Core: Prometheus generates "could not convert" error messages for lights

[runningman84]

Home Assistant release with the issue:
0.103.4

Last working Home Assistant release (if known):
none

Operating environment (Hass.io/Docker/Windows/etc.):
docker

Integration:
light / prometheus

Description of problem:
The prometheus export should just ignore this sensor or value. It could also just store the state as -1 or something...

Problem-relevant configuration.yaml entries and (fill out even if it seems unimportant):

prometheus:
  namespace: hass
hue:

Logs (if applicable):

Could not convert <state light.licht_badezimmer=unavailable; friendly_name=Licht Badezimmer, supported_features=0 @ 2019-12-23T21:00:45.793845+01:00> to float

Additional information:
This bug is similar https://github.com/home-assistant/home-assistant/issues/29015

[springstan]

The prometheus export should just ignore this sensor

You could exclude this sensor by specifying it in your configuration. See here for more details.

[runningman84]

The problem is that once a sensor is unavailable you get this message. In order to solve this I would have to exclude everything....

[springstan]

Okay I completely understand, I will look into it.

[runningman84]

Maybe it make sense to use the up metric in order to show if a sensor is available or not...

[springstan]

What do you mean by _up metric_ exactly?

[springstan]

Furthermore, I am pretty sure this is intended behaviour since unavailable does not mean the same as on or off which are represented as 1 and 0 in prometheus. unavailable is a the correct status when HA cannot reach or connect to a particular device and therefore cannot access and convert its state correctly.

[TheEggi]

What is the reason that we are not returning null/nothing if a device is not available? Current behavior with logging an error message and returning 0 seems strange. How would I currently identify in Prometheus that a sensor is not available and trigger an alert afterwards?

[kevinkk525]

I would like to ping this again.
This problem bothers me too. Every time a sensor becomes unavailable, prometheus logs a "0". That is not good behaviour and makes the stored values horrible to work with.

[knyar]

Note #32112 that is trying to improve the way unavailable entities are handled by the prometheus component.

[knyar]

I believe #37456 fixed this, and this issue can be resolved.