* Stephen Hahn <sch at sun.com> [2007-06-28 13:28]:
> Well, the most reliable way would be to have a monitoring framework of
> some kind, that tested for liveness, rather than just existence. That
> contract duration is the default is based on our assessment of how
> much software daemonizes. If I were targetting an OpenSolaris OS with
> smf(5) as a daemon author, I would seriously consider not daemonizing.
Okay, I need to clarify that last part, since Dave and Dan came by to
beat me up. The best thing to do, as a daemon author, is to make your
daemon's start-up precise: the first process shouldn't exit until the
second process has successfully started answering requests (or is
otherwise performing its service). There are a couple of examples:
- daemonize_start() in svc.configd uses a pipe to send a "success
byte" back to the parent:
http://src.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/cmd/svc/configd/configd.c
- main() in zoneadmd uses a semaphore in a shared page to coordinate
parent and child (because they didn't want any open file descriptors
to cross the zone boundary, I expect):
http://src.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/cmd/zoneadmd/zoneadmd.c
What I was too quickly commenting on is that I don't believe closing
stdout/stderr are a part of daemonizing any longer--smf(5) would
really like to capture and present diagnostic output your daemon emits
to the administrator.
- Stephen
--
sch at sun.com http://blogs.sun.com/sch/
