Optional watchdog for "adapter disconnected"-type events (non-node-crash)

[Nerivec]

Add basic, optional, watchdog to handle "soft failures" for bare Z2M users instead of requiring the use of a process watchdog. Should handle any Z2M failure that would result in the node process being stopped (like "adapter disconnected" event).

npm run start => no watchdog
Z2M_WATCHDOG=default npm run start => watchdog with default delays (1min, 5min, 15min, 30min, 60min)
Z2M_WATCHDOG=minutes_csv npm run start => watchdog with custom delays (example: Z2M_WATCHDOG=0.5,3,6,15 npm run start)

• The number of configured delays is the de facto number of times the watchdog will retry, past that, the node process will be stopped (to avoid endlessly retrying when clearly, something's requiring the user's attention). With the default delays, the watchdog will retry after 1min on first failure, then after 5min on second failure, then after 15min on third failure, then after 30min on fourth failure, then after 60min on fifth failure, then exit if sixth start fails. Any successful start resets that to the beginning.
• The watchdog will only trigger on failure after the initial (manual) start is successful.
• A problem with settings will always ignore the watchdog and stop Z2M.
• A manual stop/restart will ignore the watchdog to comply with user intent.
• Z2M_WATCHDOG accepts default as value or a list of comma-separated integers/floats. Other values/formats (invalid) will prevent Z2M from starting.

Note: This does not handle node crashes, that's better suited for a process watchdog (and already widely available).

• Docs: Add watchdog page. zigbee2mqtt.io#2825

[Koenkk]

Looks good! I think the best place to document this is above here

[Nerivec]

That would only provide the information on Linux side though. Do you want to duplicate the paragraph on all OS pages?

[Koenkk]

I think for each setup the way to enable this is a bit different, but Im not sure if we want it for the Docker setup for example.

[Nerivec]

For containerized setups, I guess it can remove some overhead, by avoiding container reset, and just trying to launch the controller again.

[Koenkk]

Im thinking that it might be easier to enable this through an env var, Z2M_WATCHDOG, that makes it easier to configure

[Nerivec]

Z2M_WATCHDOG=default => use default delays
Z2M_WATCHDOG=1,2,3 => use 1,2,3 for delays
Something like this?

[Koenkk]

Yes, and with everything that are not csv numbers, z2m should refuse to start

[Koenkk]

For the docs, Im wondering if a new Watchdog page under Configuration makes sense

[Nerivec]

How about under Installation? Right above Zigbee2MQTT fails to start looks good.

[Koenkk]

Thats also fine