R devel - Aug 2018 - Detecting whether a process exists or not by its PID?

On Fri, Aug 31, 2018 at 2:51 PM Tomas Kalibera <tomas.kalibera at
gmail.com> wrote:
[...]
> kill(sig=0) is specified by POSIX but indeed as you say there is a race
> condition due to PID-reuse.  In principle, detecting that a worker
> process is still alive cannot be done correctly outside base R.
I am not sure why you think so.
> At user-level I would probably consider some watchdog, e.g. the parallel
> tasks would be repeatedly touching a file.
I am pretty sure that there are simpler and better solutions. E.g. one
would be to
ask the worker process for its startup time (with as much precision as possible)
and then use the (pid, startup_time) pair as a unique id.

With this you can check if the process is still running, by checking
that the pid exists,
and that its startup time matches.

This is all very simple with the ps package, on Linux, macOS and Windows.

Gabor
> In base R, one can do this correctly for forked processes via
> mcparallel/mccollect, not for PSOCK cluster workers which are based on
> system() (and I understand it would be a useful feature)
>
>  > j <- mcparallel(Sys.sleep(1000))
>  > mccollect(j, wait=FALSE)
> NULL
>
> # kill the child process
>
>  > mccollect(j, wait=FALSE)
> $`1542`
> NULL
>
> More details indeed in ?mcparallel. The key part is that the job must be
> started as non-detached and as soon as mccollect() collects is,
> mccollect() must never be called on it again.
>
> Tomas
>
> >
> > I can the PID of each cluster nodes by querying them for their
> > Sys.getpid(), e.g.
> >
> >      pids <- parallel::clusterEvalQ(cl, Sys.getpid())
> >
> > Is there a function in core R for testing whether a process with a
> > given PID exists or not? From trial'n'error, I found that on
Linux:
> >
> >    pid_exists <- function(pid) as.logical(tools::pskill(pid, signal
= 0L))
> >
> > returns TRUE for existing processes and FALSE otherwise, but I'm
not
> > sure if I can trust this.  It's not a documented feature in
> > ?tools::pskill, which also warns about 'signal' not being
standardized
> > across OSes.
> >
> > The other Linux alternative I can imagine is:
> >
> >    pid_exists <- function(pid) system2("ps", args =
c("--pid", pid),
> > stdout = FALSE) == 0L
> >
> > Can I expect this to work on macOS as well?  What about other *nix
systems?
> >
> > And, finally, what can be done on Windows?
> >
> > I'm sure there are packages on CRAN that provides this, but
I'd like
> > to keep dependencies at a minimum.
> >
> > I appreciate any feedback. Thxs,
> >
> > Henrik
> >
> > ______________________________________________
> > R-devel at r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-devel
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel

On 08/31/2018 03:13 PM, G?bor Cs?rdi wrote:
> On Fri, Aug 31, 2018 at 2:51 PM Tomas Kalibera <tomas.kalibera at
gmail.com> wrote:
> [...]
>> kill(sig=0) is specified by POSIX but indeed as you say there is a race
>> condition due to PID-reuse.  In principle, detecting that a worker
>> process is still alive cannot be done correctly outside base R.
> I am not sure why you think so.
To avoid the race with PID re-use one needs access to signal handling, 
to blocking signals, to handling sigchld. system/system2 and 
mcparallel/mccollect in base R use these features and the interaction is 
still safe given the specific use in system/system2 and 
mcparallel/mccollect, yet would have to be re-visited if either of the 
two uses change. These features cannot be safely used outside of base R 
in contributed packages.

Tomas
>
>> At user-level I would probably consider some watchdog, e.g. the
parallel
>> tasks would be repeatedly touching a file.
> I am pretty sure that there are simpler and better solutions. E.g. one
> would be to
> ask the worker process for its startup time (with as much precision as
possible)
> and then use the (pid, startup_time) pair as a unique id.
>
> With this you can check if the process is still running, by checking
> that the pid exists,
> and that its startup time matches.
>
> This is all very simple with the ps package, on Linux, macOS and Windows.
>
> Gabor
>
>> In base R, one can do this correctly for forked processes via
>> mcparallel/mccollect, not for PSOCK cluster workers which are based on
>> system() (and I understand it would be a useful feature)
>>
>>   > j <- mcparallel(Sys.sleep(1000))
>>   > mccollect(j, wait=FALSE)
>> NULL
>>
>> # kill the child process
>>
>>   > mccollect(j, wait=FALSE)
>> $`1542`
>> NULL
>>
>> More details indeed in ?mcparallel. The key part is that the job must
be
>> started as non-detached and as soon as mccollect() collects is,
>> mccollect() must never be called on it again.
>>
>> Tomas
>>
>>> I can the PID of each cluster nodes by querying them for their
>>> Sys.getpid(), e.g.
>>>
>>>       pids <- parallel::clusterEvalQ(cl, Sys.getpid())
>>>
>>> Is there a function in core R for testing whether a process with a
>>> given PID exists or not? From trial'n'error, I found that
on Linux:
>>>
>>>     pid_exists <- function(pid) as.logical(tools::pskill(pid,
signal = 0L))
>>>
>>> returns TRUE for existing processes and FALSE otherwise, but
I'm not
>>> sure if I can trust this.  It's not a documented feature in
>>> ?tools::pskill, which also warns about 'signal' not being
standardized
>>> across OSes.
>>>
>>> The other Linux alternative I can imagine is:
>>>
>>>     pid_exists <- function(pid) system2("ps", args =
c("--pid", pid),
>>> stdout = FALSE) == 0L
>>>
>>> Can I expect this to work on macOS as well?  What about other *nix
systems?
>>>
>>> And, finally, what can be done on Windows?
>>>
>>> I'm sure there are packages on CRAN that provides this, but
I'd like
>>> to keep dependencies at a minimum.
>>>
>>> I appreciate any feedback. Thxs,
>>>
>>> Henrik
>>>
>>> ______________________________________________
>>> R-devel at r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-devel
>> ______________________________________________
>> R-devel at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-devel

On Fri, Aug 31, 2018 at 3:35 PM Tomas Kalibera <tomas.kalibera at
gmail.com> wrote:
>
> On 08/31/2018 03:13 PM, G?bor Cs?rdi wrote:
> > On Fri, Aug 31, 2018 at 2:51 PM Tomas Kalibera <tomas.kalibera at
gmail.com> wrote:
> > [...]
> >> kill(sig=0) is specified by POSIX but indeed as you say there is a
race
> >> condition due to PID-reuse.  In principle, detecting that a worker
> >> process is still alive cannot be done correctly outside base R.
> > I am not sure why you think so.
> To avoid the race with PID re-use one needs access to signal handling,
> to blocking signals, to handling sigchld. system/system2 and
> mcparallel/mccollect in base R use these features and the interaction is
> still safe given the specific use in system/system2 and
> mcparallel/mccollect, yet would have to be re-visited if either of the
> two uses change. These features cannot be safely used outside of base R
> in contributed packages.
Yes, _in theory_ this is right, and of course this only works for
child processes.

_In practice_, you do not need signal handling. The startup time stamp
method is
completely fine, because it is practically impossible to have two
processes with the
same pid and the same (high precision) startup time. This method also
works for any
process (not just child processes), so for PSOCK clusters as well.

Gabor

[...]