Re: Built-in connection pooler

Hi Konstantin,

Thanks for your work on this. I'll try to do more testing in the next few
days, here's what I have so far.

make installcheck-world: passed

The v8 patch [1] applies, though I get indent and whitespace errors:

<stdin>:79: tab in indent.
"Each proxy launches its own subset of backends. So
maximal number of non-tainted backends is "
<stdin>:80: tab in indent.
"session_pool_size*connection_proxies*databases*roles.
<stdin>:519: indent with spaces.
char buf[CMSG_SPACE(sizeof(sock))];
<stdin>:520: indent with spaces.
memset(buf, '\0', sizeof(buf));
<stdin>:522: indent with spaces.
/* On Mac OS X, the struct iovec is needed, even if it points to
minimal data */
warning: squelched 82 whitespace errors
warning: 87 lines add whitespace errors.

In connpool.sgml:

"but it can be changed to standard Postgres 4321"

Should be 5432?

" As far as pooled backends are not terminated on client exist, it will not
be possible to drop database to which them are connected."

Active discussion in [2] might change that, it is also in this July
commitfest [3].

"Unlike pgbouncer and other external connection poolera"

Should be "poolers"

"So developers of client applications still have a choice
either to avoid using session-specific operations either not to use
pooling."

That sentence isn't smooth for me to read. Maybe something like:
"Developers of client applications have the choice to either avoid using
session-specific operations, or not use built-in pooling."

[1]
https://www.postgresql.org/message-id/attachment/102610/builtin_connection_proxy-8.patch

[2]
https://www.postgresql.org/message-id/flat/CAP_rwwmLJJbn70vLOZFpxGw3XD7nLB_7%2BNKz46H5EOO2k5H7OQ%40mail.gmail.com

[3] https://commitfest.postgresql.org/23/2055/

Ryan Lambert

On Tue, Jul 16, 2019 at 12:20 AM Konstantin Knizhnik <
k(dot)knizhnik(at)postgrespro(dot)ru> wrote:

>
>
> On 15.07.2019 17:04, Konstantin Knizhnik wrote:
> >
> >
> > On 14.07.2019 8:03, Thomas Munro wrote:
> >>
> >> On my FreeBSD box (which doesn't have epoll(), so it's latch.c's old
> >> school poll() for now), I see the connection proxy process eating a
> >> lot of CPU and the temperature rising. I see with truss that it's
> >> doing this as fast as it can:
> >>
> >> poll({ 13/POLLIN 17/POLLIN|POLLOUT },2,1000) = 1 (0x1)
> >>
> >> Ouch. I admit that I had the idea to test on FreeBSD because I
> >> noticed the patch introduces EPOLLET and I figured this might have
> >> been tested only on Linux. FWIW the same happens on a Mac.
> >>
> > I have committed patch which emulates epoll EPOLLET flag and so should
> > avoid busy loop with poll().
> > I will be pleased if you can check it at FreeBSD box.
> >
> >
> Sorry, attached patch was incomplete.
> Please try this version of the patch.
>
>