Share one pure provider selection plan between real unrestrict routing and cooldown projection, including direct 1Fichier and DDownload paths, disabled Mega aliases, and the secondary-provider exception used when Real-Debrid is cooling down. Attach terminal observers to raw serialized web jobs before evaluating already-aborted signals so late rejections stay handled while the queue remains available. Add RED-to-GREEN regressions for each provider-selection counterexample and pre-aborted Real-Debrid, AllDebrid, and BestDebrid jobs.
Advance each serialized web-provider queue when an aborted caller is released while retaining terminal observation of the underlying request. Preserve an accepted pending start across repeated stop requests and dispatch it exactly once after drain. Derive provider retry deadlines from eligible queued items, configured fallback chains, enabled accounts and keys, and matching provider or hoster cooldowns. Add RED-to-GREEN coverage for queue progress, repeated stop behavior, disabled accounts, alternative providers, unrelated hosters, and post-processing-only state.
Release aborted Real-Debrid, AllDebrid, and BestDebrid web callers even when underlying requests ignore cancellation while retaining terminal rejection observers. Publish provider cooldown deadlines and emit a fresh idle snapshot at the earliest expiry. Abort and visibly drain post-processing before dispatching one pending restart. Surface lifecycle phase, reason, retry countdown, and remaining work in the download controls. Add focused regressions for provider abort races, cooldown expiry, post-processing drain, pending start visibility, and the updated rapid stop contract.
Reserve the actively delivered outbox event while the external sender is pending so concurrent persistence cannot evict it through expiry or capacity enforcement. Apply the delivery outcome under the exclusive chain before normal cleanup resumes, preserving retry state, delivery acknowledgements, and cooldown timing without duplicate sends. Add deterministic concurrency regressions for expiry-success and capacity-failure races.
Enforce one shared shutdown deadline while persisting late digests during blocked sends. Start stall cooldowns from serialized delivery acknowledgements, atomically rewrite cleaned legacy outboxes, retain run summaries for 24 hours, and prevent duplicate history or synthetic post-process start times.
Bind the diagnostics flow to bearer-only authentication and reject query token attempts with controlled responses. Compare bearer token bytes with timing-safe equality, remove wildcard CORS, and mark diagnostics/support responses as no-store.
Move trace configuration mutation behind POST, add method failure handling, add a small per-IP/loopback in-memory request limit, and keep generated setup and support-manifest URLs token-free while pointing support access at a local bridge/tunnel flow.
Sanitize backup remote diagnostics on export and restore so legacy token, endpoint, host mode, and port values are not persisted; restores only keep the allowlist and force local binding.
Tests cover bearer accept/reject, query rejection, GET mutation rejection, no-store/CORS behavior, loopback default binding, rate limiting, token-free hints, and backup sanitation.
Sucukdeluxe