On 8 Prosinec 2011, 14:20, Andrew Sullivan wrote: > On Thu, Dec 08, 2011 at 01:53:45PM +0100, Tomas Vondra wrote: >> I wonder if the prioritisation could be done using nice - each backend >> is >> a separate process, so why not to do 'nice(10)' for low priority >> processes >> or something like that. > > This won't work because if you are holding a lock on something someone > else needs, your low nice score is going to cause them problems. It > could make things worse rather than better. (This suggestion comes up > a lot, by the way, so there's been a lot of discussion of it > historically.) I'm aware of that, but there are cases when this may actually work. For example we do have an OLTP system, but we need to build exports to other systems regularly. The export may need to read a lot of data, but I don't want to annoy the people who are using the system. So I could lower the priority for the backend generating the report. Yes, there are cases where this "priority inversion" makes it unusable. Tomas -- Sent via pgsql-general mailing list (pgsql-general@xxxxxxxxxxxxxx) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-general
