Path: csiph.com!x330-a1.tempe.blueboxinc.net!usenet.pasdenom.info!aioe.org!news.stack.nl!newsfeed.xs4all.nl!newsfeed6.news.xs4all.nl!xs4all!post.news.xs4all.nl!not-for-mail
MIME-Version: 1.0
In-Reply-To: <CAGGWHE7=LTQ==HzqroetGUvk=kdCYJCnNm=pKsNUtVd+9a9kKA@mail.gmail.com>
References: <CAGGWHE7b-4UvtinLW2U4jWv_uD7_nLMCQBxLjZ2rFEjsTxkZAQ@mail.gmail.com> <mailman.3224.1322838818.27778.python-list@python.org> <31747430.412.1322841476524.JavaMail.geo-discussion-forums@prfx15> <mailman.3228.1322842458.27778.python-list@python.org> <27696236.393.1322843237576.JavaMail.geo-discussion-forums@prfb10> <CAAdqOWHY0S0uaX-OuXR8KQCkAtUPvYZZMoDBwhxO0PhTsVoF8g@mail.gmail.com> <CAGGWHE7=LTQ==HzqroetGUvk=kdCYJCnNm=pKsNUtVd+9a9kKA@mail.gmail.com>
Date: Wed, 7 Dec 2011 11:50:01 -0800
Subject: Re: Multiprocessing: killing children when parent dies
From: Dan Stromberg <drsalists@gmail.com>
To: Mihai Badoiu <mbadoiu@gmail.com>
Content-Type: text/plain; charset=ISO-8859-1
Cc: python-list@python.org, comp.lang.python@googlegroups.com
Precedence: list
Newsgroups: comp.lang.python
Message-ID: <mailman.3384.1323287403.27778.python-list@python.org>
Lines: 36
NNTP-Posting-Host: 2001:888:2000:d::a6
Xref: x330-a1.tempe.blueboxinc.net comp.lang.python:16790

On 12/7/11, Mihai Badoiu <mbadoiu@gmail.com> wrote:
> ok, so the code is something like
> #process A
>   p = Process(...)
>   p.daemon = 1
>   p.start()   # starts process B
> ...
>
> If process A dies (say error, or ctrl-c), or finishes, then process B also
> dies.  But if process A is killed with the "kill" command, then process B
> soldiers on...
>
> Any idea on how to make process B die when process A gets killed by the
> "kill" command?

1) If all you care about is SIGTERM, SIGHUP and the like (and
specifically NOT SIGKILL), you could just install a signal handler
that catches any catchable signals you're interested in.  Then the
signal either kills the children directly, or sets a flag that tells
the main process to do some killing shortly.  Note that threads and
signal handlers don't mix very well - the combination tends to make
the main thread immune to control-C, whether you want it to be or not.
 Also, signal handlers tend to complicate performing I/O, as you're
more likely to read short blocks.

2) If you need to handle SIGKILL gracefully, and you have access to
the code of the child process, you could make sure that the child
isn't setting a SID (?).  ssh, I believe, likes to start a new SID,
making it immune to signals to the parent.  Alternatively, you could
add something to the child process' main loop that polls the parent,
exiting if the parent no longer exists.

3) If you need to handle SIGKILL gracefully, and you don't have access
to the code of the child process, you could use a single extra process
that checks for the presense of the parent, and if it doesn't exist
any more, then kill the children before exiting itself.