[pve-devel] [PATCH manager] services: add restart on-failure to pvescheduler, pvestatd and spiceproxy

[Fiona Ebner]

Am 26.05.25 um 12:38 schrieb Thomas Lamprecht:
> Am 26.05.25 um 10:45 schrieb Fiona Ebner:
>> Same rationale as 4fd2027e ("service: add restart on-failure to
>> pveproxy and pvedaemon") which added the setting for the pveproxy and
>> pvedaemon services.
>>
>> Suggested for pvestatd in the community forum:
>> https://forum.proxmox.com/threads/165597/post-773210
> 
> Fine by me in general, but might be good to recheck if the overall behavior
> of the mechanism makes sense, especially with the default RestartSec=100ms
> (man systemd.service) and the default StartLimitBurst=5 (man systemd.unit),
> which basically means that if the problematic condition is still present,
> it will be restart 5 times in a total span of 500 ms, and then not get
> restarted anymore. The StartLimitIntervalSec=10s default is also a limiting
> factor, but when the services fails fast early it's unlikely to be hit.
> 
> Maybe increasing the interval between restarts a bit (0.5 to 1s?) and/or
> the burst rate (10 to 20 times) might make sense to survive more temporary
> issues would make more sense – there certainly isn't one size fits all here,
> but 5 times in 500 ms is IMO not that ideal for our services here.
> 
> That said, applying this now should not make the status quo worse, beside
> filling the logs with restart failures, making the limited output included
> in the systemctl status commands less useful, but that's hardly a real
> problem.

Yes, this can be better fine-tuned.

Should there be a limit? AFAIU, if we pick e.g. StartLimitBurst=11,
RestartSec=1, then with the default StartLimitIntervalSec=10s, the limit
will never be hit and the service would be tried to be restarted
perpetually. Do we want to keep a limit by also increasing the
StartLimitIntervalSec value accordingly?

I suppose this should be adapted for pvedaemon and pveproxy too then?