Yesterday a tiny startup saw its production app disappear after DigitalOcean rolled out a managed PostgreSQL update. The change broke private VPC connectivity to the provider’s managed Kubernetes clusters, leaving the public endpoint reachable while the private one timed out. Engineers traced the failure to a Cilium bug (issue #34503) that lets ARP entries go stale after infrastructure modifications.

Support responded within twelve hours and deployed a community‑authored DaemonSet that pings stale ARP entries every ten seconds. An upstream fix has been merged but has not yet reached DOKS, and DigitalOcean gave no deployment ETA. The outage forced the founders to spend late‑night hours debugging a networking layer they do not control, contradicting their expectation that “managed” equals hands‑off reliability.

Their experience underscores a broader truth: outsourcing infrastructure swaps one set of failure modes for another, not eliminates them. Despite the incident, the team remains on DigitalOcean, now with a clearer view of what “managed” really delivers.