Subject: Re: [linux-audio-dev] pthreads and signals
From: Ingo Oeser (ingo.oeser_AT_informatik.tu-chemnitz.de)
Date: Thu May 15 2003 - 21:58:07 EEST
On Thu, May 15, 2003 at 10:39:45AM +0200, David Olofson wrote:
> On Wednesday 14 May 2003 23.52, Ingo Oeser wrote:
> [...]
> > Re-exec is an alternative I could envision for realtime.
>
> Further thoughts on that:
>
> If you have a setup with multiple redundant processes and/or machines,
> the obvious job for the "crash handler" is to tell some other process
> or machine to take over. (This could be a matter of sending some
> commands to a hardware mixer or "router".) After that, you should
> just get out of the way ASAP, and maybe restart.
If you are talking HA, then the other processes (or some kind of
monitor in a more fragile solution) should detect the failure and
take over.
If you crash, then you cannot trust your addressspace anymore.
Detecting a crash is very simple for the other process: Just have
a never to be used pipe open and detect SIGCHLD or SIGPIPE. Then
you even know, which one of processes died.
Cannot do that with threads? What a pity...
Anything more sophisticated, which involves data from the
segfaulting process asks for trouble.
Re-exec means, that you just trust your read-only segment, which
is a good assumption until hardware memory errors exist. If they
exist, you have other problems. I'm talking about exec called
with read-only constants, to clarify that.
Regards
Ingo Oeser
This archive was generated by hypermail 2b28 : Thu May 15 2003 - 22:09:37 EEST