Approaching the limit on PV entries, consider increasing either the vm.pmap.shpgperproc or the vm.pmap.pv_entry_max sysctl.

[Evren Yurtesen]

Thomas Hurst wrote:
> * Evren Yurtesen (yurtesen at ispro.net) wrote:
> 
>> How do I see what process is sharing memory and how much memory?
> 
> Guessing is normally sufficient; typically it's processes with the same
> name and similar size/res.  On 7-STABLE you can use procstat -v to look
> at the VM mappings for a process, but typically that'll be overkill.

Thanks, I will try to check procstat -v when I start seeing the error message 
coming. When the system is showing "Approaching the limit on PV entries" is it 
related to number of allocations I see in procstat -v output? Is each line an PV 
entry? How can one obtain same information from 6.x i386? (just asking to 
compare similar systems)

>> There are a bunch of apache 2.2 processes working normally about 20-30 
>> processes. This box doesnt do much more than that...
>>
>> I just checked the machine and here is what it looks like:
>>  2:32PM  up 18 days,  5:40, 3 users, load averages: 0.41, 0.36, 0.27
>>
>> web:/root#ps ax |grep http
>> 21429  ??  Ss     0:18.08 /usr/local/sbin/httpd
>> 86473  ??  S      0:00.09 /usr/local/sbin/httpd
>> 86659  ??  S      0:00.09 /usr/local/sbin/httpd
>>
>> Although I see now that for 2 days the PV entries error did not appear. I 
>> wonder if it is spooling up somehow...
> 
> They do look a bit small to be triggering it; assuming they're sharing
> most of that, that's still only about 400k pv entries; 5MB or so (12
> bytes per entry).  The systems I've seen pv entries run out on run to a
> couple of orders of magnitude more than that.
>> There is a cron job restarting apache everyday at midnight so it cant
>> be apache leaking perhaps.
> 
> Load spikes maybe?  Child count running into the stratosphere?  Big PHP
> opcode cache?

The machine is primarily serving perl pages through mod_perl2 and there are a 
few PHP sites too but they are negligible. It is just that I never saw this kind 
of message coming out on 6.3-stable i386 with the same sites which is kind of weird.

However a load spike might be the cause of course, I will try to catch this when 
PV entry warning appear. Now that I know that procstat can show me more info, I 
can try to collect some info to see if something looks weird. I will let you 
guys know of my findings.

Thanks for the great help! This information was very useful.

Thanks,
Evren