Job-level kill switch should fall back to a repository variable when the input is unset

[plengauer]

Currently the kill switch for job-level instrumentation is controlled solely by the internal __kill_switch input in actions/instrument/job/action.yml, which defaults to '':

  __kill_switch:
    description: 'INTERNAL ONLY - DO NOT SET MANUALLY' # 'Set OTEL_GITHUB_KILL_SWITCH variable for the repository or organization to any non-empty value to disable all monitoring in case of an emergency. DO NOT SET MANUALLY.'
    default: '' # '${{ vars.OTEL_GITHUB_KILL_SWITCH }}'

The commented-out default shows an earlier attempt to source this from a repository/org variable, but it's currently inert — the live default is just ''. Downstream, config_validation.sh only ever sees INPUT___KILL_SWITCH (the env var GitHub Actions derives from the input):

if [ -n "${INPUT___KILL_SWITCH:-}" ]; then
  echo "::warning ::OpenTelemetry for GitHub actions disabled by kill switch!" && exit 0
fi

Request: more generally, allow configuration to be sourced from repository variables when the corresponding input isn't explicitly set — starting with this kill switch as the concrete case:

• If the __kill_switch input is set (non-empty) → unchanged, kill switch is active.
• If the input is not set (empty/default) → check the repository variable OTEL_KILL_SWITCH. If it exists and is non-empty → treat it the same as if the input had been set to a non-empty value (kill switch active).
• If the variable doesn't exist or is empty → unchanged, kill switch inactive.

Naming to confirm: the ask above uses OTEL_KILL_SWITCH, but note the existing commented-out line references OTEL_GITHUB_KILL_SWITCH instead. Worth deciding which name is canonical before implementing, since it's a repo/org-facing setting.

Related: actions/instrument/workflow/action.yml has the identical commented-out pattern for its own __kill_switch input, while actions/instrument/checksuite/action.yml doesn't reference a variable at all yet. Flagging in case the fix should be applied consistently across all three, even though this issue is scoped to the job-level action.

[github-actions]

Triage Analysis

This is a well-specified enhancement request. The implementation path is clear:

Proposed Solution

1. Naming: Use OTEL_SHELL_CONFIG_GITHUB_KILL_SWITCH to stay consistent with the existing OTEL_SHELL_CONFIG_* namespace, or keep OTEL_GITHUB_KILL_SWITCH from the commented-out code. The commented-out code uses OTEL_GITHUB_KILL_SWITCH — that seems like the canonical choice to preserve backward compatibility intent.

2. In config_validation.sh: After the existing INPUT___KILL_SWITCH check, add a fallback that reads $OTEL_GITHUB_KILL_SWITCH (which GitHub Actions exposes as $OTEL_GITHUB_KILL_SWITCH from vars.OTEL_GITHUB_KILL_SWITCH).

3. In all three action YMLs (job/action.yml, workflow/action.yml, checksuite/action.yml): Restore (or add) the ${{ vars.OTEL_GITHUB_KILL_SWITCH }} as the default for __kill_switch — but this may not be possible since GitHub doesn't allow expressions in default: for inputs. The alternative is to pass it explicitly via env: or let config_validation.sh read a dedicated env var.

Simplest Implementation

In config_validation.sh, change to:

if [ -n "${INPUT___KILL_SWITCH:-}" ] || [ -n "${OTEL_GITHUB_KILL_SWITCH:-}" ]; then
  echo "::warning ::OpenTelemetry for GitHub actions disabled by kill switch!" && exit 0
fi

Then in each action's action.yml, expose the variable via env::

env:
  OTEL_GITHUB_KILL_SWITCH: ${{ vars.OTEL_GITHUB_KILL_SWITCH }}

This avoids the default: expression limitation and keeps logic in the shared config_validation.sh.

Assigning to Copilot agent for implementation.

Generated by Autotriage for #3877 · sonnet46 15.9 AIC · ⌖ 4.09 AIC · ⊞ 5.2K · ◷