Hi Michael,

>daemon-reload is a costly operation, so we should try to avoid calling it 
too excessively.
>
>So I'm not convinced it is a good idea to generate a daemon-reload (via 
dh_installsystemd in postinst) for packages which do not actually (re)start 
a unit as part of the upgrade process.
>
>By using "--no-enable --no-start" you are basically leaving the 
management of the life cycle of the service entirely to the administrator. 
Wouldn't you agree?
>
>Shouldn't the administrator then call "systemctl daemon-reload" as well?
>systemd will even warn him in cases a .service file has changed (which 
thankfully doesn't happen too often)
>
>Is this actually an issue in practice? Do you have bug reports where 
users of your xpra package have asked for this?

I realize daemon-reload is a costly operation, and I also realize the unit
in question is used by proxy module of Xpra, which is optional.

That's why Dmitry intentionally left it for system administrators to 
configure.

Your point that typically units are not changed daily is also perfectly 
valid.

However, speaking of "Shouldn't the administrator then call "systemctl 
daemon-reload" as well?"… 
No, if one uses unattended-upgrades or any other automation like auter or 
puppet etc.

Server gets rebooted or service gets restarted and start-up fails due to 
requirement of
"daemon-reload" or worse, starts with previous configuration which fails too.

What I suggest is slightly different Debian-wise. We do have list of conffiles 
and list of files
installable by the package. What prevents us from scanning the list for known 
locations
of systemd units ({/usr}/lib/systemd/{system,user}?) and compare them by hash 
or by diff
just like we do it with conffiles? If the units, timers, etc differ, we call 
daemon-reload and
issue a warning to stderr so both user or script catches it. Of course it may 
need some additions
to dpkg and debhelper (or to plain debhelper only?) but this issue will be 
eradicated for all packages
rebuilt after the update.

Of course I can implement the same check in postinst script of Xpra, but I 
decided to discuss
my idea with involved colleagues.

What do you think?
-- 
Vasyl Gello
==================================================
Certified SolidWorks Expert

Mob.:+380 (98) 465 66 77

E-Mail: vasek.ge...@gmail.com

Skype: vasek.gello
==================================================
호랑이는 죽어서 가죽을 남기고 사람은 죽어서 이름을 남긴다

Reply via email to