Re: freebsd + postgresql 8.3.3 = 100% of cpu usage on stats collector?

hubert depesz lubaczewski <depesz(at)depesz(dot)com> writes:
> 58241 postgres CALL poll(0x7fffffffd4e0,0x1,0x7d0)
> 58241 postgres RET poll -1 errno 4 Interrupted system call
> 58241 postgres CALL poll(0x7fffffffd4e0,0x1,0x7d0)
> 58241 postgres RET poll -1 errno 4 Interrupted system call
> 58241 postgres CALL poll(0x7fffffffd4e0,0x1,0x7d0)
> 58241 postgres RET poll -1 errno 4 Interrupted system call
> 58241 postgres CALL poll(0x7fffffffd4e0,0x1,0x7d0)
> 58241 postgres RET poll -1 errno 4 Interrupted system call
> 58241 postgres CALL poll(0x7fffffffd4e0,0x1,0x7d0)
> 58241 postgres RET poll -1 errno 4 Interrupted system call

Hmm. The wait-for-input loop in the stats collector assumes that
if it gets EINTR, it should just retry the poll() call immediately.
Which AFAIK is correct. I think you're looking at a kernel or libc
bug here --- somehow the EINTR failure is recurring continuously.
It's probably time to ask about this on a FreeBSD-specific list.

regards, tom lane