http://bugzilla.suse.com/show_bug.cgi?id=1112991 Zejin Xu changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |jxu@suse.com, ro@suse.com Assignee|bnc-team-screening@forge.pr |systemd-maintainers@suse.de |ovo.novell.com | Summary|syatemd-rpm-macros: missing |systemd-rpm-macros: missing |method for reload instead |method for reload instead |of restart on update |of restart on update -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=1112991 Franck Bui changed: What |Removed |Added ---------------------------------------------------------------------------- Flags| |needinfo?(ro@suse.com) -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=1112991 http://bugzilla.suse.com/show_bug.cgi?id=1112991#c3 --- Comment #3 from Franck Bui --- (In reply to Ruediger Oertel from comment #2)

it's basically the old "ExecRestart" discussion.

Any links to share ?

Some services want to implement the Restart action themselves, in this case because they want to pass on file descriptors to their successor.

What do you mean by "themselves" ? PID1 offers an API to services/daemons for sending and storing their fds for this case. See logind and journald for examples. Do you mean that in your case this API can't be used ? if so can you explain why ? -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=1112991 http://bugzilla.suse.com/show_bug.cgi?id=1112991#c5 Michael Schröder changed: What |Removed |Added ---------------------------------------------------------------------------- Flags|needinfo?(mls@suse.com) | --- Comment #5 from Michael Schröder --- As Rudi said: we wan't a graceful restart with no downtime at all. Regarding #3: what interface? Currently a restart is hardcoded as first stop then start, so how would such an interface be useful at all? The service doesn't know that it's going to be started again. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=1112991 http://bugzilla.suse.com/show_bug.cgi?id=1112991#c7 --- Comment #7 from Michael Schröder --- And how would this work from a perl program? -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=1112991 http://bugzilla.suse.com/show_bug.cgi?id=1112991#c9 --- Comment #9 from Michael Schröder --- (It's also not clear to me that stopping the service will make systemd close all stored fds. Is that the case?) -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=1112991 http://bugzilla.suse.com/show_bug.cgi?id=1112991#c11 --- Comment #11 from Michael Schröder ---

The daemon is implemented in perl ? that's not really common...

Now that's a helpful comment...

Well, how do you plan to implement that exactly otherwise ?

Exactly what Rudi proposed.

Sure.

(I couldn't find that in the systemd code, that's why I was asking.) -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=1112991 http://bugzilla.suse.com/show_bug.cgi?id=1112991#c13 --- Comment #13 from Michael Schröder --- Well, "true" restarting is in the eye of the beholder. It's not "reloading its configuration", it's actually restarting, aka re-executing the new binary. It's a mis-use of "reload" because systemd's restart means a hard stop/start. Your suggestions are not really helping. Please show me how to do that filedescriptor sending/receiving in perl. I'm fine with simply overwriting your systemd macros in our packages, so don't worry. I find your attitude a bit strange, though. Have a nice day! -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=1112991 http://bugzilla.suse.com/show_bug.cgi?id=1112991#c17 --- Comment #17 from Franck Bui --- (In reply to Michael Schröder from comment #16)

Anyway, here's some more information from me:

Thanks, please next time include such description earlier.

The package we're talking about is the OBS backend server, serving requests on a socket. Our goal is to do a graceful restart on package update. This means:

1) we do not want to close the server socket 2) we want the current jobs to finish (this might take a couple of minutes for some types, the server has special code to deal with longrunning jobs)

So we can't use systemd's restart mechanism which does a hard start/stop.

Well as discussed previously, you're supposed to use sd_notify() in this case. But it appears that no perl bindings exists currently so they should implemented like python-systemd does for python.

Instead, we opted to 'hijack' the reload mechanism. (A graceful restart is also the only way to reload the server config, so it's not that far-fetched.)

We do this since quite some time with sysv-init style scripts, but now we want to migrate to systemd service files. That's why Rudi is asking for some way to make the package update to a reload instead of a restart.

I understand that the current proposed change is somewhat invasive for this special case, but maybe we can do something simpler: How about adding

The problem is not about how invasive the changes would be but making a somehow 'special' case officially available. Again the official way for a service to preserve its file-descriptors is to send them to PID1. IMHO until such API becomes available for perl, you should just call 'try-reload-or-restart' command in the %postun section of your package. Hope that makes sense. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=1112991 http://bugzilla.suse.com/show_bug.cgi?id=1112991#c19 --- Comment #19 from Franck Bui --- How does this relate to systemd rpm macros ? I mean the jobs part is not something that systemd is supposed to provide support for you. This actually suggests that the restart of your service is not properly implemented or designed and doesn't support re-executing. PID1 or other services have the same problems to deal with. Nevertheless PID1, for instance, is not doing "reload instead of restart" even though it manages a lot of long standing services which must not be stopped while PID1 is restarted. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=1112991 http://bugzilla.suse.com/show_bug.cgi?id=1112991#c21 --- Comment #21 from Franck Bui --- (In reply to Michael Schröder from comment #20)

The point is that we do not want a "restart on update" in the systemd sense. We want something more graceful on package update.

The problem here is that different packages will need different custom actions. For example one will need to be re-executed with a command of its own (such as fooctl reexecute) another one might want to receive a signal, in your case you want to reload and so on.. I don't see how all of these different cases could be hidden via a rpm macro and the point of doing it actually. The rpm macros have been created to deal with the generic cases only where stop/start is enough. If you need to do more involved stuff, then it's better to write the commands explicitely instead of trying to hide them behind macros. -- You are receiving this mail because: You are on the CC list for the bug.

http://bugzilla.suse.com/show_bug.cgi?id=1112991 http://bugzilla.suse.com/show_bug.cgi?id=1112991#c23 --- Comment #23 from Franck Bui --- Well I hope the arguments made sense. -- You are receiving this mail because: You are on the CC list for the bug.