On 1/23/05 1:01 PM, "Tom Lane" <tgl@xxxxxxxxxxxxx> wrote:
> If you like you can try the operation with "set
> enable_seqscan = off", but I bet it will take longer.
Ouch! That was a big mistake. Since inquiring minds want to know, I
decided to give that a try. The expected outcome is to beat the heck out of
the index disks as it read one index and referenced the other to see if the
value existed. What appears to have happened is that it went through the
same process as before, but read each data record via the index. It still
created all the pgsql_tmp files, the data disk was still the heaviest hit
(expected no or little access there), and it beat the living daylights out
of my swap - pageins/outs like crazy. The I/O on the index disks was
negligible compared to the data and swap disks. I won't try that again...
Wes
---------------------------(end of broadcast)---------------------------
TIP 6: Have you searched our list archives?
http://archives.postgresql.org
