[systemd-devel] Restart best practices

["Jóhann B. Guðmundsson"]

On 06/06/2014 11:56 AM, Lennart Poettering wrote:
> On Fri, 06.06.14 09:48, Jóhann B. Guðmundsson (johannbg at gmail.com) wrote:
>
>> >On 06/05/2014 05:00 PM, Lennart Poettering wrote:
>>> > >I think it would make sense to recommend that packagers and upstream
>>> > >packages adopt Restart=on-failure or Restart=on-abnormal (where the
>>> > >former is not appropriate) for all long-running daemons. However, this
>>> > >recommendation should be taken with a grain of salt, as for some
>>> > >services it might be a bad choice to just tape over problems and
>>> > >restarting it again on failure... Hence, this is really not something to
>>> > >blindly apply to all services on the world, but just a recommendation
>>> > >that individually should be considered.
>> >
>> >We already have units in the wild upstream where the internet has
>> >effectively spread ignorance once again which contain
>> >"Restart=Always" when they should not so I would think it would be
>> >better send the signal to upstream not to set these settings et all
>> >but rather expected downstream distribution to set this based on
>> >their own policy so those downstream distribution can provide
>> >consistent end user expectation for their administrative user base.
> It is actually our intention to unify distributions, and thus encourage
> developers to ship unit files upstream that need no modification. I'd
> thus enjoy if Debian and Fedora could adopt similar guidelines here.

I'm very well aware of the intention, it was the reason I signed up in 
the first place but I dont think you will be able to achieve that 
unification with the direction Fedora/Red Hat are taking and probably 
never with distribution that have corporation dictating and deciding 
things in it's background since you have conflict of interest between 
corporates and the community one always.

But let's put that fact aside and say that conflict did not exist then 
we are still faced with the fragmentation between various upstreams 
shipping daemon and or services where for example some upstream refuse 
to ship init scripts of anykind since they are writing their software to 
be used between *nixes and they think this should be handled entirely 
downstream.

So as far as I can tell the closest and the only way we can effectively 
achieve the goal of unifying distribution in that regard is for those of 
us doing systemd integration work in distribution and other places, to 
collaborative maintaining unit files in a centralized repository for 
upstream component, which then upstream and or distribution directly 
would sync from git repository as I think I have already mentioned 
before since this become more or less clear in F15/F16 era.

I created a repository for such thing a while back where I had removed 
the idiotic and no longer necessary Red Hat specific bits which I know 
for a fact will be hard to remove from Fedora with similar things 
existing in other distribution to make the unit files I created truly 
cross distribution and reusable.

Tom did as well create a repository for similar purpose a year or two 
later after some discussion at some conference or hackfest and then 
merged what I had in mine but I think that work has stagnated since he 
started working full time on Systemd and I myself was to busy doing the 
systemd integration into Fedora

With centralized managed repository for type systemd units we ought to 
be better and more efficiently equipped to push changes to type units 
into distribution but first all of us need to agree that should be the 
way forward for type systemd units and start doing the required work to 
make that happen if that agreement is made.

JBG