On Wed, Jun 9, 2010 at 3:12 AM, Alexander V Openkin <open at immo.ru> wrote:
> 09.06.2010 11:54, Scott Marlowe пишет:
>>
>> On Wed, Jun 9, 2010 at 1:37 AM, Alexander V Openkin<open at immo.ru>  wrote:
>>
>>>
>>> Tnx for the quick answer,
>>> but if you run all slon process (for your replication cluster)
>>> you must have standalone server with 32G ?
>>> I think this is unnatural )
>>>
>>
>> No, definitely not.  VIRT is everything the process has ever touched,
>> including shared memory and all libs whether or not they've actually
>> been loaded or not.  It's not uncommon to have literally a hundred
>> processes with 8G+VIRT on my 32Gig db servers, because most of that 8G
>> is shred buffers.  It's not using that much memory individually, it's
>> using shared_memory, and each process reports that it has access to
>> and has touched that 8G.
>>
>>
>
> small experement, the second instance for ather Db was added...
>
> [root at vpsXXXX /]# /etc/init.d/slon stop
> Stopping slon service: [ DONE ]
> Stopping slon service: [ DONE ]
> [root at vpsXXXX /]# ps axuf |grep slon
> root 28419 0.0 0.0 6044 580 pts/0 S+ Jun08 0:00 \_ grep slon
> [root at vps6147 /]# free
> total used free shared buffers cached
> Mem: 10485760 99192 10386568 0 0 0
> -/+ buffers/cache: 99192 10386568
> Swap: 0 0 0
> [root at vpsXXX /]# /etc/init.d/slon start
> Starting slon service: [ DONE ]
> [root at vpsXXX /]# ps axuf |grep slon
> root 29737 0.0 0.0 6040 584 pts/0 S+ Jun08 0:00 \_ grep slon
> postgres 28555 0.0 0.0 40636 1832 pts/0 S Jun08 0:00 /usr/bin/slon
> postgres 28556 0.0 0.0 4042880 1508 pts/0 Sl Jun08 0:00 \_ /usr/bin/slon
> postgres 28592 0.0 0.0 40636 1836 pts/0 S Jun08 0:00 /usr/bin/slon
> postgres 28594 0.0 0.0 4042880 1512 pts/0 Sl Jun08 0:00 \_ /usr/bin/slon
> [root at vpsXXXX /]# free
> total used free shared buffers cached
> Mem: 10485760 8119136 2366624 0 0 0
> -/+ buffers/cache: 8119136 2366624
> Swap: 0 0 0
> [root at vpsXXXX /]#
>
> The different between "no slon process" and "2 pair slon process" ~8G, that
> we have a problem,
> because it`s not a shared segment.....

Oh whoa, I thought you were talking about the postgres backend that
slony connects to using up that much memory.

I wonder if there's some accounting difference in how your vps works
versus running right on the server.

> Besides OpenVZ divide shared memory and resident memory
>
> [root at vps6147 /]# cat /proc/user_beancounters |grep -E
> 'privvmpages|shmpages'
> privvmpages 2029830 2033439 2621440 2621440 5
> shmpages 17632 17632 412000 412000 0
> [root at vps6147 /]#
>
> first column - the current value in 4k pages, it`s indicates very small
> shared segment and huge resident segment,

Yeah, that's different from what I was thinking was going on.

> Do you have a expirience using slony1 on x86_64 servers ?

Quite a bit actually.

> We using slony1 replication about 3 year on i686 architecture and we hav`t

Is that a "have" or "haven't" ?

> similar problem....
>
> PS we using the same OpenVZ template for application servers, and
> probability error in template or in the current VPS is minimum.

I've never run dbs inside vms before (seems counter productive to me)