On Oct 27, 2004, at 0:57, TTK Ciar wrote:
> At a guess, it looks like the data set is being buffered in its
> entirety by psql, before any data is written to the output file,
> which is surprising. I would have expected it to grab data as it
> appeared on the socket from postmaster and write it to disk. Is
> there something we can do to stop psql from buffering results?
> Does anyone know what's going on here?
Yes, the result set is sent back to the client before it can be used.
An easy workaround when dealing with this much data is to use a cursor.
Something like this:
db# start transaction;
db# declare logcur cursor for select * from some_table;
db# fetch 5 in logcur;
This will do approximately what you expected the select to do in the
first, place, but the fetch will decide how many rows to buffer into
the client at a time.
> If the solution is to just write a little client that uses perl
> DBI to fetch rows one at a time and write them out, that's doable,
> but it would be nice if psql could be made to "just work" without
> the monster RSS.
It wouldn't make a difference unless that driver implements the
underlying protocol on its own.
SPY My girlfriend asked me which one I like better.
pub 1024/3CAE01D5 1994/11/03 Dustin Sallings <dustin(at)spy(dot)net>
| Key fingerprint = 87 02 57 08 02 D0 DA D6 C8 0F 3E 65 51 98 D8 BE
L_______________________ I hope the answer won't upset her. ____________
In response to
pgsql-performance by date
|Next:||From: Markus Bertheau||Date: 2004-10-31 10:27:01|
|Subject: Re: psql large RSS (1.6GB)|
|Previous:||From: Markus Bertheau||Date: 2004-10-30 21:12:25|
|Subject: Re: can't handle large number of INSERT/UPDATEs|