Stefan Manegold wrote:
On Mon, Apr 13, 2009 at 02:30:44AM +0200, Stefan de Konink wrote:
Martin Kersten wrote:
You can trace the execution even for a server started already using the stethoscope.
What does this/you you want to tell/show us? I.e., does your query still hang, or did it finish now?
The query finishes as I wrote in the previous email, the output was merely an attempt to show were the pain might be.
I see that aggr.count() (not unexpectedly) is the most expensive one of the operations shown below --- but "only" ~100 sec --- not too extreme for counting the size of 362105182 groups over 365652137 BUNs ...
The sum of all shown times is "only" 116865583 usec (i.e., just under 2 min).
Then it is likely the output misses the last actual 'big time consumer', while Martin suggested it would show that event.
Did you compile from CVS or from tarball?
I use CVS head.
Did you configure with either --disable-assert (CVS) or --enable-assert (tarball)?
assert, profile, debug
If you compiled from CVS without --disable-assert or from tarball with --enable-assert, you also might want to start your mserver5 with --performance (or at least --debug=33554432) to disable "cleaning" of to-be-freed memory just before free() is called --- this might make the ":= nil;" assignments (a bit?) faster ...
I am more than happy to try this. Stefan