Repository navigation
ateapi crashes actors when atelet rejects a stale snapshot generation #2808
Description
atelet's Checkpoint and PromoteSnapshot return FailedPrecondition when the requested snapshot generation is older than the actor's newest snapshot on the node (or matches it under a different snapshot UUID).
In ateapi, Checkpoint and PromoteSnapshot errors go through handleAteletError (cmd/ateapi/internal/controlapi/workflow_suspend.go), which crashes the actor via ateletErrorCrashesActor (cmd/ateapi/internal/controlapi/crash.go) on every gRPC status code except Unavailable, Canceled, and DeadlineExceeded.
A stale-generation FailedPrecondition only means ateapi's view of the actor's snapshot state is behind atelet's (for example, a late retry of generation N after generation N+1 already committed on atelet), not that the actor itself failed. ateapi should handle that stale-generation rejection without crashing the actor.
Follow-up from #2803 (comment).
Activity
cloudonly commentedon Oct 18, 2026
I'll take this on. I plan to return
Abortedfor stale snapshot generations and same-generation UUID conflicts, and have ateapi report the conflict without crashing the actor. OtherFailedPreconditionerrors will retain their current handling; I'll add regression tests for error propagation and preserved actor/snapshot state.Preserve actors when stale snapshot requests are rejected (agent-subs…