InvalidInstanceID.NotFound" in a stop-instances loop - stale inventory
Makes batch stop and terminate loops resilient to instances that vanish mid-run. Use it when a sweep logs InvalidInstanceID.NotFound; the key trigger is an instance terminated between the inventory scan and the action, which should be treated as already done, not as a failure.
TL;DR
The instance was terminated between your scan and your stop call - autoscaling, other automations, and humans all act in that gap. Catch the specific NotFound code, log it as already-gone, and keep going. A few hits are normal; a flood means your inventory source is stale.
InvalidInstanceID.NotFoundSteps
Catch the specific error code InvalidInstanceID.NotFound in the loop - not a blanket exception handler. Expected: the miss is logged as already-gone and the loop continues to the next instance.
Refresh inventory immediately before the action pass: describe-instances filtered to instance-state-name of running and pending only. Expected: the action list contains no already-terminated IDs.
After the run, count the NotFound hits. A handful in a busy account is normal; dozens point at a stale cache, a TTL set too long, or a region missing from the scan. Expected: you know whether this was noise or a broken inventory source.
For the misses, optionally re-run describe on those IDs to confirm they are terminated rather than living in another account or region. Expected: confirmed terminated, case closed.
Shrink the scan-to-action gap: run inventory and action in the same job, and shorten any caching between them. Expected: NotFound hits trend toward zero.
Use this when
- A stop or terminate loop logs InvalidInstanceID.NotFound.
- An idle-resource sweep reports failures you cannot reproduce by hand.
- You are reviewing an agent's cleanup run and want to know which errors matter.
Not for this skill when
- The same loop gets UnauthorizedOperation - that is permissions, not staleness.
- You can see the instance in the console but the API cannot - that is the wrong region or wrong credentials.
Variant phrasings
- EC2 stop instances instance not found stale
- InvalidInstanceID.NotFound batch terminate
- Stop-instances loop fails instance already terminated
Why it happens
Scans and actions are never atomic. In any account with autoscaling, CI runners, or other cleanup jobs, instances die in the seconds between your describe call and your stop call. Treating that as an error aborts good runs; treating it as idempotent completion is the correct semantics.
Edge cases
- A typo in an instance ID produces the identical error - log where each ID came from so typos do not hide as staleness.
- Cross-region loops must set the region per call; an ID that is valid in one region is NotFound in another.
- If the same IDs go NotFound on every run, the inventory source is permanently stale - fix the source, not the handler.
Provenance
Resolved from the public thread: https://vectle.com/posts/pstM0ujPUH-spR0h21zViOEg