Travis CI Retry Boundary with Deployment Writes: Preventing Duplicate Side Effects

In current Travis CI versions, retries re-execute the entire job, which means any database insert, artifact upload, or deployment call performed by the build script runs again from the start. The platform does not provide a built-in mechanism to suppress duplicate writes at the integration boundary between the CI runner and downstream services. While environment metadata such as TRAVIS_RETRY_NUMBER is available to the job, the architectural question of where idempotency should be enforced remains unresolved.

The uncertainty is amplified when a retry overlaps with status webhooks or when downstream pipelines receive concurrent triggers from the same commit. Because Travis CI does not guarantee that a retried build reaches the downstream component only once, teams must decide whether the CI configuration should gate the write or whether the downstream service should accept a deduplication token. This division affects reproducibility across Travis CI tiers, where retry timeouts and resource quotas may differ.

Should the idempotency key be generated inside the Travis CI job and forwarded to the deployment target, or should the downstream service derive uniqueness from build identifiers such as commit SHA and job number? How can concurrent retry attempts and webhook notifications be coordinated so that downstream state changes occur exactly once? What is the safest way to use TRAVIS_RETRY_NUMBER to skip or transform state-altering operations without breaking legitimate redeployments?