Self-Healing CI: Recovering Flaky DNS Inside a Service Container
A container that briefly cannot resolve a sibling service by name hit an internal-DNS blip, not a wiring mistake -- the next lookup resolves fine.
The problem
A job that talks to a service container by hostname fails with could not resolve host for the service name. The container network and aliases are configured correctly; the embedded DNS that resolves service names had a brief hiccup. A human re-runs the job and the name resolves with no change.
getaddrinfo EAI_AGAIN db ... Temporary failure in name resolution
# resolving the service container alias "db"Why it happens
Container networks resolve service names through an embedded DNS resolver. Early in a job, or under brief load, that resolver can return a temporary failure for a name that is valid and resolves correctly on the next query.
It is a transient internal-DNS blip, not a networking misconfiguration: the alias is correct, the service is up, and the same lookup succeeds moments later.
The manual fix
Manual handling for flaky internal DNS:
- Re-run the job so the lookup is retried.
- Add retry-with-backoff around the first calls that resolve service names.
- Wait for the service to be reachable by name before the dependent step starts.
How this gets automated
An internal name-resolution failure has the same recognizable transient signature as any DNS blip, and the safe response is to wait briefly and retry the lookup. A self-healing CI pipeline detects the resolution failure, retries the step with backoff, and only escalates if the name stays unresolvable, which is the real signal of a misconfigured network rather than a blip.