[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: slow shutdowns



On 1/25/07, Donn Cave <donn@u.washington.edu> wrote:

On Jan 25, 2007, at 6:49 AM, Aaron Richton wrote:

>
> On Thu, 25 Jan 2007, Howard Chu wrote:
>
>> matthew sporleder wrote:
>>> I'm running a pretty large slapd 2.3.23 (3G-5G or so) and
>>> shutdown is
>>> taking anywhere from twenty minutes to an hour.  (I'm still waiting
>>> for a shutdown and hoping for an hour)
>>> I was just wondering what exactly was happening.  I suspect that
>>> it's
>>> something in bdb_db_close(), but I'm not sure what's really going on
>>> since truss is just sitting around and prstat reports the pid in
>>> sleep.
>>> My guess is that it's walking the entire cache and free-ing it,
>>> but I
>>> would expect the memory to change or the truss to return free's in
>>> that case.  Any ideas?  My last log messages are:
>>> daemon: shutdown requested and initiated.
>>> slapd shutdown: waiting for 0 threads to terminate
>>
>> Attach to it with gdb and get a stack trace.
>
> Always good advice, but given that heavy disk activity, do you have
> an appropriate checkpoint directive set?

Symptoms would be 1) problem occurs only after some period of real
activity, not when stopped immediately after start, and 2) large
fraction of time in I/O wait state (cf. top, vmstat.)  I don't know
for sure what an appropriate checkpoint would be, but I know what
happens without one.


My checkpoint on all three subordinate databases is:
checkpoint      5120 15