Error Handling in Automations

[depoll]

I've had trouble with the backend for one of my integrations being flaky and giving 504s (in this case Nest, but it's not the only one), which has resulted in some of my automations failing. I'd love to be able to handle failures in my automations, but continue_on_error is insufficient because there's no way to determine if some step (or sequence of steps) failed and the failure was ignored (e.g. to backoff and retry).

My proposal is to add a try/catch/finally script action that would allow a user to attempt a sequence of actions in the try, handle errors (with a script variable that captures the error message and maybe the alias of the action that failed), and do cleanup work if necessary. It would look something like:

- try:
    - ...
  catch: #optional -- if not provided, the script still errors
    - ...
  finally: #optional -- runs regardless of errors in the catch.  Errors in this sequence are likely logged but not propagated up.
    - ...

Alternatively, one could add similar metadata to a script variable if a continue_on_error occurs, though I think the try/catch/finally is more robust (as it can handle whole sequences and doesn't get overwritten when a second error occurs).

I'm open to helping provide an implementation but wanted to first vet the proposed solution.

[frenck]

Try catch is hard to UI and write sensible from that perspective. Additionally, this brings programming to our automation/YAML, which we really generally try to avoid.

The original idea was to add an on_error parameter to the places the support continue_on_error (automation/sequences/scripts), allowing to provide a list of actions to run when an error occurs (e.g., send notify).

When the on_error script fires, the error that caused the trigger, could be passed in as a variable (for use in, e.g., messages, templates and what else).

I have a branch here somewhere local that partly implements this (which was part of my original effort in adding continue_on_error.

[depoll]

Interesting -- would love to know more about avoiding bringing programming into the automation/YAML. In my mind this isn't that different from If/Then and Repeat Until. Is there a reason the same rough UI for if/then wouldn't work here?

The on_error solution seems ok as well, but then makes control flow harder.

Could be that being a programmer myself makes me want these constructs a lot more -- totally open to that notion ;)

[allenporter]

I like the on_error proposal here since (1) I believe it can be used to implement any other proposal discussed here (2) It is straight forward to understand in the UI what its going to do.

An early sketch of a proposal of the smallest scope I can think of:

• add on_error that acts like a script action.
• errors within on_error are propagated. (To suppress them, use continue_on_error within the on_error)
• timeouts are considered errors (no separate on_timeout)
• initial scope would not include other details about the error

action:
  - service: alarm_control_panel.alarm_arm_home
     data:
      entity_id: alarm_control_panel.security
      code: "{{ code }}"
    on_error:
      - service: persistent_notification.create
        data:
          title: "Alarm was not armed"
          message: "Failed to arm alarm, please check error logs"
      # An error from within on_error is propagated so you can still mark the automation as failed
      - stop: "Failed to arm alarm"
        error: true

Alternative: We could go ahead and include error details in a variable error that contains the string contents of the error. (Is this good to go or needs more thought?)
Alternative: Fire events on automation errors, and handle in separate automations. This would be less flexible, but could keep this out of the script syntax. I don't think this works as well.

[rikroe]

Following up on this, as I have also looked into solving this.

My first try (home-assistant/core#97857) was just adding an additional variable to the script output if continue_on_error was set and an exception raised. This would enable users to write their own "catch" logic using the existing if/then conditions.

However in this PR frenck suggested to rather go with a try/except route which I started in home-assistant/core#98129. There, it was suggested to first discuss this as an architecture discussion.

Should this continue here or should I create a new discussion with a proposal?

To be honest, I don't see a try/catch looking much different than an if/then, with the only difference that the if part is not a condition, but a script itself. However I might be biased as I fully know what to expect from this.

[MartinHjelmare]

I think this is a good place to continue discussing.

[michaelmarconi]

@rikroe, I've just run into this too and I'm surprised to find no explicit error-handling mechanism. I appreciate the argument about not turning this into a programming platform @frenck but I think this is a bit of a grey area, as I cannot find a good way to make my automation reliable or tell me when it fails, so I can sort things out manually.

[MadSpindel]

I also need something similar. In my case I've made my own alarm with Manual Alarm control panel and a Ring Alarm Keypad in Home Assistant. In my Disarm Automation I want to do something like this:

Then do:
Try:

• Set Alarm control panel to 'Disarm' on HA Alarm (with the provided code from Ring Keypad)

Catch:

• Set Invalid code state on Ring Alarm Keypad (it will flash and make a sound)

I don't know how I should do this today. Can't use continue_on_error because it will also say Invalid code when no errors. I could continue_on_error and then create a new event, and a new automation where I check if not Disarmed, and Set invalid code on Ring Keypad. But this would be super easy to do if I could just use a Try Catch action.

[andyfrei]

Would still be great to have such an option. Finding out more likely only "by accident" and definitely much later than it started to fail is quite unhelpful.

How about an option for turning a separate script to monitor them all?
That script catches defined script- and/or automation entities (the ones to monitor) and sends a message to a notify service with the entity-id of the automation/script that failed or produced an error, maybe enhanced with the log title or something?

Maybe "Spook" could be an option ;-)

[r1si]

Since in homeassistant you use components (HASS for example) of which you have no control, you need to have a try/catch or continue_on_error that works even without necessarily using scripts.
As a workaround every time I launch a sensitive service I have to put it in a script that acts as middleware and is launched with continue_on_error.

At the time of writing, notify.alexa_media is down and all my automations that ask Alexa to talk are no longer working.
This is a problem that cannot be contained.

[AJediIAm]

Homey has an "on error" thingy to which actions can be assigned. Maybe it's worth looking into this for inspiration.

I can see this working similar to the expandable "else" option of the "if" action.