Can stale global registrations persist after a node crash in a multi‑node Erlang cluster?

Stale global registrations can persist after a node crash

After a node crash in a multi-node Erlang cluster, a name registered via global:register_name/2 can remain registered on surviving nodes until the global module's background cleanup cycle runs or an operator explicitly unregisters it. The global module does not guarantee immediate removal upon node death.

Likely explanation vs confirmed facts

• The global module runs an internal timer that periodically checks for names whose owning node is no longer alive and purges the entry.
• Cleanup is triggered by the timer and by explicit calls to global:unregister_name/1; there is no automatic immediate removal when a node crashes.

Verification and safe removal steps

1. Query the current state on any surviving node: global:whereis(Name). If it returns undefined, the entry has been purged; otherwise it persists.
2. If the entry persists and you need to reclaim the name, force removal with: global:unregister_name(Name).
3. During a rolling upgrade, monitor the cleanup process or restart the global supervisor on a different node and re-query global:whereis/1 to confirm the name is released.

Uncertainty note: The exact cleanup timing depends on OTP version, system load, and whether network partitions are present. The knowledge brief reports empirical cleanup within 1-10 seconds, but this is not a guaranteed contract.

To refine the recommendation, please confirm: is the global cleanup process currently running on your surviving nodes? (Answering yes or no will determine whether passive waiting or active unregistration is the safer next step.)