Re: Monitoring number of backends

On 10/22/2013 3:44 PM, Tom Lane wrote:
> andy <andy(at)squeakycode(dot)net> writes:
>> On 10/22/2013 2:18 PM, John R Pierce wrote:
>>> that style of php programming, you're getting some HUGE overhead in
>>> connect/disconnect per web page. putting pg_bouncer in the middle
>>> will make a HUGE improvement, possibly a second per page load on a busy
>>> server.
>
>> No, actually, I don't think my connect overhead is huge. My apache and
>> postgres are on the same box, and it connects using unix socket.
>
> You're ignoring the fact that PG backends have a pretty considerable
> startup transient. By the time a backend has gotten its caches populated
> enough to be efficient, it's expended a lot of cycles. You might be
> getting away with this approach under low load, but it will bite you in
> painful places eventually.
>
> regards, tom lane
>
>

> but it will bite you in
> painful places eventually.

:-) heh.

Well I think PG is even more impressive now. My server is on a VM, and
I'm pretty much doing things the slow way, and I get a page back in
500ms. And this is a busy time of day.

Of course, I'm right next to the server. Anyone wanna check page times
for me?

http://jasper.iowaassessors.com/parcel.php?gid=99680

I'm talking JUST parcel.php ... the maps and photos don't count.

Thanks all.

-Andy