Re: squid-1.2beta22 eating gobs of CPU

It appears FD 43 (in the machine I'm playing with), which is ICP socket is
always returning as being writable, but squid never seems to write on this
FD, hence it loops.

Off to look at the code now... (well, recent diffs first).

-cw

On Mon, Jul 20, 1998 at 11:47:20AM +1200, Chris Wedgwood wrote:
> Something funny (peculiar) happened last week.
>
> After running squid-1.2beta22 for a week or so (previously beta20) one of
> the caches starting using 100% CPU (well, as near to it as possible).
>
> No big deal, I just restarted it and assumed something evil had gone wrong.
> I now find, coming in this morning, both active caches are using 100% CPU,
> which is unusually high for caches which normally using at most 5% or so
> when quite busy.
>
> They are still functioning, and still fairly responsive, so I'll leave them
> a little longer, but if anyone has any suggestions on what to check before I
> restart them, please get in quick.
>
> I'll probably grab an strace before restarting, and I notice I'm presently
> doing about 6600 select loop/s, which indicates perhaps on of the FDs is
> hosed and squid hasn't realised?
>
> I'll rebuild with debugging symbols to next time this happens I can attach
> with gdb and try to figure out what FD is screwing select up....
>
> Suggestions or comments?
>
>
>
> -cw
Received on Tue Jul 29 2003 - 13:15:51 MDT