[Freeipa-users] IPA Load Problems?
Rich Megginson
rmeggins at redhat.com
Wed Sep 4 13:53:34 UTC 2013
On 09/04/2013 07:51 AM, Martin Kosek wrote:
> Ah, ok. One of the reasons why I was poking to this thread is exactly this
> ticket. It does not contain much information _what exactly_ is making IPA
> performance poor - whether it is missing indices (which ones?) or some issue
> in IPA plugins during binds, etc.
>
> Without more information, we do not know what to fix, what to improve.
Right. It's a big, complicated problem. So we start with what we know
- the JIRA LDAP sync with IPA is much, much too slow. We don't know why
it's too slow. But we can at least set it up and begin profiling it.
>
> Martin
>
> On 09/04/2013 02:01 PM, John Moyer wrote:
>> Martin,
>>
>> I apologize there was a large offline conversation between Rich and
>> myself. Rich was kind enough to help me through some of my issues. We
>> did a lot more tests and poking and prodding. We discovered that IPA is
>> not as efficient when dealing with large number of connections. Most of
>> my load inefficiently reconnect to IPA over and over and over and though
>> LDAP can deal with this fairly efficiently, IPA apparently drops to it's
>> knees.
>>
>> A ticket was opened to addressed this issue.
>>
>> https://fedorahosted.org/freeipa/ticket/3892
>>
>>
>> Thanks, _____________________________________________________ John Moyer
>> Director, IT Operations Digital Reasoning Systems, Inc.
>> John.Moyer at digitalreasoning.com Office: 703.678.2311 Mobile: 240.460.0023
>> Fax: 703.678.2312 www.digitalreasoning.com
>>
>> On Sep 4, 2013, at 3:44 AM, Martin Kosek <mkosek at redhat.com> wrote:
>>
>>> On 08/30/2013 11:08 PM, John Moyer wrote:
>>>> Well IPA has machine entries on some test clusters that I'm rolling
>>>> IPA out on (20 machines maybe) but the user base is the same (about 80
>>>> ~ 100) accounts with maybe 40 to 50 groups?
>>>>
>>>> I've stood up a clone of the jira server along with IPA. I cleared
>>>> my logs and then did the sync and ran the log analyzer on it. These
>>>> stats are pretty much ONLY for that jira sync I don't have any other
>>>> connections pointed to it.
>>>>
>>>>
>>>> Start of Log: 30/Aug/2013:15:57:13 End of Log:
>>>> 30/Aug/2013:16:01:14
>>>>
>>>> Processed Log Time: Hours, 4 Minutes, 1 Seconds
>>>>
>>>> Restarts: 1 Total Connections: 824 SSL
>>>> Connections: 824 Peak Concurrent Connections: 6 Total
>>>> Operations: 1806 Total Results: 1805
>>>> Overall Performance: 99.9%
>>>>
>>>> Searches: 968 (4.02/sec) (241.00/min)
>>>> Modifications: 5 (0.02/sec) (1.24/min) Adds:
>>>> 0 (0.00/sec) (0.00/min) Deletes: 0
>>>> (0.00/sec) (0.00/min) Mod RDNs: 0
>>>> (0.00/sec) (0.00/min) Compares: 0
>>>> (0.00/sec) (0.00/min) Binds: 833
>>>> (3.46/sec) (207.39/min)
>>>>
>>>> Proxied Auth Operations: 0 Persistent Searches: 1
>>>> Internal Operations: 0 Entry Operations: 0
>>>> Extended Operations: 0 Abandoned Requests: 0 Smart
>>>> Referrals Received: 0
>>>>
>>>> VLV Operations: 0 VLV Unindexed Searches: 0 SORT
>>>> Operations: 0
>>>>
>>>> Entire Search Base Queries: 0 Unindexed Searches: 1
>>>>
>>> This looks like a promising way to find out the reason, thanks John.
>>> However, I see just one unindexed search. Is the access log complete?
>>> Previously I see that the sync takes 900 seconds/15 minutes, but there
>>> is only 4 minutes the access log. Note that it it may take some time
>>> until the log is dumped.
>>>
>>> I think it would be also useful to run the analyzer with "-ula" flags as
>>> Rob suggested earlier to find out the unindexed searches (if any).
>>>
>>> What I find interesting is that JIRA does a lot of LDAP BINDs. Can the
>>> problem be in longer BINDs than with than expected (compared to for
>>> example plain LDAP servers)? Performance-wise, it would be I think
>>> better if JIRA does just one BIND and run all the LDAP searches the
>>> established connection. But I do not know if it can be configured this
>>> way.
>>>
>>> Rich, Rob, I am wondering if the slow up is not really caused by the
>>> binds, we have several DS plugins tied to the BIND operation, it may be
>>> useful to analyze if they do not take too long.
>>>
>>> Martin
>>
More information about the Freeipa-users
mailing list