David Powell wrote:
> Mark Fenwick wrote:
> > wadworth# ctwatch -r 1718
> > CTID EVID CRIT ACK CTTYPE SUMMARY
> >
> > *** kill 455 - nothing *****
> >
> > *** kill 453 *****
> > 1718 1071 crit no process process 453 received a fatal signal
> > 1718 1072 crit no process contract empty
> >
> >
> > BTW if I change my method so that it only launches a single daemon
> > then killing this process results in the service being restarted.
>
> It sounds like PID 455 isn't dying from the signal. Instead it is
> voluntarily exiting after receiving the signal. This cannot be
> detected by the kernel, and therefore won't result in a contract
> event.
FYI, I've filed
6668167 explicitly generated process contract events
to cover a possible solution to the general problem of process
failures that cannot be detected by external observers.
Another solution -- frequently-mentioned but unimplemented -- are
service monitors. Roughly speaking, these would let you provide a
method that would verify all the necessary pieces of your service are
running and providing the required service. See
6533008 support service monitors
for more.
Dave
