On Tue, Sep 21, 2004 at 04:24:21PM -0400, Christopher Petrilli wrote:
> Has anyone investigated having either high, or low urgency queries? A
> system I'm working on has a constant inflow of data, which has some
> queries gainst it which might require long sequential scans. I'm not
> that worried about how long those queries take, just that they don't
> interfere with other insertions.
>
> This is a bit DSSish, I guess, but I would think it could be managed
> by nicing processes?
I'd like this feature on some boxes that are being pushed a bit too
close to the limit for comfort.
I've played around with some of the crude ways of doing it. Disk I/O
tends to be the resource that's limited, and process niceness won't
affect that. You'd need to do something like explicitly do a nanosleep
for every X blocks read in by a query or somesuch. Perhaps a
generalization of the vacuum-sleep hack.
Cheers,
Steve
