Restart a workflow from a numbered and/or named step

[pk-nb]

Have a question? Please checkout our Slack Community or visit our Slack Archive.

Describe the Feature

We are starting to use workflows to apply a large number of components.

Given the size of a workflow, it is common for a step in the middle to fail due to user-error (copied the wrong command; missing command) or a configuration error for some step in the middle. Once corrected, it would be nice to restart the workflow from the failed step. This failed step may be a different number if the workflow steps were corrected.

It would be great to pass in an argument to the atmos workflow command (e.g. --from) that took a numbered or named step to start from. For example:

atmos workflow apply-all -f networking --from "terraform deploy tgw/spoke"
atmos workflow apply-all -f networking --from 5

Expected Behavior

Given a named --from argument:

atmos workflow apply-all -f networking --from "terraform deploy tgw/spoke"

• (Happy path) When command is found by name, it starts the workflow from the command (inclusive).
• When command is not found, it gives a helpful error message that the command is not found with exit code 1.
• When multiple commands match, it fails with exit code 1 and suggests using the numeric form of --from. As a bonus, atmos could let you know the step numbers from the multiple matching steps.

Given a numbered --from argument:

atmos workflow apply-all -f networking --from 5

• Steps are 1-indexed for humans (where --from 1 is the first step).
• (Happy path) It starts the workflow from the indexed step (inclusive)
• When the steps is out of range (0, large values) a helpful error message is printed with exit code 1.
• Negative numbers are either disallowed or allow for reverse indexing from the end of the list.

Use Case

Use-cases include:

• Sanity check components that have downstream updates (e.g. apply all 20 networking components in a specific order)
• Recreate a cluster in a new region (create a cluster and apply all N common controllers)

Describe Ideal Solution

See above.

Alternatives Considered

Having to comment out steps and risk of accidentally committing.

Additional Context

Other notes

--from could nicely mirror a --to step.

[osterman]

I like this proposal. Let's support an optional id to identify each step, and that the step # can also be used. Additionally, when a workflow is run, each step id should be emitted so the end user knows how to restart easily from any point.

workflows:
  apply-all:
    steps:
    - id: deploy-vpc # alternative, this id defaults to "1", since it's the first step.
      command: terraform deploy vpc
      stack: core-use1-network