Subject Re: SS 1.0 on Solaris SPARC
Author Adam
--- In, "robert_difalco"
<rdifalco@t...> wrote:
> Very occasionally, we find that our server (when seemingly doing
> nothing -- maybe sweeping?) will peg the CPU at near 100%. It will not
> relinquish its grip on the CPU until we restart FB. Doing so returns
> the system back to normal with no ill effects. What is odd is that
> there are no operations in progress when this happens. Odd, hu?
> Am I alone in this or has this ever been experienced by anyone before?
> It seems very much like what happens when I inadvertantly write some C
> code that infinitely recurses. :)
> R.

Smells like a garbage collection thread to me. If so, the question
becomes why are you accumulating so much garbage. Possibly it might
help if you schedule a sweep during low activity period. In the long
run, identify old transactions (generally long running ones) and make
sure they are committed as soon as they have finished. Commit
retaining is no good, commit needs to be run.

Having old transactions lying around prevents garbage from being
collected as it is stumbled upon. I believe low selectivity indices
can also make sweep take a lot longer.