diff --git a/MANAGED_TARGET.md b/MANAGED_TARGET.md index c8e1f85..85a3768 100644 --- a/MANAGED_TARGET.md +++ b/MANAGED_TARGET.md @@ -39,9 +39,16 @@ old SS1080/Xray were untouched. Historical LIVE observations below are not curre production availability. No network operations or deployment accompany this update. Check-only neither selects/applies strategies nor writes strategy DB entries or -declares full-service READY. Generic periodic controller/config/strategy DB, -plugins/autotuning, maximum strategy collection, safe PID1 BLOCKED stay-alive and -Prometheus metrics remain TODO. Three-CDN discovery/qualification is a YouTube +declares full-service READY. Periodic checks/state/status/reload/independent retry +and generic PID1 BLOCKED foundations now have narrow remote fault acceptance; +unlike the earlier check-only run, this monitor ran without SIGSTOP. +Use `managed monitor --resource-config FILE`, or PID1 environment +`ZAPRET_AUTOTUNE_RESOURCE_CONFIG`; `managed resource-status` reads that same env +source. See [RESOURCE_MONITOR.md](RESOURCE_MONITOR.md) for ownership and repair. +HEALTHY is CHECKS_ONLY, not automatic strategy assignment. Generic strategy DB, +domain qualification matrix, runtime coverage/profile-group rendering and apply, +plugins/autotuning, maximum strategy collection and Prometheus remain TODO. +Three-CDN discovery/qualification is a YouTube adapter contract, not a universal arbitrary-service requirement. ## Managed DoH live acceptance (2026-10-13): DNS verified, NOT READY diff --git a/PROGRESS.md b/PROGRESS.md index 171c62a..4abfe8d 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -1,16 +1,26 @@ # Прогресс реализации -Срез на **2026-10-13**. Этот файл описывает фактическое состояние кода, а не весь объём требований из `Requirements.md` и `Architecture.md`. +Срез на **2026-10-14**. Этот файл описывает фактическое состояние кода, а не весь объём требований из `Requirements.md` и `Architecture.md`. ## Текущий статус +### Generic periodic monitor fault acceptance: narrow PASS, not autotuning/READY + +- Supplied orchestrator-verified privileged read of `/home/agent/zapret2-autotune-monitor-5c85f3f/field-private/report.safe.json` covers `4efc066` plus **`5c85f3f0b9ef04e600f67593d02cae526fac901a`**. Not independently reread here; native full/race checks and independent security review were reported GO. This docs-only update performs no probes/deployment; later docs commits are not tested source. +- Actual periodic monitoring ran **without SIGSTOP**: latest window 21:50:22–21:54:30 retained PID132 and three rows across two services (alpha stable/flaky, beta slow). Healthy rows progressed while flaky was DEGRADED, then repair made all healthy; earlier `4efc066` ran stably for 305s with independent failure/retry and valid reload/removal. +- Snapshot/key inode replacement, wrong UID/mode, symlink, corrupt/missing files and readiness tampering yielded **BLOCKED, stale=true**; private DNS/SNI/server counters were unchanged during faults. Invalid source/wrong ownership made zero DNS requests. Foreign bytes/unknown files were preserved; original-inode/operator repair resumed the same PID. Root wrong-owner field test passed (local unit case skipped). +- Public state was authenticated with a private 32-byte root-owned mode0600 HMAC key; secret-marker/address scans were clean. Idle 30s left content/inode/mtime unchanged. Cancelling an actual active curl left the unfinished row's LastCompleted/Failures/NextDue untouched; completed rows remained durable. +- Graceful stop: **1.45s**, exit0, all owned PIDs joined and network namespace gone; restart PID121 accepted HMAC/prior rows, force-recreate PID126 retained the state-directory inode and accepted HMAC. Additional stops: 0.59s/0.57s. Offline native worst-case cleanup verifies bounded final reaps within 90s, not a measured remote 80s worst-case wait. +- Strategy config/journal hashes were unchanged. Old SS1080, stopped proxy and unrelated stopped RC11082/exit78 were unchanged; legacy YouTube exit78 is **not fixed**. HEALTHY means **CHECKS_ONLY**, not generic strategy recovery, whole-production READY or full P0 acceptance. +- Periodic controller/state/status/reload/independent retry foundations are DONE. Next P0: actual per-domain qualification matrix, then runtime coverage/profile groups, renderer transactions and saving multiple qualified candidates; generic strategy DB/migration/apply remain TODO. Prometheus P1 docs were researched, but no metrics code exists; future labels must stay bounded to configured IDs/enums. + ### Declarative resource-check acceptance (2026-10-13): check-only PASS, not controller acceptance - Supplied orchestrator-verified remote evidence binds source **`4d7c4b33c36186d43ea89fd24354b1834b21692c`** to image **`sha256:be5a1c45c6d40b390a29bfc9a5ab3975c3a8e0c015b98a524696312807dad9a4`**. Private report: `/home/agent/zapret2-autotune-resources-4d7c4b3/field-private/final.safe.json`; privileged-read by the orchestrator, **not independently reread here**. Reported `checker_gates_passed=true`; later documentation commits are not the tested source. No new network checks or deployment accompany this update. - JSON v1 schema/loader and executable HTTPS GET/HEAD checks are unit-verified and now remotely accepted: public GET/HEAD on TLS1.2/1.3; private opt-in plus CIDR with a staging-only trusted test CA, original TLS/SNI hostname and numeric CONNECT pin. No `--insecure`, production trust-store pollution or DoH implementation change. Checks reuse the owned SS/configured wire-DNS backend; the earlier DoH outage acceptance recorded below remains separate evidence. - Default/private-policy, mixed DNS answers, metadata/link-local/ULA/loopback denials made **zero destination connections**; rebinding used the first pinned address and denied the later answer. Ownership-negative cases made zero DNS/resource probes. Cancellation exited1 with joined workers in **0.38s**; the secret sentinel was absent from argv, output and persisted artifacts. - **Check-only isolation:** staging YouTube monitor was SIGSTOP-isolated during checks, then SIGCONT-resumed **before** graceful stop; reported `monitor_resumed=true`, Docker exit0. Cleanup left zero owned host PIDs, no owned namespace or temporary containers/networks, and an unchanged foreign baseline. An earlier existing managed monitor **exit78** contributed before isolation: this does **not** prove whole-controller resilience, generic recovery or production availability. LIVE11082 was already **stopped** before the task and was **not upgraded or restarted**; old SS1080/Xray were untouched. Earlier LIVE snapshots below are historical, not current uptime claims. -- Remaining P0: generic periodic controller integration, config/strategy DB integration, plugins/autotuning. Maximum strategy collection and PID1 safe stay-alive/BLOCKED mode remain TODO; Prometheus is P1 TODO, not an implemented endpoint. Three-CDN qualification belongs to the YouTube adapter, not every arbitrary service. See [MANAGED_TARGET.md](MANAGED_TARGET.md#declarative-https-resource-checks-v1-check-only) and the tracked [resource-check.example.json](resource-check.example.json). +- At this earlier check-only acceptance, periodic integration was pending; the narrow controller/BLOCKED foundations are now accepted above. Remaining P0: config/strategy DB integration, matrix and plugins/autotuning; maximum strategy collection and Prometheus P1 remain TODO. Three-CDN qualification belongs to the YouTube adapter, not every arbitrary service. See [MANAGED_TARGET.md](MANAGED_TARGET.md#declarative-https-resource-checks-v1-check-only) and the tracked [resource-check.example.json](resource-check.example.json). ### Managed DoH acceptance (2026-10-13): live DNS path and staged outage verified, NOT READY diff --git a/RESOURCE_MONITOR.md b/RESOURCE_MONITOR.md index 53037e0..134fb96 100644 --- a/RESOURCE_MONITOR.md +++ b/RESOURCE_MONITOR.md @@ -16,7 +16,11 @@ retry after its cooldown. There is **no strategy selection, recovery or apply**, and no no-strategy two-hour pause. HEALTHY means configured checks passed, not playback health, generic recovery or whole-production READY: HEALTHY is **CHECKS_ONLY**, never READY. The scheduling, status, persistence and lifecycle -base is unit/fixture-verified; isolated remote fault acceptance remains pending. +base passed native full/race checks, independent security review and isolated +remote fault acceptance for `4efc066` plus ownership-status fix `5c85f3f`. +See [PROGRESS.md](PROGRESS.md) for supplied evidence and its narrow scope. +Version1 is bounded to 16 services / 64 resources with stable configured IDs; +valid configuration/state does not imply an automatic strategy assignment. Invalid resource JSON initially or after reload is BLOCKED, with no DNS or transfers until repaired. Bad static target role, proxy or DNS settings still @@ -63,4 +67,6 @@ BLOCKED pending operator repair/restart; no unsafe daemon sourcing is allowed. Generic cleanup shares an 80-second budget, including watchdog setup and command kill escalation, with bounded final reaping; it is designed to fit within a 90-second Docker stop limit (the supplied Compose grants 240 seconds). +Bounded final reaps/worst-case cleanup were verified offline; measured remote +normal stops were 1.45s, 0.59s and 0.57s, not an 80-second runtime worst-case test. These changes do not claim to fix legacy YouTube exit 78.