Milter-greylist

Ok, my 3.0rc6 memory consumption issue is still happening in 3.0 (final).

After less than 24 hours my milter-greylist had it's "size" (via top) grow to
86852. 5 minutes later it had grown another 448.

so, I tried reducing the number of DNSRBL ACL's from 18 (3 actual sites) to 2 (2
actual sites) in my config and restarted. Now my growth rate is 44 per 5 mins
instead of 448.

Next I'll try disabling DNSRBLs entirely, but this begs the question: is anyone
else seeing milter-greylist grow noticeably when using DNSRBLs?

Matt Kettler <mkettler@...> wrote:

> Next I'll try disabling DNSRBLs entirely, but this begs the question: is
> anyone else seeing milter-greylist grow noticeably when using DNSRBLs?

I did not noticed. But I would not be surprised if there was a leak
here: the code is rather new.

-- 
Emmanuel Dreyfus
http://hcpnet.free.fr/pubz
manu@...

manu@... wrote:
> Matt Kettler <mkettler@...> wrote:
> 
>> Next I'll try disabling DNSRBLs entirely, but this begs the question: is
>> anyone else seeing milter-greylist grow noticeably when using DNSRBLs?
> 
> I did not noticed. But I would not be surprised if there was a leak
> here: the code is rather new.
> 

I gave dnsrbl.c a casual eyeball-check yesterday, and I didn't see any code
paths that would cause leaks, but your right, they might be there.

I also ran it through splint, which was perfectly happy with it.

Maybe tonight I'll give it a deep eyeball-check after I get home..

Matt Kettler <mkettler@...> wrote:

> I gave dnsrbl.c a casual eyeball-check yesterday, and I didn't see any code
> paths that would cause leaks, but your right, they might be there.
> 
> I also ran it through splint, which was perfectly happy with it.
> 
> Maybe tonight I'll give it a deep eyeball-check after I get home..

Maybe a problem in a library?

-- 
Emmanuel Dreyfus
http://hcpnet.free.fr/pubz
manu@...

manu@... wrote:
> Matt Kettler <mkettler@...> wrote:
> 
>> I gave dnsrbl.c a casual eyeball-check yesterday, and I didn't see any code
>> paths that would cause leaks, but your right, they might be there.
>>
>> I also ran it through splint, which was perfectly happy with it.
>>
>> Maybe tonight I'll give it a deep eyeball-check after I get home..
> 
> Maybe a problem in a library?
> 

Maybe.. libbind has been known to have leaks before. That might be why my
release of RH doesn't include libbind by default.

FWIW, I ran overnight last night with only one DNSBL enabled. Overnight MG did
grow, but very little.