>>> John Jasen <[email protected]> schrieb am 20.07.2018 um 15:41 in Nachricht
<[email protected]>:

> 
> On 07/20/2018 04:41 AM, [email protected] wrote:
>> Hi!
>>
>> Stupid question: could it be your load-balancer that had a problem?
>> How does the netstat look like (sockets opened, queued data, etc.?)
> 
> I do not believe it to be the load-balancer. They log loss of contact
> with the LDAP servers and drop them from the relay group shortly after
> one of these events start; and when it gets cleaned up, they're added
> back in. I also do not suspect network between the load balancers and
> the LDAP servers.
> 
> During such an event, ps -efT will usually show slapd running at full
> thread capacity. Comparing that to threads in cn=monitor is not
> possible, as those ldap searches fail.

Still I don't know the internals on slapd, but could it runs out of worker 
threads? DO you monitor the theads' activity up to the problem?

> 
> Open sockets does not substantially change until after the event
> subsides. The servers will show 1200-2000 open sockets before an event,
> and drop lower when it clears up -- to quickly scale back up to
> pre-event levels.

The other idea is to try to run "strace ... -p pid" on the hanging process to 
see what it is doing, or maybe even try to attach gdb to the process (most 
useful if the binary still contains debug info).

> 
> The queues will show data being held until the socket(s) time out.

OK, so it des not look like a problem in the load-balancer.

Regards,
Ulrich






Reply via email to