[DRBD-user] linstor controller vm in HA on proxmox: missing links in /dev/drbd/by-res after boot

[Roberto Resoli]

Il 05/09/2018 16:03, Roland Kammerer ha scritto:
> On Wed, Sep 05, 2018 at 03:35:44PM +0200, Roberto Resoli wrote:
>> Il 05/09/2018 09:58, Roberto Resoli ha scritto:
>>> [Unit]
>>> After=
>>> After=network-online.target drbd.service
>> A correction: according to
>> man 5 systemd.unit :
>> "Note that dependencies (After=, etc.) cannot be reset to an empty list, so
>> dependencies can only be added in drop-ins. If you want to remove
>> dependencies, you have to override the entire
>> unit."
>>
>> I have re-edited linsor-satellite unit ovverride as follows:
>>
>> [Unit]
>> After=drbd.service
>>
>> [Service]
>> ExecStart=
>> ExecStart=/usr/share/linstor-server/bin/Satellite
>> --logs=/var/log/linstor-satellite --config-directory=/etc/linstor
>> --keep-res=vm-200
> I thought we only have to handle the case for the resource where the
> controller VM is stored to. My reasoning was that basically both
> services can start at the same time. The important res is the one from
> the controller VM. It is not deleted and certainly brought up by
> drbd.service. For the rest I thought "don't care" is good enough. If the
> res files are there, drbd.service brings them up, if not, the satellite
> brings them up (actually does an adjust) when first contacted by the
> controller.

I agree; I read in "man 8 drbd":

====
SYNOPSIS
        /etc/init.d/drbd [resource]
                         {{start} | {stop} | {status} | {reload} | 
{restart} | {force-reload}}

INTRODUCTION
        The /etc/init.d/drbd script is used to start and stop drbd on a 
system
        V style init system.

        In order to use /etc/init.d/drbd you must define a resource, a host,
        and any other configuration options in the drbd configuration 
file. See
        /etc/drbd.conf for details. If resource is omitted, then all of the
        resources listed in the config file are configured.
====

So it should be possible to start/stop une resource only. 
Unfortunatetely, that man page seems no more actual, and the current 
drbd script used in drbd unit (/lib/drbd/drbd) does not support anymore 
the resource poaram.

> So at least I thought that after the first contact from the controller,
> all resources should be up.
> 
> For now, maybe for ever, we should just serialize startup the way you
> did it. That is quite obvious a solution, but I thought it is not
> necessary. Having two services operate on the same files probably is not
> a good idea anyways.

Yes

> So thank you for testing this whole thing, I will add the "After" part
> to the doc.

You're welcome.

rob

> Regards, rck