AIKIDO-2026-513228

TL;DR

The worker generated in define_schedulers.ex runs a trigger's on_error action on a job's final attempt inside a rescue clause. When that action deterministically fails, the rescue re-invokes the same error handler with the unchanged job, and because the call is not in tail position each frame retains a freshly formatted stacktrace. This produces unbounded recursion that consumes CPU, database load, and process memory until the worker is killed, denying service. The fix reraises the error instead of recursively retrying the failing on_error action.