RunsOn: Spot Interruptions, Auto-Retry, and the Circuit Breaker
RunsOn uses spot instances by default. When AWS reclaims one, you see a CloudTrail BidEvictedEvent; per RunsOn's docs it auto-retries on on-demand once and has a spot circuit breaker.
Per RunsOn's troubleshooting docs, RunsOn runs spot instances by default for cost, and AWS can interrupt them. A reclaimed instance shows up as a CloudTrail BidEvictedEvent; RunsOn automatically retries the job on an on-demand instance once, and a spot circuit breaker steps in when spot capacity is repeatedly unavailable (runs-on.com/docs/maintenance/troubleshooting). Verify current behavior on runs-on.com.
What is happening
- Spot instances can be reclaimed by AWS at any time when capacity or price shifts.
- A reclaim surfaces as a CloudTrail
BidEvictedEventfor the affected instance. - Per RunsOn's docs, the job is automatically retried once on an on-demand instance.
- A spot circuit breaker guards against repeated spot failures.
How to reason about it (per RunsOn docs)
- Expect occasional spot interruptions since spot is the default for cost savings.
- Look for
BidEvictedEventin CloudTrail to confirm an interruption rather than another failure. - Understand that the single on-demand retry is automatic, not something you trigger.
- If spot is chronically unavailable for your instance types, the circuit breaker and your family selection matter.
The managed contrast
This is the classic trade of spot: lowest raw cost in exchange for interruptions you manage. RunsOn softens it with a single on-demand retry and a circuit breaker. Latchkey takes a different approach: it is fully managed and self-healing, so transient and mechanical failures across a job are detected, fixed, and retried automatically rather than relying on a one-shot spot fallback.