Saslauthd constantly increasing memory use, solved by enabling caching. Why?

Dan White dwhite at olp.net
Tue Nov 1 12:27:09 EDT 2011


On 01/11/11 11:19 -0500, ktm at rice.edu wrote:
>On Tue, Nov 01, 2011 at 12:14:28PM -0400, Mark London wrote:
>> ktm at rice.edu wrote:
>> >On Tue, Nov 01, 2011 at 11:57:57AM -0400, Mark London wrote:
>> >>Hi - On RHEL 6, with the latest updates, I have SASLAUTHD configured
>> >>to use PAM authentication.  I'm also running SSSD. U sing this
>> >>configuration, the SASLAUTHD processes would gradually increase
>> >>memory usage.  After running for several days, each process was
>> >>using up about 680M.  Are there any known memory leaks when using
>> >>PAM?  I've found posts on the web from people complaining about PAM
>> >>memory leaks, but am not sure they still exists.  In any event, I'm
>> >>also experiencing that about once a week, SASLAUTHD starts recording
>> >>time out errors when trying to contact SSSD, i.e.
>> >>"pam_sss(imap:auth): Request to sssd failed. Timer expired."   I
>> >>decided to enable SASLAUTHD caching with the -c flag, and was
>> >>surprised to discover that the SASLAUTHD processes no longer use up
>> >>significant memory (i.e. they are now using < 10M)!  Can anyone
>> >>explain this behavior?  Thanks. - Mark
>> >Each trip through the PAM stack loses some memory. When you turn on
>> >caching, you make a single trip for each authentication via SASL
>> >and then it uses the cached copy from then on. This bounds your
>> >memory use to N x num-users. Without caching, the growth as you
>> >found is unbounded.
>>
>> Thanks for the info!  But without caching, does the Mailman related
>> memory use, eventually get freed up?
>
>Do not quote me, but there is a problem with the SASL spec and the
>needs of the PAM stack that cause the leak and the only way to free the
>space is to restart saslauthd.

More than likely a memory leak in saslauthd, when using the pam backend, is
caused by a bug in one of the pam modules being used. At least that's been
my experience.

-- 
Dan White


More information about the Cyrus-sasl mailing list