Re: (ITS#6331) slapd freezing, not accepting further 'write' operations while 'read' operations are unaffected

[quanah@zimbra.com]

--On Wednesday, October 14, 2009 1:11 AM +0000 j@telepaths.org wrote:

> Full_Name: Jeff Doyle
> Version: 2.4.17
> OS: Debian/Lenny
> URL: ftp://ftp.openldap.org/incoming/
> Submission from: (NULL) (130.94.106.251)
>
>
> Greetings,
>
> Occasionally, I will find one of my servers (running Debian/Lenny with
> back-ported slapd version 2.4.17) in a state where the following occurs:
>
>   * Write operations will 'freeze' - leaving the requester at a frozen
> prompt or frozen LDAP-client of some kind (from wherever the
> write-operation originated).   * slapd refuses to 'stop', even though it
> indicates that its waiting for "0" processes to complete.
>   * Logs do not indicate a segfault nor any poorly terminated processes.
>   * Read operations on same server continue to function normally. Any
> data that existed before the "freeze" is returned as it should be.
>
> Modules used:
>
>   * accesslog
>   * back_hdb
>   * back_monitor
>   * back_relay
>   * back_rwm
>   * syncprov
>
> Servers are all fairly new, and reasonably beefy for their purposes:
>
>   * 6GB RAM
>   * 2+ GHz CPU, all QuadCore Xeon and all running amd64 kernel (current
> with Lenny)
>   * Plenty of Disk Space (gigs and gigs)
>   * Servers are SLEEPING (0.00 avg load)
>
> slapd is making use of 'tool-threads' with a value of '4' - however
> changing this value has had no effect - Just a side note.

tool threads only affects offline operations (slapadd).

How many threads do you have configured?  What version of BDB is slapd 
linked against?  Is the bdb version fully patched?  What is your checkpoint 
directive set to in both the accesslog and hdb databases?

--Quanah

--

Quanah Gibson-Mount
Principal Software Engineer
Zimbra, Inc
--------------------
Zimbra ::  the leader in open source messaging and collaboration