Re: reducing statistics write overhead

Tom Lane escribió:
> Alvaro Herrera <alvherre(at)commandprompt(dot)com> writes:

> > - Maybe we oughta have separate files, one for each database? That way
> > we'd reduce unnecessary I/O traffic for both the reader and the writer.
>
> The signaling would become way too complex, I think. Also what do you
> do about shared tables?

They are already stored in a "separate database" (denoted with
InvalidOid dbid), and autovacuum grabs it separately. I admit I don't
know what do regular backends do about it.

As for signalling, maybe we could implement something like we do for the
postmaster signal stuff: the requestor stores a dbid in shared memory
and sends a SIGUSR2 to pgstat or some such. We'd have enough shmem
space for a reasonable number of requests, and pgstat consumes them from
there into local memory (similar to what Andrew proposes for
LISTEN/NOTIFY); it stores the dbid and PID of the requestor. As soon as
the request has been fulfilled, pgstat responds by <fill in magical
mechanism that Martin is about to propose> to that particular backend.

--
Alvaro Herrera http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.