docs: record bounded recovery and release acceptance
Co-authored-by: GPT-6 <gpt-6@openai.com>
This commit is contained in:
1 parent
f5a3c67a68
commit
226d93e2aa
1 file changed
+5
@@ -4,6 +4,11 @@
|
||||
|
||||
## Готово
|
||||
|
||||
- **2026-10-11 production-candidate update (remote checks dated 2026-10-10 UTC):** `ed4298d` adds one 90-minute budget per managed bootstrap/recovery job, shared across phases. Watchdog expiration is abnormal/fatal, never ordinary nonpass/cooldown continuation; clean timeout permits technical exit 75 only after worker join and bounded 30-second restoration. Any guard/DB/rollback/cleanup or mixed unknown error vetoes restart (78); configuration errors use 64, graceful shutdown 0. Independent review reproduced and closed mixed-timeout, fatal-discovery and fatal-health/deadline races. Preferred frontend seed is now the layout-safe alternative verified in prior 60/60 requests; cached definitions are not rewritten.
|
||||
- **2026-10-10 restart/startup hardening:** optional `deploy/managed-target/zapret-autotune.service` uses foreground Compose, no blanket Docker restart, systemd allowlisted 75 with bounded starts (3 per 6h); shutdown grace 120s / unit stop budget 180s. `61e6e00` adds guarded fixed-DB `managed reconcile` before initial activation; init/reconcile/start are bounded (45/45/30s plus 5s kill grace), all startup refusals map to 78. Unknown state is not sourced even by cleanup. Actual systemd tests: technical 75 retries bounded, 78 does not retry; killing monitor in final image caused one whole-target restart, preserving both groups and Player API OK. Permanent production service was not installed/enabled; transient acceptance units were stopped.
|
||||
- **2026-10-10 interrupted-apply acceptance:** real isolated fixtures from SQLite backup/qualified cached rows tested final image. Known prepared config (pending hash `507b2d…`) was restored to committed `1f9696…` before first real sysv restart/start; journal at first start was committed with old IDs [1,15]. Unknown config hash exited 78 with zero activations/restarts and preserved diagnostic state. No qualification evidence was fabricated; no main state was modified. Private report: `/home/agent/prepared-acceptance-g2p72yjt/acceptance-report.json`; disposable containers cleaned after validation.
|
||||
- **2026-10-10 build/docs/result:** Go builder and runtime are digest-pinned; `4824570` adds native BuildKit compilation-cache mounts. Built release candidate `zapret-autotune-managed-target:rc-4824570`, local immutable image ID `sha256:011b8e82422a83686322eafddbd4203f9117bf4d9830fa3ec900f0f31538c1ae`. Atomically persisted only its selector in private remote `.env`; image binding, env0600 and SS password preservation verified without printing credentials. New target runs this image, generation29 active[1,15], Player API OK, only queue300; old container PID/start-time unchanged, Xray untouched. README, MANAGED_TARGET and deployment README cover provisioning, actual config forwarding, immutable promotion, service install, backup/restore, failure handling and limitations. `go test ./...`, affected `-race`, shell/static checks and native unit syntax passed. No real server reboot or full 90-minute expiry wait was performed; expiry tested with shortened injected budgets. Registry publication, permanent service enablement and final production rollout remain operator decisions; this is a production candidate, not a blanket production-ready claim.
|
||||
|
||||
- **2026-10-10 transport RCA:** captured packet/strace diagnostics only in disposable containers on the target bridge, fixed destination `142.251.153.4`. TCP establishment succeeds (~18ms); failed passes retransmit ClientHello without server ACK advancement or first TLS response, not an HTTP-body timeout. Full-snapshot, conflict-checked reassembly of 24 connections showed the old frontend seed passes all 22 non-final-SNI layouts and times out on both final-SNI layouts (one TLS1.2, one TLS1.3; zero bytes after SNI). Rustls extension order changes between requests. This strongly establishes observed strategy layout sensitivity, not a proven Rustls/Lua bug or exact network drop location. An independent heavy-tier review confirmed the conclusion/limits.
|
||||
- **2026-10-10 replacement candidate:** `--payload=tls_client_hello --lua-desync=fake:blob=fake_default_tls:tcp_ts=-1000:repeats=1 --lua-desync=multidisorder:pos=1,midsld` passed 24/24 diagnostic connections, including final-SNI in both protocols, then 10 standard checker jobs with 3/3 for TLS1.2 and TLS1.3 (60/60 requests using auto DNS). This is a tested replacement hypothesis, not a universal guarantee. Qualification/evidence rules are unchanged. No working profile/image was switched during this tracing task; target Player API remains OK, old target/Xray unchanged. Temporary trace containers were removed after copying private artifacts; raw captures/reports retained in `/home/agent/checker-transport-trace`, `/home/agent/checker-full-trace`, `/home/agent/checker-alternative-trace`, and `/home/agent/alternative-qualification-*.json/.stderr`. Next implementation step, if approved: prefer this candidate through the normal managed verify/save/apply pipeline instead of weakening checks or modifying TLS libraries/timeouts.
|
||||
|
||||
|
||||
Reference in new issue
Block a user