[SSSD] Memory leak with LDAP and frequent enumeration refreshes?

Mark London mrl at psfc.mit.edu
Fri Jul 6 02:41:09 UTC 2012


sgallagh at redhat.com wrote:
>> Here is my solution to have a persistant uptodate local cache of all ldap entries, so as to avoid very long delays when a user issues a command that causes a large number of LDAP lookups, i.e. by doing a "ls -l /home":
>>
>>  enumerate = true
>>  enum_cache_timeout = 86400 (every 24 hours)
>>  ldap_purge_cache_timeout = 0
>>  ldap_enumeration_refresh_timeout = 300
>>
>> What I would like to know, is why long delays occur when doing an "ls -l /home".  Is it because it has to write out each entry into the local database?    
> How many users are we talking about, and are they all represented in /home?
>
> The initial enumeration can take up to a few minutes the first time (after that we only update on differences). Once the full enumeration has been completed, it should take about 0.02 seconds per lookup against the local cache. On ordinary lookups, there are no writes going on. Writes to the cache only occur on cache misses.

"ls -l /home" is fast after enumeration.  But if I allow the enum cache 
to time out, then "ls -l /home" take minutes.  We have 430 directories.  
I get around this problem, by setting the cache timeout to be very 
long.  However, it appears that using this configuration, both sssd_nss 
and sssd_be are constantly growing in size.  I.e., here are the memory 
size after about 3 days of use.  The resident memory has grown from 
about 21m to over 200m.  The ldap refresh (every 5 minutes), appears to 
be constantly adding to the cache.  Or maybe there is a problem with my 
configuration that is causing this? 

10468 root      20   0  402m 236m  14m S  0.7  1.5   2:50.74 
sssd_nss          
10467 root      20   0  424m 222m 8164 S  0.0  1.4  10:20.20 sssd_be

Thanks. - Mark

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.fedorahosted.org/pipermail/sssd-devel/attachments/20120705/5c525177/attachment.html>


More information about the sssd-devel mailing list