Awesome

Retry

Smart homes include a network of devices. A case of a failed command can happen due to temporary connectivity issues or invalid device states. The cost of such a failure can be high, especially for background automation. For example, failing to shutdown an irrigation system which should run for 20 minutes can have severe consequences.

The integration increases the automation reliability by implementing 2 custom actions - retry.actions and retry.action. retry.actions is the recommended and UI friendly action which should be used. retry.action is the engine behind the scenes. It's being used by retry.actions but can also be used directly by advanced users.

retry.actions

Here is a short demo of using retry.actions in the automation rule editor:

https://github.com/amitfin/retry/assets/19599059/318c2129-901f-4f6c-8e79-e155ae097ba4

retry.actions wraps any action inside the sequence of actions with retry.action. retry.action performs the original action with a background retry logic on failures. A complex sequence of actions with a nested structure and conditions is supported. retry.actions traverses through the actions and wraps any action step. There is no impact or changes to the rest of the steps. The detailed behavior and the list of optional parameters of retry.action is explained in the section below. All features and parameters of retry.action are also supported by retry.actions, so there is no reason to use a YAML configuration. A straightforward UI usage as demonstrated above should be the way to go.

Note: retry.actions and retry.action are not suitable for the following scenarios:

1. When the order of the actions matters: the background retries are running independently to the rest of the actions.
2. For a relative state change: for example, homeassistant.toggle and fan.increase_speed are relative actions while light.turn_on is an absolute action. The reason is that a relative action might change the state and only then a failure occurs. Performing it again might have an unintentional result.
3. If any action response data is needed: the actions are running in the background and therefore it's not possible to propagate responses.

retry.action

(retry.call is the previous and obsolete name but can still be used.)

This action warps an inner action with background retries on failures. It can be useful to mitigate temporary issues of connectivity or invalid device states.

For example, instead of:

action: homeassistant.turn_on
target:
  entity_id: light.kitchen

The following should be used:

action: retry.action
data:
  action: homeassistant.turn_on
target:
  entity_id: light.kitchen

It's possible to add additional parameters to the data section. The extra parameters will be passed to the inner action.

The inner action will get performed again if one of the following happens:

1. The inner action raises an exception.
2. The target entity is unavailable. Note that this is important since HA silently skips unavailable entities (here).

Here is the list of parameters to control the behavior of retry.action (directly or via retry.actions). These parameters are not passed to the inner action and are consumed only by the retry.action itself.

action parameter (mandatory)

The action parameter is the only mandatory parameter. It contains the name of the inner action. It supports templates.

(service is the previous and obsolete parameter name but can still be used. action and service are mutually exclusive. One of them must be provided.)

retries parameter (optional)

Controls the amount of retries. The default value is 7. For example:

action: retry.action
data:
  action: homeassistant.turn_on
  retries: 10
target:
  entity_id: light.kitchen

backoff parameter (optional)

The amount of seconds to wait between attempts. It's expressed in a special template format with square brackets "[[ ... ]]" instead of curly brackets "{{ ... }}". This is needed to prevent from rendering the expression in advance (relevant core's code is 1, 2, 3, 4). attempt is provided as a variable, holding a zero-based counter - it's zero for 1st time the expression is evaluated, and increasing by one for subsequence evaluations. Note that there is no delay for the initial attempt, so the list of delays always starts with a zero.

The default value is "[[ 2 ** attempt ]]" which is an exponential backoff. These are the delay times of the first 7 attempts: [0, 1, 2, 4, 8, 16, 32] (each delay is twice than the previous one). The following are the attempt offsets from the beginning: [0, 1, 3, 7, 15, 31, 63].

Linear backoff is a different strategy which can be expressed as a simple non-template string (without brackets). For example, these are the delay times of the first 7 attempts when using "10": [0, 10, 10, 10, 10, 10, 10]. The following are the attempt offsets from the beginning: [0, 10, 20, 30, 40, 50, 60].

Another example is "[[ 10 * 2 ** attempt ]]" which is a slower exponential backoff. These are the delay times of the first 7 attempts: [0, 10, 20, 40, 80, 160, 320]. The following are the attempt offsets from the beginning: [0, 10, 30, 70, 150, 310, 630].

action: retry.action
data:
  action: homeassistant.turn_on
  backoff: "[[ 10 * 2 ** attempt ]]"
target:
  entity_id: light.kitchen

expected_state parameter (optional)

Validation of the entity's state after the inner action. For example:

action: retry.action
data:
  action: homeassistant.turn_on
  expected_state: "on"
target:
  entity_id: light.kitchen

If the new state is different than expected, the attempt is considered a failure and the loop of retries continues. The expected_state parameter can be a list and it supports templates.

validation parameter (optional)

A boolean expression of a special template format with square brackets "[[ ... ]]" instead of curly brackets "{{ ... }}". This is needed to prevent from rendering the expression in advance. entity_id is provided as a variable. For example:

action: retry.action
data:
  action: light.turn_on
  brightness: 70
  validation: "[[ state_attr(entity_id, 'brightness') == 70 ]]"
target:
  entity_id: light.kitchen

The boolean expression is rendered after each inner action attempt. If the value is False, the attempt is considered a failure and the loop of retries continues.

Note: validation: "[[ states(entity_id) == 'on' ]]" has an identical logic and impact as setting expected_state: "on". Therefore, the later is preferable from simplicity reasons.

state_delay parameter (optional)

Controls the amount of seconds to wait before the initial check of expected_state and validation (has no impact if both are absent). The default value is zero (no delay). This option can be used if the new state is being updated immediately but later on getting reverted to the previous state since the operation failed on the remote device. It's not the common behavior as integrations should update the state only when getting a new state from the device. Here is a configuration example:

action: retry.action
data:
  action: light.turn_off
  state_delay: 2
  expected_state: "off"
target:
  entity_id: light.kitchen

state_grace parameter (optional)

Controls the amount of seconds to wait before the final check of expected_state and validation (has no impact if both are absent). The default value is 0.2 seconds. The 2nd (final) check is done if the initial check fails. The action attempt is considered a failure only if the 2nd check fails. Here is a configuration example:

action: retry.action
data:
  action: light.turn_off
  transition: 5
  state_grace: 5.1
  expected_state: "off"
target:
  entity_id: light.kitchen

on_error parameter (optional)

A sequence of actions to perform if all retries fail.

Here is an automation rule example with a self remediation logic:

alias: Kitchen Evening Lights
mode: parallel
trigger:
  - platform: sun
    event: sunset
action:
  - action: retry.action
    data:
      action: light.turn_on
      entity_id: light.kitchen_light
      retries: 2
      on_error:
        - action: homeassistant.reload_config_entry
          data:
            entry_id: "{{ config_entry_id(entity_id) }}"
        - delay:
            seconds: 20
        - action: automation.trigger
          target:
            entity_id: automation.kitchen_evening_lights

(This example can be configured in UI mode by using retry.actions. YAML is not needed.)

entity_id is provided as a variable and can be used by on_error templates.

Note that each entity is running individually when the inner action has a list of entities. In such a case on_error can get performed multiple times, once per each failed entity. Similarly, retry.actions has a sequence of actions which might include multiple actions. This can also cause on_error to get performed multiple times, once per each failed inner action.

retry_id parameter (optional)

An action cancels a previous running action with the same retry ID. This parameter can be used to set the retry ID explicitly but it should be rarely used, if at all. The default value of retry_id is the entity_id of the inner action. For an inner action with no entity_id, the default value of retry_id is the action name (e.g. homeassistant.reload_all).

An example of the cancellation scenario might be when turning off a light while the turn on retry loop of the same light is still running due to failures or light's transition time. The turn on retry loop will be getting canceled by the turn off action since both share the same retry_id by default (the entity ID).

Note that each entity is running individually when the inner action has a list of entities. Therefore, they have a different default retry_id. However, an explicit retry_id is shared for all entities of the same action. Nevertheless, retry loops created by the same action (retry.action or retry.actions) are not canceling each other even when they share the same retry_id.

It's possible to disable the cancellation logic by setting retry_id to an empty string (retry_id: "") or null (retry_id: null). In such a case, the action doesn't cancel any other running action and will not be canceled by any other future action. Note that it's not possible to set retry_id to an empty string or null via the "UI Mode" but instead the "YAML Mode" in the UI should be used.

Notes

1. The action does not propagate inner action failures (exceptions) since the attempts are done in the background. However, the action logs a warning when the inner function fails (on every attempt). It also logs an error and issue a repair ticket when the maximum amount of attempts is reached. Repair tickets can be disabled via the integration's configuration dialog.
2. Action supports a list of entities either by providing an explicit list or by targeting areas and devices. It's also possible to specify a group entity. The inner action is performed individually per entity to isolate failures. Group entities are expanded (recursively.)

Install

HACS is the preferred and easier way to install the component, and can be done by using this My button:

Otherwise, download retry.zip from the latest release, extract and copy the content under custom_components directory.

Home Assistant restart is required once the integration files are copied (either by HACS or manually).

The Retry integration should also be added to the configuration in order to use the new custom actions. This can be done via the user interface, by using this My button:

It's also possible to add the integration via configuration.yaml by adding the single line retry:.

Contributions are welcome!

If you want to contribute to this please read the Contribution guidelines

Link to post in Home Assistant's community forum