[systemd-devel] Should a unit be able to start, while its OnFailure unit is active?

[Zbigniew Jędrzejewski-Szmek]

On Wed, Jan 14, 2015 at 04:24:15PM +0100, Jean-Pierre.Bogler at continental-corporation.com wrote:
> Hi all,
> 
> I'd like to have your opinion on the following problem: 
> 
> In case a unit fails, we are using an OnFailure unit to 
> handle the error (e. g. reset the config of the failed 
> unit) and restart it.
> 
> In one case the failed unit had dependencies to other 
> units. Therefore, the failed unit was (re-)started when 
> the other units started.
> 
> This way, the OnFailure unit was active (which could 
> delete the config), *while* the failed unit, which reads 
> the config, was restarting!
> 
> Is this behavior intended or could it be an advantage to 
> let a unit "conflict" to its OnFailure unit in some way? 
Yes, it's intended. 

> A first idea for a workaround is to add an "After" 
> dependency to the OnFailure unit in the real unit's 
> service file. This way a job for the unit should be 
> created but the unit would not start until the 
> OnFailure unit finbished. Is this correct?
That's should work.

Zbyszek