Hello Ole and GNU Parallel readers,
I am staring anew because I have taken a different approach than the one
I documented in the other emails I have sent.
This time, I updated the documentation first, so as to describe the new
features I am proposing.
I have pushed the feature description to github:
https://github.com/martinda/gnu-parallel/compare/sigterm-2?expand=1
The problems I try to address are:
- I would like GNU parallel to send SIGTERM to the jobs when kill -TERM
is issued to GNU parallel, because I do not want GNU Parallel to wait
for those jobs to run to full completion, I want them to terminate but
do so gracefully, then pass control back to GNU Parallel.
- I need a longer grace period than the hardcoded 200ms delay. Because
some jobs are part of a bigger chain of scripts where delays can be long.
To provide a flexible and general answer to termination handling, I am
proposing to optionally let the user choose a custom signal termination
sequence, but keep the default one in place if the user does not need a
different one. The default termination sequence is TERM, wait 200ms,
TERM, wait 200ms, KILL.
Should I also publish the documentation changes I posted to github on
this mailing list?
Question:
The termination sequence is used when --timeout expires. It is currently
used under other conditions?
Thank you,
Martin
