ANN: Closing AC

Tomasz Pala gotar at polanet.pl
Sun Dec 11 02:38:56 CET 2005


On Sun, Dec 11, 2005 at 01:23:12 +0100, Arkadiusz Miskiewicz wrote:

> > Small chances for such think to work, as many of our services shall
> > return HGW instead of DONE.
> Huh? You seem to have no idea how supervising is usually done. The processes 

Indeed.

How long does it wait until assumes that service is running (it should
wait to eliminate risk of races - init part of one program can take
longer than starting another)? What does it do when required service
dies after some other (which depends on it) has already started (or is
starting)? - I'm particullary interested in service restart/reload, as
they're triggered automatically from logrotate.

> are run in foreground as childs of supervising process. If it dies then 
> process knows about that.

Mmhhhmm... 'background'? Hm, why don't we use such a 'supervising
process' now?

Date: Thu, 2 Sep 2004 21:11:32 +0200
Subject: service start
Message-ID: <20040902191132.GA2506 at os>

This problem occours not only with bind, AFAIR I had the same with
quagga (when logging via files not syslog - they still have wrong
permissions in AC), ircd, jabber (broken database), innd (indexes from
other arch, innwatch counting free i-nodes on reiserfs), mailgraph (no
read access to logs), mysql (many many times, I've seen DIED while
server was started, and DONE when died...).

In short: some services can start just-to-die-SOON giving DONE.

> No idea. Maybe bind requires network while zebra becomes part of network when 
> it's installed?

bind requires interfaces (to bind to) - that's for sure. But how about
routing? Dependency means 'can stand up and WILL work if...' (bind to
interfaces, wait for routing and network access) or 'can do it's job
NOW' (needs routing for serving requests)?

> No one started the project means almost everything is 
> unknown.

And I doubt someone will. I won't put it into production machine,
because don't want to end up with everything shut down because of
log rotation while temporary network shutdown (by the way - I've
got quagga-ospf dying a moment after configuration changes...).

After all - what are the advantages? Faster startup (due to
backgrounding jobs)? Sounds coming from my disks say no (maybe if
someone has broken DNS, didn't initialize postgresql database or has
some other misconfiguration). Dependencies? Well, in fact we can imagine
a situation when virtually every service works for it's own without
anything else.

-- 
GoTaR <priv0.onet.pl->gotar>        http://vfmg.sourceforge.net/
                                    http://tccs.sourceforge.net/



More information about the pld-devel-en mailing list