This is the mail archive of the systemtap@sourceware.org mailing list for the systemtap project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: timestamp problems


Hi -

> [...] considering that I do nothing apart from
> just calling a gettimeofday_ms(), am quite convinced that it is not a
> problem with my script [...]

It seems like a systemtap (runtime) bug.

> The error:  The timestamps after a sometime  jump back by a huge number.
>
> on Fedora 8
> The timestamps starts off with something like 1220226924420
> and goes on till 1220226928771
> (process executing rc.sysinitentry number 133, time in ms since  first
> probe : 4351 ms )
> then suddenly jumps back to 1220207130000

That is "only" 0.229 days (20 million units if we're talking about
milliseconds), but of course still unacceptable.  (I'm assuming the
main system clock does not jump similarly.)

Can you try something like this:

probe timer.s(10) { now=gettimeofday_ns()
                    printf("systemtap %s %6d\n",ctime(now/1000000000),now%1000000000)
                    system("date -u +\"system    %a %b %d %H:%M:%S %Y %N\"")
                    system("grep -i mhz /proc/cpuinfo") }'

and watch for a divergence over time.  The timer.s(NN) parameter could
be changed to narrow in on a suspected trigger time.


- FChE


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]