[ClusterLabs] Antw: Re: Antw: Re: Antw: Re: pacemaker resources under systemd

[Ulrich Windl]

>>> Andrei Borzenkov <arvidjaar at gmail.com> schrieb am 12.09.2019 um 15:00 in
Nachricht
<CAA91j0UQVA=v9Qe9jvgi+7GsNsR7+63+9kz3VtQfJAG4qRW1Zg at mail.gmail.com>:
> On Thu, Sep 12, 2019 at 3:45 PM Ulrich Windl
> <Ulrich.Windl at rz.uni‑regensburg.de> wrote:
>>
>> >>> Andrei Borzenkov <arvidjaar at gmail.com> schrieb am 12.09.2019 um 14:21
in
>> Nachricht
>> <CAA91j0UnCSS08ZKyjYn26eysVKwVVuh4hryF98HN5jK7SEU2UQ at mail.gmail.com>:
>> > On Thu, Sep 12, 2019 at 12:40 PM Ulrich Windl
>> > <Ulrich.Windl at rz.uni‑regensburg.de> wrote:
>> >>
>> >> Hi!
>> >>
>> >> I just discovered an unpleasant side‑effect of this:
>> >> SLES has "zypper ps" to show processes that use obsoleted binaries. Now
if
>>
>> > any
>> >> resource binary was replaced, zypper suggests to restart pacemaker
(which
>> is
>> >> nonsense, of course).
>> >>
>> >> Example:
>> >> # zypper ps
>> >> The following running processes use deleted files:
>> >>
>> >> PID    | PPID  | UID | User  | Command           | Service   | Files
>> >>
>> >
>>
‑‑‑‑‑‑‑+‑‑‑‑‑‑‑+‑‑‑‑‑+‑‑‑‑‑‑‑+‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑+‑‑‑‑‑‑‑‑‑‑‑+‑‑‑‑‑‑‑‑‑‑‑‑‑‑‑
>> > ‑‑‑‑‑‑‑‑‑‑‑‑‑‑
>> >> 2558   | 92480 | 0   | root  | isredir (deleted) | pacemaker |
>> >> /usr/bin/isredir (deleted)
>> >>
>> >> The file definitely is not a part of pacemaker!
>> >>
>> >
>> > Neither zypper tells you that it is. All that zypper tells you is that
>> > binary of some process that was started as part of pacemaker *SERVICE*
>> > was deleted and that if you want to refresh it and ensure that process
>> > is using updated binary you need to restart *SERVICE* pacemaker. Which
>>
>> That is exactly the point: You DO NOT need to restart pacemaker; it is
>> sufficient to restart the corresponding resource.
>>
>> > is absolutely correct. Restarting pacemaker service makes sure
>> > everything used by pacemaker is up to date. That pacemaker is capable
>> > of restarting only some processes is not known to zypper. While it
>> > sure is possible to extend zypper to recognize pacemaker, parse
>> > current configuration and suggest to restart specific resource, this
>> > time is probably better spent somewhere else.
>>
>> You absolutely miss the point:
> 
> I do not.
> 
>> In a 24x7 environment (a classic  cluster
>> application) restarting pacemaker means that every resource is stopped and
>> restarted. If you do it in maintenance mode, the resources will not be
>> restarted, but also the problem will not go away. Not restarting pacemaker

> is
>> worth the time!
>>
> 
> This is common misconception. Your 24x7 environment must have planned
> maintenance windows exactly so you can do maintenance that requires
> stopping and starting process/services/systems. And in such critical
> environments updates are not installed at arbitrary times, but during
> these maintenance windows.
> 
> But sure, if zypper can do better, that's fine. I am just not sure
> whether zypper developers can be reached on this forum ...

Maybe it can be fixed in pacemaker's unit file also. I'don't know.
As said before, it feels just wrong to me to run all cliuster resources in the
pacemaker slice.

Regards,
Ulrich