Re: how scalable is openldap

[asr@ufl.edu]

=> On Fri, 4 Apr 2003 15:11:56 -0800, "Howard Chu" <hyc@highlandsun.com> said:


> A fix for this was in release 2.1.15. There was a bug in that fix, which
> itself was fixed in 2.1.16. This should no longer be a problem, as long as
> your data fits inside the sockbuf_max_incoming limits.

Thanks, I'll give this a shot.


>> You will run into locking limitations on modifications of some large
>> entries.  I haven't been able to elicit the problem on demand, but I fairly
>> regularly find that OpenLDAP stops responding to queries while it processes
>> a modify for a smaller, but still large, indexed group (60,000 members or
>> so)

> Stops? Does it just hang forever, or does it eventually come back? If it
> comes back, it's probably just overhead from writing the transaction log.
> Make sure the log is on a separate disk from the database files. Any large
> write that ties up a lot of data like this has the potential to get in
> everything else's way, but it should only be temporary.

It's only temporary, but temporary is "wait until I finish updating the
indices for all these entries", not "wait until I flush these 30MB to disk".
i.e. four or five minutes of unresponsiveness.

I don't consider that ludicrous time to delete and then recreate 60,000 index
entries, but it's unacceptable as an outage. :)

I've been working on generating this condition repeatably so I can issue a
sane problem report.


- Allen S. Rout