog max age to 7 days:
>>>>>>>
>>>>>>> https://access.redhat.com/documentation/en-us/red_hat_directory_server/11/html/administration_guide/trimming_the_replication_changelog
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> Once that the trimming cleans up the changelog the database
>>>>>>> compaction
>>>>>>> should work without issue.  You can also increase the database
>>>>>>> locks to
>>>>>>> 1 million (that does require a restart of the server to take
>>>>>>> effect).
>>>>>>
>>>>>> Let's hope that 1 million is enough :-)
>>>>>
>>>>> It might not be, you can keep bumping it up if needed. There is no
>>>>> limit, or negative impact on db performance.
>>>>>
>>>>> Mark
>>>>>
>>>>>>
>>>>>>>
>>>>>>> HTH,
>>>>>>>
>>>>>>> Mark
>>>>>>>
>>>>>> [1]
>>>>>> https://directory.fedoraproject.org/docs/389ds/FAQ/changelog-trimming.html
>>>>>>
>>>>>>
>>>>>
>>>>> --
>>>>> Directory Server Development Team
>>>>>
>>>> _______________________________________________
>>>> 389-users mailing list -- 389-users@lists.fedoraproject.org
>>>> To unsubscribe send an email to 389-users-leave@lists.fedoraproject.org
>>>> Fedora Code of Conduct:
>>>> https://docs.fedoraproject.org/en-US/project/code-of-conduct/
>>>> List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
>>>> List Archives:
>>>> https://lists.fedoraproject.org/archives/list/389-users@lists.fedoraproject.org
>>>>
>>>> Do not reply to spam on the list, report it:
>>>> https://pagure.io/fedora-infrastructure
>>>
>>> --
>>> Directory Server Development Team
>>>
>>
> --
> Directory Server Development Team
>
_______________________________________________
389-users mailing list -- 389-users@lists.fedoraproject.org
To unsubscribe send an email to 389-users-leave@lists.fedoraproject.org
Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: https://lists.fedoraproject.org/archives/list/389-users@lists.fedoraproject.org
Do not reply to spam on the list, report it: https://pagure.io/fedora-infrastructure


--
--

389 Directory Server Development Team
--   Directory Server Development Team

No comments:

Post a Comment