Re: [pcp] pmie and system clock changes

[Greg Banks <gnb@xxxxxxxxxxx>]

Martin Hicks wrote:
> 1272757518.386530 gettimeofday({1272757518, 386539}, NULL) = 0
> 1272757518.386562 nanosleep({18446744073709013957, 18446744073070619627}, 
> 0x7fff22cbaf90) = -1 EINVAL (Invalid argument)
>
>
>
> I believe what has happened is:
>
> - Machine's clock is set far in the past
> - pmie starts and notes the epoch for polling the pmcd.  It then polls at even
>   intervals where:  next poll == epoch + poll_count * poll_delay
> - ntp starts, and warps time forward by a couple years.
> - pmie spins, failing nanosleep() because the "next poll" is still a long way in
>   the past and advancing by poll_delay at each subsequent call until it gets to
>   the present time.
>
>
> Attached is my first crack at fixing it, but I don't love it because:
>
>   
The right way to fix this is to use the POSIX API 
clock_gettime(CLOCKID_MONOTONIC) instead of gettimeofday() to calculate 
time intervals for nanosleep().  That is widely available on modern 
OSes, although you should check for it at configure time.
--
Greg.