Thanks. We do have it set to 15 mb. I would think that 16 mb would not make a big difference. Do you have any other ideas?
-----Original Message----- From: Kevin Barnard [mailto:[EMAIL PROTECTED] Sent: Tuesday, October 26, 2004 2:32 PM To: Brian Maguire Cc: [EMAIL PROTECTED] Subject: Re: [GENERAL] what could cause inserts getting queued up and db locking?? This sounds like a WAL segment recycling issue. I believe 8.0 should relieve some of the stress. I've had this problem before and found that increasing the number of check point segments has helped a little. You still get the a wallop when this happens, increasing the size should make that happen less often. The downside is each of the files, segments, is 16meg. On Tue, 26 Oct 2004 13:27:53 -0400, Brian Maguire <[EMAIL PROTECTED]> wrote: > One observation that we made was right when the statements pile up there > is a large increase in the number of disk reads. The entire issue lasts > approx. 20 secs and then everything recovers. There will be a backlog > of 300+ statements and then all a sudden it seems to get resolved. > > We though there might be locking, but noticed that there were not any > queries in wait mode indicating that no statements were blocked by > another statement's lock. > > Can anyone suggest what may be causing the pile up/locking like > occurrences? > > Brian > > -----Original Message----- > From: Richard Huxton [mailto:[EMAIL PROTECTED] > Sent: Tuesday, October 26, 2004 4:46 AM > To: Brian Maguire > Cc: [EMAIL PROTECTED] > Subject: Re: [GENERAL] what could cause inserts getting queued up and db > locking?? > > Brian Maguire wrote: > > > > What could cause the database to lock up and queue up all the > > queries? > > You'll want to check the lock details (pg_locks: see "Monitoring > Database Activity" in the reference manuals) and also what the system as > > a whole is doing (vmstat/iostat). > > I seem to recall some configuration of ext3 could cause bursts of > activity with certain write patterns, but don't recall PG being affected > > by this. > > There have been discussions of context-switching issues with Xeons IIRC > - don't know the details, so check the -performance/hackers archives for > > details. > > HTH > -- > Richard Huxton > Archonet Ltd > > ---------------------------(end of broadcast)--------------------------- > TIP 1: subscribe and unsubscribe commands go to [EMAIL PROTECTED] > ---------------------------(end of broadcast)--------------------------- TIP 9: the planner will ignore your desire to choose an index scan if your joining column's datatypes do not match
