[OpenIndiana-discuss] complaint about dovecot in sfe IPS not starting.

[Jim Klimov]

On 2013-02-25 16:02, Jonathan Adams wrote:
> actually, probably the issue is more with the fact that the package
> went into maintenance, it would probably be better to go offline if
> the configuration files wasn't there ...
>
> however it trying to run allowed the script to output that the config
> files didn't exist ... so maybe it _is_ best leaving it the way it is.


For moderately integrated packages - i.e. used almost "as is" from
an upstream repo which has little idea about SMF - this is likely
a good approach. At least, it works to forbid running unconfigured
services and attracts administrative attention. Possibly, some more
awareness is needed for admins that just began working with this OS
about how to recover from such errors and where to look for hints
(SMF logs and all). For example, a "restart" of a service in the
"maintenance" state could at least complain that it's not gonna
really restart the service (alternately see RFE #3596).

However, IF the set of required config files is known, they can and
should be listed as dependencies of the SMF instance (or reverse deps
if some block-files are used to require an admin to revise configs
and remove the block-file when he's done).

I don't quite agree that in general, an enabled service which can't
work should be made "offline". There are cases when it might be a
proper thing to do, but in general "maintenance" is just for that,
whatever the reason of a failure.

I am however annoyed by some services that fail remote dep checks
in their method scripts (i.e. can't contact a database server) and
fail quickly several times in a row - and are thus put into the
"maintenance" state. This often mis-happens during poweron of some
larger distributed systems, when clients happen to start before
a DBMS or LDAP server completes its integrity checks.

But this is at least a known case, and we have scripts to work
around that in some cases, or try to configure SMF's tolerance to
quick restarts (or induce delays into start-scripts) in others...
I've long thought about how cool it would be to have a simple means
to check remote systems' SMF states and define dependencies onto
those (including "suffices to have one of that replicated service's
instances running")...

What is other list members' mileage? ;)


//Jim