[Pkg-dns-devel] Bug#826241: Bug#826241: Bug#826241: Bug#826241: Bug#826241: unbound: Provide $named facility under systemd

[Michael Biebl]

Am 14.07.2016 um 12:34 schrieb Martin Pitt:
> Hello Michael,
> 
> Michael Biebl [2016-07-14 12:27 +0200]:
>> I don't see a reason to only apply --ignore-dependencies only on boot
>> and for reload. Maybe we should just run it unconditionally (line 65).
>> That would make the behaviour more consistent.
>> I think I mentioned that already in an earlier email.
> 
> I think this will introduce breakage/regressions. If a package runs
> "invoke-rc.d foo start" in its postinst, it is currently fully
> entitled to assume that foo's dependencies will be started along with
> it. I don't want to break that.

This was never the case for SysV, fwiw.
/etc/init.d/foo or invoke-rc.d foo only affect the service itself.
So I don't think we actually regress here, we make it actually more
(bug) compatible with the old SysV behaviour.

My concern is, that the more conditionals we add to the lsb hook, the
harder it get's to debug issues as you don't necessarily know anymore
which behaviour was applied. Doing it always makes it simple and
straightforward.

 This is not an issue with "reload" as
> you can assume that the service itself is already running and thus all
> of its dependencies should. But that isn't true for "start".


> We can get away with --ignore-dependencies in early boot since that
> mostly just affects if-up.d/ and similar hooks, not package installs
> or admin/user commands.

As you see, hooks can be run at any time. So I don't really agree with that.


>> That said, once the postfix hook (or any other hook for that matter)
>> uses invoke-rc.d, unbound-resolvconf.service would be affected again.
> 
> Right, invoke-rc.d needs a corresponding fix. I did one this morning
> as part of https://launchpad.net/bugs/1584393 (which is essentially
> the same issue).

Ok, seems consistent to apply the same fix to both invoke-rc.d and the
lsb hook.

>> This is the reason I switched this service to Type=simple, so it
>> shouldn't actually block for completion and we'd be independent from
>> what the hooks are doing.
> 
> Right, that would be a more specific fix that works with current
> systemd/i-s-h. But this should only be done if the service is actually
> conceptually a daemon instead of a oneshot-service. Doing this change
> merely to work around this deadlock sounds a bit ugly.
> 

I don't think using Type=simple is actually that ugly and only reserved
for daemons.

Regards,
Michael


-- 
Why is it that all of the instruments seeking intelligent life in the
universe are pointed away from Earth?

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: OpenPGP digital signature
URL: <http://lists.alioth.debian.org/pipermail/pkg-dns-devel/attachments/20160714/4fbf6cd2/attachment.sig>