Use dbus signals to track service health, instead of checking ActiveState in regular intervals

stackabletech / agent

Stackable Agent - a kubelet written in Rust which uses systemd as its backend

Apache License 2.0

15 stars 9 forks source link

Use dbus signals to track service health, instead of checking ActiveState in regular intervals #182

Open soenkeliebau opened 3 years ago

soenkeliebau commented 3 years ago

The agent currently checks the ActiveState of its managed services every 10 seconds and reacts if a service reports anything else than active at that time.

This method has the potential of missing failures, if the service for example failed and restarted within that 10 second window and is back "in order" by the time the agent checks again.

I believe systemd exposes signals about failed jobs on dbus, which the agent should subscribe to in order to keep track of unit health.

soenkeliebau commented 3 years ago

Decided to remove this from release planning until zbus has released the new version, which will give us reliable information to plan.