autorespawn for crashed modules

[Klaus Knopper]

On Tue, Oct 05, 2010 at 12:25:38PM +0200, Halim Sahin wrote:
> Hi,
> On Thu, Sep 30, 2010 at 10:07:05AM +0200, Hynek Hanke wrote:
> > On 30.9.2010 01:01, Halim Sahin wrote:
> > > What do you think about an autorespawn feature for crashed modules?
> > > This would be a great improovment because the new autorespawn feature in
> > > sd is only useful when the whole server crasshes.
> > >    
> > 
> > Well, autospawn is not designed because of eliminating
> > crashes :) but you are right that it is its side effect and that
> > in a hackish way, it kind of improves the user experience.
> 
> Yes :-).
> 
> > On the other hand, reloading crashed modules makes quite
> > good sense. Especially in the situation when they have to use
> > third party libraries, which can be quite unstable (as with IBM TTS)
> > and are beyond our control.
> 
> ACK.

+yes.

> Stopping resuming after the counter has reached 3 would be possible and
> should work in most situations.
> 
> /* detected festival crash */
> If (counter < 3) 
> {
> resume("ibmtts");
> counter++;
> else {
> counter = 0;
> set_output_module(NEXT_AVAILABLE_MODULE); // fallback from festival to espeak;
> }

I have to disagree here. A better indicator of a "non-working module"
would be the number of crashes WITHIN a certain time period, like

/* Pseudo code */
if ((number_of_crashes / ten_seconds) > 5); then
 kill_module(); replace_by_alternative_module();
else
 respawn_module();
endif

This would honour both cases, unstable (proprietary?) modules that
occasionally need an automatic respawn, as well as completely
non-functioning ones that must be removed if non-starting.

It would surely increase the overall stability feeling of sd.

Regards
-Klaus