On Mon, Jun 14, 2010 at 2:25 PM, Hernan Saltiel <hsaltiel at gmail.com> wrote:
>
>
> On Mon, Jun 14, 2010 at 5:01 PM, Scott Marlowe <scott.marlowe at gmail.com>
> wrote:
>>
>> On Mon, Jun 14, 2010 at 9:56 AM, Dhaval Jaiswal
>> <bablu_postgres at yahoo.com> wrote:
>> >
>> > I am working on PostgreSQL 8.0.2. with slony I.
>> >
>> > Whenever there is a update, insert, delete happened on primary it will
>> > take
>> > some time to replicate the same on slave. We came to know about this
>> > using
>> > sl_status table, where its lagging time showing 1 hr or 2 hrs. However,
>> > sl_confirm table shows last replicated events is before 5 mins.
>> > We have also seen there is vacuum analyze running on replications
>> > schema.
>>
>> Happens to me when there's too much IO for my hardware (which is quite
>> a bit on my hardware).
>
>
> How can I meassure how much is too much use of my hardware when Slony is in
> place?
> I can meassure the CPU, memory, disk IO, and network use, but how much is
> needed in order to let Slony work well?
> Is there any way to calculate this on a transaction number and size basis?

size isn't so much important as how much disk io you're using, and
whether or not that's too much is very dependent on your hardware.  I
find that when

iostat -xd /dev/sd? 60

consistently shows 100% utilization, slony starts to fall behind.
Load testing / benchmarking your system will tell you how many
requests / second of your typical load it takes to hit that.  Once you
hit that load, the only solution is faster hardware, or reducing load,
often by optimizing your code.