[DTrace-devel] [PATCH] test: increase child process runtime

[Kris Van Hees]

On Wed, Mar 06, 2024 at 11:07:24PM +0000, Nick Alcock wrote:
> On 6 Mar 2024, Kris Van Hees via DTrace-devel verbalised:
> 
> > Withdrawn - Eugene pointed out correctly that this cannot be the correct
> > fix because the test already run with a timeout of 41s.
> >
> > Looking at another possiblity, which is that one of the child processes is
> > not fully started by the time we try to put a probe on it, causing the probe
> > not to be found.  That would mean we may need a delay between starting the
> > child processes and starting dtrace.
> 
> Isn't this just what evaltime's default value of 'main' is supposed to
> solve, without our having to do anything?

Hm, but is that actually being honoured here?  How could it be, since we do
not actually pass pids to dtrace as processes to monitor/trace?  All we do is
specify pid probes - and that you can do with processes no matter where they
are in their execution cycle.

> (However... that assumes we can successfully drop breakpoints on things.
> On your ksplice-afflicted system, we almost certainly don't have working
> symbol resolution, so we can't, so evaltime is broken and we free-spin
> and probably don't get hold of it until it's too late.)

The system that had ksplice-aware libs isn't the one that shows the issue :)