From 08c67bde1aed718a8df86dcd0f7ef9c2bd6c83cd Mon Sep 17 00:00:00 2001
From: igor04091968 <igor04091968@gmail.com>
Date: Fri, 3 Jul 2026 08:51:11 +0300
Subject: [PATCH] Add DetMir primary recovery guard

---
 ansible/templates/proxmox-web-gateway.conf.j2 |   6 +-
 docs/DAILY_MAINTENANCE_RU.md                  |  19 +-
 docs/DETMIR_SERVICE_RELIABILITY_RUNBOOK_RU.md | 233 +++++++++++
 docs/OPERATIONS_RUNBOOK_RU.md                 |  18 +
 .../detmir-aw-primary-recovery.service        |  18 +
 ops/systemd/detmir-aw-primary-recovery.timer  |  13 +
 scripts/detmir-aw-primary-recovery.sh         | 367 ++++++++++++++++++
 7 files changed, 672 insertions(+), 2 deletions(-)
 create mode 100644 docs/DETMIR_SERVICE_RELIABILITY_RUNBOOK_RU.md
 create mode 100644 ops/systemd/detmir-aw-primary-recovery.service
 create mode 100644 ops/systemd/detmir-aw-primary-recovery.timer
 create mode 100755 scripts/detmir-aw-primary-recovery.sh

diff --git a/ansible/templates/proxmox-web-gateway.conf.j2 b/ansible/templates/proxmox-web-gateway.conf.j2
index 3910b8a..ffa0ce7 100644
--- a/ansible/templates/proxmox-web-gateway.conf.j2
+++ b/ansible/templates/proxmox-web-gateway.conf.j2
@@ -250,6 +250,10 @@ server {
     }
 
 {% for route in proxmox_web_gateway_routes %}
+    location = /go/{{ route.slug }} {
+        return 302 {{ route.target_url }};
+    }
+
 {% if route.proxy_target_url is defined %}
     location = {{ route.proxy_path }} {
         proxy_set_header Authorization "";
@@ -274,7 +278,7 @@ server {
     location ^~ /portal/api/readiness {
         proxy_set_header Authorization "";
         proxy_set_header X-Remote-User $remote_user;
-        proxy_pass http://192.0.2.13:8721/api/readiness;
+        proxy_pass http://127.0.0.1:8720/readyz;
         proxy_redirect off;
     }
 
diff --git a/docs/DAILY_MAINTENANCE_RU.md b/docs/DAILY_MAINTENANCE_RU.md
index 08f28b7..1c02fdd 100644
--- a/docs/DAILY_MAINTENANCE_RU.md
+++ b/docs/DAILY_MAINTENANCE_RU.md
@@ -1,6 +1,6 @@
 # Ежедневное обслуживание AWatch-rus
 
-Дата актуализации: 2026-07-02
+Дата актуализации: 2026-07-03
 
 Документ описывает ежедневный операторский цикл обслуживания AWatch-rus /
 DetMir и безопасное использование Pollinations AI как вспомогательного
@@ -87,6 +87,22 @@ node scripts/deployment-readiness-smoke.mjs
 Критерий: нет новых hard-fail проверок. Warning допустим только при известном
 и задокументированном operational constraint.
 
+### 3.1. Primary recovery guard
+
+Проверить, что автоматический recovery первичного AW API включен и не скрывает
+новые incident:
+
+```bash
+systemctl is-active detmir-aw-primary-recovery.timer
+systemctl status detmir-aw-primary-recovery.timer --no-pager
+sudo /usr/local/bin/detmir-aw-primary-recovery --check-only
+sudo jq . /var/lib/detmir-aw-primary-recovery/latest.json 2>/dev/null || true
+```
+
+Критерий: timer активен, `--check-only` возвращает `status=ok`, новые incident
+не имеют `outcome=failed`. Если latest incident отсутствует, это нормально для
+контура, где recovery еще ни разу не требовался.
+
 ### 4. Очереди, backlog и storage growth
 
 Проверить отсутствие неконтролируемого роста:
@@ -305,6 +321,7 @@ unset ALL_PROXY all_proxy
 
 - [Operations Runbook](OPERATIONS_RUNBOOK_RU.md)
 - [Эксплуатационная проверка контура](OPERATIONS_VALIDATION_RUNBOOK_RU.md)
+- [DetMir service reliability runbook](DETMIR_SERVICE_RELIABILITY_RUNBOOK_RU.md)
 - [Retention and Cleanup Policy](RETENTION_POLICY_RU.md)
 - [Production readiness](PRODUCTION_READINESS_RU.md)
 - [Autonomous validation](AUTONOMOUS_VALIDATION_RU.md)
diff --git a/docs/DETMIR_SERVICE_RELIABILITY_RUNBOOK_RU.md b/docs/DETMIR_SERVICE_RELIABILITY_RUNBOOK_RU.md
new file mode 100644
index 0000000..81f589e
--- /dev/null
+++ b/docs/DETMIR_SERVICE_RELIABILITY_RUNBOOK_RU.md
@@ -0,0 +1,233 @@
+# DetMir Service Reliability Runbook
+
+Дата актуализации: 2026-07-03
+
+Документ описывает production-safe контур снижения повторных выпадений сервисов
+AWatch-rus/DetMir. Главная цель — отделять первичный отказ ActivityWatch API от
+вторичных падений ingest/report/health jobs и выполнять ограниченное
+автовосстановление только для подтвержденного безопасного сценария.
+
+## План реализации
+
+1. Считать `activitywatch-server` первичным сервисом для AW API.
+2. Считать `aw-workforce-ingest`, `aw-worktime-autoheal`, `aw-worktime-prewarm`,
+   `aw-worktime-ui-bridge`, `aw-worktime-influx-exporter`, `aw-rus-healthd` и
+   `aw-slo-monitor` вторичными зависимыми jobs.
+3. Проверять AW API через `/api/0/settings/` и `/api/0/buckets/`.
+4. Автоматически восстанавливать только подтвержденный класс отказа:
+   `poisoned datastore lock` или неактивный `activitywatch-server`.
+5. Перед restart первичного сервиса временно остановить вторичные jobs, чтобы
+   они не усиливали нагрузку на AW SQLite.
+6. Перезапустить только `activitywatch-server` внутри CT `203`.
+7. Дождаться восстановления AW API.
+8. Вернуть вторичные timers и записать incident JSON.
+
+## Что реализовано
+
+Реализован script:
+
+```text
+scripts/detmir-aw-primary-recovery.sh
+```
+
+Systemd units для Proxmox host:
+
+```text
+ops/systemd/detmir-aw-primary-recovery.service
+ops/systemd/detmir-aw-primary-recovery.timer
+```
+
+Контур предназначен для запуска на Proxmox host `10.10.10.2`, потому что там
+доступны одновременно:
+
+- управление PVE jobs через локальный `systemctl`;
+- управление AW CT через `pct exec 203`;
+- проверка внешнего AW API `http://10.10.10.13:5600`.
+
+## Safety Rules
+
+Recovery script:
+
+- не удаляет SQLite database, lock files или journal files;
+- не включает DLP, Loki или always-on Velociraptor;
+- не перезапускает Windows host и Windows collectors;
+- не меняет конфигурацию AW server;
+- не выполняет recovery при произвольном HTTP timeout без подтвержденной
+  причины;
+- имеет cooldown между restart попытками;
+- пишет incident evidence в JSON.
+
+## Установка
+
+На Proxmox host:
+
+```bash
+sudo install -m 0755 scripts/detmir-aw-primary-recovery.sh \
+  /usr/local/bin/detmir-aw-primary-recovery
+
+sudo install -m 0644 ops/systemd/detmir-aw-primary-recovery.service \
+  /etc/systemd/system/detmir-aw-primary-recovery.service
+
+sudo install -m 0644 ops/systemd/detmir-aw-primary-recovery.timer \
+  /etc/systemd/system/detmir-aw-primary-recovery.timer
+
+sudo systemctl daemon-reload
+sudo systemctl enable --now detmir-aw-primary-recovery.timer
+```
+
+Опциональная конфигурация:
+
+```bash
+sudo install -d -m 0755 /etc/detmir
+sudoedit /etc/detmir/aw-primary-recovery.env
+```
+
+Пример:
+
+```bash
+DETMIR_AW_RECOVERY_URL=http://10.10.10.13:5600
+DETMIR_AW_RECOVERY_CT_ID=203
+DETMIR_AW_RECOVERY_CONFIRM_ATTEMPTS=2
+DETMIR_AW_RECOVERY_COOLDOWN_SECONDS=900
+DETMIR_AW_RECOVERY_HTTP_TIMEOUT_SECONDS=8
+```
+
+## Проверка без изменений
+
+```bash
+/usr/local/bin/detmir-aw-primary-recovery --self-test
+/usr/local/bin/detmir-aw-primary-recovery --check-only
+systemctl start detmir-aw-primary-recovery.service
+journalctl -u detmir-aw-primary-recovery.service -n 80 --no-pager
+```
+
+Ожидаемо для здорового контура:
+
+```text
+event=probe attempt=1 status=ok
+```
+
+## Evidence
+
+Incident files:
+
+```text
+/var/lib/detmir-aw-primary-recovery/incidents/*.json
+/var/lib/detmir-aw-primary-recovery/latest.json
+/var/lib/detmir-aw-primary-recovery/status.json
+```
+
+Проверка последнего incident:
+
+```bash
+sudo jq . /var/lib/detmir-aw-primary-recovery/latest.json
+```
+
+Ключевые поля:
+
+- `outcome`: `recovered`, `failed`, `skipped`, `observed_no_action`, `dry_run`;
+- `reason`: `poisoned_lock`, `service_inactive`, `cooldown_active_after_*`;
+- `recovered`: boolean;
+- `pve_pause_units`;
+- `ct_pause_units`;
+- `details`.
+
+## Rollback
+
+Отключить automation:
+
+```bash
+sudo systemctl disable --now detmir-aw-primary-recovery.timer
+sudo systemctl reset-failed detmir-aw-primary-recovery.service
+```
+
+Удаление установленных файлов не требуется для rollback. Если нужно убрать
+полностью:
+
+```bash
+sudo rm -f /etc/systemd/system/detmir-aw-primary-recovery.service
+sudo rm -f /etc/systemd/system/detmir-aw-primary-recovery.timer
+sudo rm -f /usr/local/bin/detmir-aw-primary-recovery
+sudo systemctl daemon-reload
+```
+
+## Manual Recovery Sequence
+
+Если automation disabled или recovery не помог:
+
+```bash
+sudo systemctl stop aw-workforce-ingest.timer aw-workforce-ingest.service
+sudo pct exec 203 -- systemctl stop \
+  aw-worktime-autoheal.timer aw-worktime-autoheal.service \
+  aw-worktime-prewarm.timer aw-worktime-prewarm.service \
+  aw-worktime-ui-bridge.timer aw-worktime-ui-bridge.service \
+  aw-worktime-influx-exporter.timer aw-worktime-influx-exporter.service \
+  aw-rus-healthd.timer aw-rus-healthd.service \
+  aw-slo-monitor.timer aw-slo-monitor.service
+
+sudo pct exec 203 -- systemctl restart activitywatch-server
+
+curl -fsS http://10.10.10.13:5600/api/0/settings/ >/dev/null
+curl -fsS http://10.10.10.13:5600/api/0/buckets/ >/dev/null
+
+sudo pct exec 203 -- systemctl start \
+  aw-worktime-autoheal.timer \
+  aw-worktime-prewarm.timer \
+  aw-worktime-ui-bridge.timer \
+  aw-worktime-influx-exporter.timer \
+  aw-rus-healthd.timer \
+  aw-slo-monitor.timer
+
+sudo systemctl start aw-workforce-ingest.timer
+```
+
+## Post-Recovery Validation
+
+```bash
+cd /mnt/usb_hdd2/Projects/ActivityWatch-Russian
+./check-aw-full.sh
+
+ssh igor@10.10.10.2 'systemctl --failed --no-legend || true'
+ssh igor@10.10.10.2 'sudo pct exec 203 -- systemctl --failed --no-legend || true'
+
+curl -fsS http://10.10.10.13:5610/health | jq .
+curl -fsS http://10.10.10.2:8720/readyz | jq .
+```
+
+Критерий:
+
+- `check-aw-full.sh`: `DEAD=0`, `STALE=0`;
+- failed units на Proxmox host: `0`;
+- failed units внутри CT `203`: `0`;
+- Worktime health: `status=OK`;
+- Portal readiness: `status=ready`.
+
+## Gateway Contract
+
+Операторские shortcuts должны соответствовать runbook:
+
+```text
+/go/proxmox-gui
+/go/file1c-brief
+/go/file1c-actions
+```
+
+Они формируются из `proxmox_web_gateway_routes` в nginx template и должны
+возвращать `302` на целевой внутренний URL после Basic Auth.
+
+Readiness compatibility endpoint:
+
+```text
+/portal/api/readiness -> http://127.0.0.1:8720/readyz
+```
+
+## Known Limitations
+
+- Recovery intentionally limited to AW API poisoned lock and inactive primary
+  service. Other failures remain operator-visible incidents.
+- If restart does not restore AW API before timeout, secondary timers remain
+  paused to avoid failure amplification. Operator must inspect
+  `/var/lib/detmir-aw-primary-recovery/latest.json`.
+- Windows credential or WinRM authentication problems are outside this recovery
+  loop.
+- This does not prove backup/restore. It is runtime availability recovery only.
diff --git a/docs/OPERATIONS_RUNBOOK_RU.md b/docs/OPERATIONS_RUNBOOK_RU.md
index 8e53755..a00b1b9 100644
--- a/docs/OPERATIONS_RUNBOOK_RU.md
+++ b/docs/OPERATIONS_RUNBOOK_RU.md
@@ -6,6 +6,10 @@
 использование Pollinations AI для анализа sanitized evidence описаны отдельно:
 [ежедневное обслуживание AWatch-rus](DAILY_MAINTENANCE_RU.md).
 
+Автоматическое восстановление первичного AW API от подтвержденного
+`poisoned datastore lock` описано отдельно:
+[DetMir service reliability runbook](DETMIR_SERVICE_RELIABILITY_RUNBOOK_RU.md).
+
 ## Быстрая проверка
 
 Проверить доступность:
@@ -108,6 +112,20 @@ systemctl --failed --no-pager
 - coverage;
 - errors in logs.
 
+### ActivityWatch API возвращает `503 poisoned lock`
+
+Проверить, сработал ли primary recovery guard:
+
+```bash
+systemctl status detmir-aw-primary-recovery.timer --no-pager
+journalctl -u detmir-aw-primary-recovery.service -n 80 --no-pager
+sudo jq . /var/lib/detmir-aw-primary-recovery/latest.json
+```
+
+Если guard отключен, использовать manual sequence из
+`DETMIR_SERVICE_RELIABILITY_RUNBOOK_RU.md`. Не удалять SQLite, lock или journal
+файлы ActivityWatch вручную.
+
 ### Нет данных
 
 Проверить:
diff --git a/ops/systemd/detmir-aw-primary-recovery.service b/ops/systemd/detmir-aw-primary-recovery.service
new file mode 100644
index 0000000..8050d47
--- /dev/null
+++ b/ops/systemd/detmir-aw-primary-recovery.service
@@ -0,0 +1,18 @@
+[Unit]
+Description=DetMir ActivityWatch primary recovery guard
+Documentation=https://github.com/igor04091968/AWatch-rus/blob/main/docs/DETMIR_SERVICE_RELIABILITY_RUNBOOK_RU.md
+Wants=network-online.target
+After=network-online.target
+ConditionPathExists=/usr/sbin/pct
+
+[Service]
+Type=oneshot
+EnvironmentFile=-/etc/detmir/aw-primary-recovery.env
+ExecStart=/usr/local/bin/detmir-aw-primary-recovery --once
+User=root
+Group=root
+Nice=5
+TimeoutStartSec=5min
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=detmir-aw-primary-recovery
diff --git a/ops/systemd/detmir-aw-primary-recovery.timer b/ops/systemd/detmir-aw-primary-recovery.timer
new file mode 100644
index 0000000..46a5fcc
--- /dev/null
+++ b/ops/systemd/detmir-aw-primary-recovery.timer
@@ -0,0 +1,13 @@
+[Unit]
+Description=Run DetMir ActivityWatch primary recovery guard every 2 minutes
+
+[Timer]
+OnBootSec=3min
+OnUnitActiveSec=2min
+AccuracySec=30s
+RandomizedDelaySec=15s
+Persistent=false
+Unit=detmir-aw-primary-recovery.service
+
+[Install]
+WantedBy=timers.target
diff --git a/scripts/detmir-aw-primary-recovery.sh b/scripts/detmir-aw-primary-recovery.sh
new file mode 100755
index 0000000..8541699
--- /dev/null
+++ b/scripts/detmir-aw-primary-recovery.sh
@@ -0,0 +1,367 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# Production-safe primary recovery loop for DetMir ActivityWatch.
+#
+# Scope:
+# - Detects ActivityWatch API poisoned datastore lock.
+# - Pauses dependent secondary jobs before restarting the primary AW service.
+# - Restarts only activitywatch-server inside the configured Proxmox CT.
+# - Writes structured incident evidence.
+#
+# Non-goals:
+# - Does not delete SQLite, journal or lock files.
+# - Does not enable DLP, Loki or Velociraptor.
+# - Does not restart Windows collectors.
+
+AW_URL="${DETMIR_AW_RECOVERY_URL:-http://10.10.10.13:5600}"
+AW_CT_ID="${DETMIR_AW_RECOVERY_CT_ID:-203}"
+AW_SERVICE="${DETMIR_AW_RECOVERY_SERVICE:-activitywatch-server}"
+STATE_DIR="${DETMIR_AW_RECOVERY_STATE_DIR:-/var/lib/detmir-aw-primary-recovery}"
+HTTP_TIMEOUT_SECONDS="${DETMIR_AW_RECOVERY_HTTP_TIMEOUT_SECONDS:-8}"
+CONFIRM_ATTEMPTS="${DETMIR_AW_RECOVERY_CONFIRM_ATTEMPTS:-2}"
+CONFIRM_SLEEP_SECONDS="${DETMIR_AW_RECOVERY_CONFIRM_SLEEP_SECONDS:-5}"
+STARTUP_TIMEOUT_SECONDS="${DETMIR_AW_RECOVERY_STARTUP_TIMEOUT_SECONDS:-120}"
+STARTUP_SLEEP_SECONDS="${DETMIR_AW_RECOVERY_STARTUP_SLEEP_SECONDS:-5}"
+COOLDOWN_SECONDS="${DETMIR_AW_RECOVERY_COOLDOWN_SECONDS:-900}"
+DRY_RUN="${DETMIR_AW_RECOVERY_DRY_RUN:-0}"
+
+PVE_PAUSE_UNITS="${DETMIR_AW_RECOVERY_PVE_PAUSE_UNITS:-aw-workforce-ingest.timer aw-workforce-ingest.service}"
+PVE_RESUME_UNITS="${DETMIR_AW_RECOVERY_PVE_RESUME_UNITS:-aw-workforce-ingest.timer}"
+CT_PAUSE_UNITS="${DETMIR_AW_RECOVERY_CT_PAUSE_UNITS:-aw-worktime-autoheal.timer aw-worktime-autoheal.service aw-worktime-prewarm.timer aw-worktime-prewarm.service aw-worktime-ui-bridge.timer aw-worktime-ui-bridge.service aw-worktime-influx-exporter.timer aw-worktime-influx-exporter.service aw-rus-healthd.timer aw-rus-healthd.service aw-slo-monitor.timer aw-slo-monitor.service}"
+CT_RESUME_UNITS="${DETMIR_AW_RECOVERY_CT_RESUME_UNITS:-aw-worktime-autoheal.timer aw-worktime-prewarm.timer aw-worktime-ui-bridge.timer aw-worktime-influx-exporter.timer aw-rus-healthd.timer aw-slo-monitor.timer}"
+
+usage() {
+  cat <<'EOF'
+Usage:
+  detmir-aw-primary-recovery.sh [--once|--check-only|--self-test]
+
+Environment:
+  DETMIR_AW_RECOVERY_URL                  default: http://10.10.10.13:5600
+  DETMIR_AW_RECOVERY_CT_ID                default: 203
+  DETMIR_AW_RECOVERY_SERVICE              default: activitywatch-server
+  DETMIR_AW_RECOVERY_STATE_DIR            default: /var/lib/detmir-aw-primary-recovery
+  DETMIR_AW_RECOVERY_CONFIRM_ATTEMPTS     default: 2
+  DETMIR_AW_RECOVERY_COOLDOWN_SECONDS     default: 900
+  DETMIR_AW_RECOVERY_DRY_RUN              default: 0
+  DETMIR_AW_RECOVERY_PVE_PAUSE_UNITS      space-separated units on Proxmox host
+  DETMIR_AW_RECOVERY_PVE_RESUME_UNITS     space-separated units on Proxmox host
+  DETMIR_AW_RECOVERY_CT_PAUSE_UNITS       space-separated units inside AW CT
+  DETMIR_AW_RECOVERY_CT_RESUME_UNITS      space-separated units inside AW CT
+
+The script never removes ActivityWatch SQLite, lock or journal files.
+EOF
+}
+
+log() {
+  printf 'ts=%s component=detmir-aw-primary-recovery %s\n' "$(date -u +%Y-%m-%dT%H:%M:%SZ)" "$*"
+}
+
+require_command() {
+  local cmd="$1"
+  if ! command -v "$cmd" >/dev/null 2>&1; then
+    log "level=error event=missing_command command=$cmd"
+    exit 127
+  fi
+}
+
+json_string_array() {
+  jq -Rsc 'split(" ") | map(select(length > 0))' <<<"$1"
+}
+
+record_incident() {
+  local outcome="$1"
+  local reason="$2"
+  local recovered="$3"
+  local started_at="$4"
+  local finished_at="$5"
+  local details="$6"
+  local incident_id
+  incident_id="$(date -u +%Y%m%dT%H%M%SZ)-$$"
+  local incident_path="${STATE_DIR}/incidents/${incident_id}.json"
+  install -d -m 0750 "${STATE_DIR}/incidents"
+  jq -n \
+    --arg generated_at_utc "$finished_at" \
+    --arg started_at_utc "$started_at" \
+    --arg outcome "$outcome" \
+    --arg reason "$reason" \
+    --arg recovered "$recovered" \
+    --arg aw_url "$AW_URL" \
+    --arg aw_ct_id "$AW_CT_ID" \
+    --arg aw_service "$AW_SERVICE" \
+    --arg details "$details" \
+    --argjson pve_pause_units "$(json_string_array "$PVE_PAUSE_UNITS")" \
+    --argjson ct_pause_units "$(json_string_array "$CT_PAUSE_UNITS")" \
+    '{
+      generated_at_utc: $generated_at_utc,
+      started_at_utc: $started_at_utc,
+      outcome: $outcome,
+      reason: $reason,
+      recovered: ($recovered == "true"),
+      aw_url: $aw_url,
+      aw_ct_id: ($aw_ct_id | tonumber? // $aw_ct_id),
+      aw_service: $aw_service,
+      details: $details,
+      pve_pause_units: $pve_pause_units,
+      ct_pause_units: $ct_pause_units
+    }' >"$incident_path"
+  ln -sfn "$incident_path" "${STATE_DIR}/latest.json"
+  log "level=info event=incident_written path=$incident_path outcome=$outcome reason=$reason recovered=$recovered"
+}
+
+classify_http_response() {
+  local code="$1"
+  local body_file="$2"
+  if [[ "$code" =~ ^2[0-9][0-9]$ ]]; then
+    printf 'ok'
+    return 0
+  fi
+  if grep -Eiq 'poisoned lock|Taking datastore lock failed|datastore lock failed' "$body_file"; then
+    printf 'poisoned_lock'
+    return 0
+  fi
+  printf 'http_%s' "$code"
+}
+
+probe_path() {
+  local path="$1"
+  local body_file
+  body_file="$(mktemp)"
+  local code rc
+  code="$(curl -sS --max-time "$HTTP_TIMEOUT_SECONDS" -o "$body_file" -w '%{http_code}' "${AW_URL}${path}" 2>/dev/null)" || rc=$?
+  rc="${rc:-0}"
+  if [[ "$rc" -ne 0 ]]; then
+    rm -f "$body_file"
+    printf 'timeout'
+    return 0
+  fi
+  local status
+  status="$(classify_http_response "$code" "$body_file")"
+  rm -f "$body_file"
+  printf '%s' "$status"
+}
+
+ct_systemctl() {
+  pct exec "$AW_CT_ID" -- systemctl "$@"
+}
+
+activitywatch_service_status() {
+  if ! pct status "$AW_CT_ID" >/dev/null 2>&1; then
+    printf 'ct_unavailable'
+    return 0
+  fi
+  if ct_systemctl is-active --quiet "$AW_SERVICE"; then
+    printf 'active'
+  else
+    printf 'inactive'
+  fi
+}
+
+probe_aw() {
+  local service_state settings_status buckets_status
+  service_state="$(activitywatch_service_status)"
+  if [[ "$service_state" != "active" ]]; then
+    printf 'service_%s' "$service_state"
+    return 0
+  fi
+  settings_status="$(probe_path '/api/0/settings/')"
+  buckets_status="$(probe_path '/api/0/buckets/')"
+  if [[ "$settings_status" == "ok" && "$buckets_status" == "ok" ]]; then
+    printf 'ok'
+  elif [[ "$settings_status" == "poisoned_lock" || "$buckets_status" == "poisoned_lock" ]]; then
+    printf 'poisoned_lock'
+  else
+    printf '%s,%s' "$settings_status" "$buckets_status"
+  fi
+}
+
+stop_units_on_host() {
+  local units="$1"
+  local unit
+  for unit in $units; do
+    log "level=info event=stop_unit scope=pve unit=$unit"
+    systemctl stop "$unit" >/dev/null 2>&1 || true
+  done
+}
+
+start_units_on_host() {
+  local units="$1"
+  local unit
+  for unit in $units; do
+    log "level=info event=start_unit scope=pve unit=$unit"
+    systemctl start "$unit" >/dev/null 2>&1 || true
+  done
+}
+
+stop_units_in_ct() {
+  local units="$1"
+  local unit
+  for unit in $units; do
+    log "level=info event=stop_unit scope=ct ct=$AW_CT_ID unit=$unit"
+    ct_systemctl stop "$unit" >/dev/null 2>&1 || true
+  done
+}
+
+start_units_in_ct() {
+  local units="$1"
+  local unit
+  for unit in $units; do
+    log "level=info event=start_unit scope=ct ct=$AW_CT_ID unit=$unit"
+    ct_systemctl start "$unit" >/dev/null 2>&1 || true
+  done
+}
+
+cooldown_active() {
+  local last_file="${STATE_DIR}/last_restart_epoch"
+  [[ -f "$last_file" ]] || return 1
+  local now last
+  now="$(date -u +%s)"
+  last="$(cat "$last_file" 2>/dev/null || printf '0')"
+  [[ "$last" =~ ^[0-9]+$ ]] || return 1
+  (( now - last < COOLDOWN_SECONDS ))
+}
+
+wait_for_aw_ok() {
+  local deadline now status
+  deadline=$(( $(date -u +%s) + STARTUP_TIMEOUT_SECONDS ))
+  while true; do
+    status="$(probe_aw)"
+    if [[ "$status" == "ok" ]]; then
+      return 0
+    fi
+    now="$(date -u +%s)"
+    if (( now >= deadline )); then
+      log "level=error event=wait_for_aw_timeout last_status=$status"
+      return 1
+    fi
+    log "level=info event=wait_for_aw status=$status"
+    sleep "$STARTUP_SLEEP_SECONDS"
+  done
+}
+
+run_recovery() {
+  local reason="$1"
+  local started_at="$2"
+  if cooldown_active; then
+    local finished_at
+    finished_at="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+    record_incident "skipped" "cooldown_active_after_${reason}" "false" "$started_at" "$finished_at" "restart suppressed by cooldown"
+    return 0
+  fi
+
+  if [[ "$DRY_RUN" == "1" ]]; then
+    local finished_at
+    finished_at="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+    record_incident "dry_run" "$reason" "false" "$started_at" "$finished_at" "dry run requested"
+    return 0
+  fi
+
+  log "level=warn event=recovery_start reason=$reason ct=$AW_CT_ID service=$AW_SERVICE"
+  stop_units_on_host "$PVE_PAUSE_UNITS"
+  stop_units_in_ct "$CT_PAUSE_UNITS"
+
+  log "level=warn event=restart_primary ct=$AW_CT_ID service=$AW_SERVICE"
+  ct_systemctl restart "$AW_SERVICE"
+  printf '%s\n' "$(date -u +%s)" >"${STATE_DIR}/last_restart_epoch"
+
+  local finished_at
+  if wait_for_aw_ok; then
+    start_units_in_ct "$CT_RESUME_UNITS"
+    start_units_on_host "$PVE_RESUME_UNITS"
+    finished_at="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+    record_incident "recovered" "$reason" "true" "$started_at" "$finished_at" "primary restarted and AW API returned ok"
+    return 0
+  fi
+
+  finished_at="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+  record_incident "failed" "$reason" "false" "$started_at" "$finished_at" "primary restart did not restore AW API; secondary timers left paused"
+  return 1
+}
+
+check_once() {
+  install -d -m 0750 "$STATE_DIR"
+  local started_at status attempt reason
+  started_at="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+  reason=""
+  for attempt in $(seq 1 "$CONFIRM_ATTEMPTS"); do
+    status="$(probe_aw)"
+    log "level=info event=probe attempt=$attempt status=$status"
+    case "$status" in
+      ok)
+        jq -n \
+          --arg generated_at_utc "$started_at" \
+          --arg status "ok" \
+          --arg aw_url "$AW_URL" \
+          '{generated_at_utc: $generated_at_utc, status: $status, aw_url: $aw_url}' \
+          >"${STATE_DIR}/status.json"
+        return 0
+        ;;
+      poisoned_lock|service_inactive|service_ct_unavailable)
+        reason="$status"
+        ;;
+      *)
+        local finished_at
+        finished_at="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+        record_incident "observed_no_action" "$status" "false" "$started_at" "$finished_at" "unhealthy state is not in automatic recovery allowlist"
+        return 0
+        ;;
+    esac
+    if [[ "$attempt" -lt "$CONFIRM_ATTEMPTS" ]]; then
+      sleep "$CONFIRM_SLEEP_SECONDS"
+    fi
+  done
+  run_recovery "$reason" "$started_at"
+}
+
+self_test() {
+  local tmp
+  tmp="$(mktemp)"
+  printf '{"message":"Taking datastore lock failed, returning 504: poisoned lock: another task failed inside"}' >"$tmp"
+  [[ "$(classify_http_response 503 "$tmp")" == "poisoned_lock" ]]
+  printf '{"ok":true}' >"$tmp"
+  [[ "$(classify_http_response 200 "$tmp")" == "ok" ]]
+  printf '{"message":"other"}' >"$tmp"
+  [[ "$(classify_http_response 503 "$tmp")" == "http_503" ]]
+  rm -f "$tmp"
+  log "level=info event=self_test status=ok"
+}
+
+main() {
+  local mode="${1:---once}"
+  case "$mode" in
+    --help|-h)
+      usage
+      ;;
+    --self-test)
+      require_command jq
+      self_test
+      ;;
+    --check-only)
+      require_command curl
+      require_command jq
+      require_command pct
+      status="$(probe_aw)"
+      log "level=info event=check_only status=$status"
+      [[ "$status" == "ok" ]]
+      ;;
+    --once)
+      require_command curl
+      require_command jq
+      require_command pct
+      require_command flock
+      install -d -m 0750 "$STATE_DIR"
+      exec 9>"${STATE_DIR}/lock"
+      if ! flock -n 9; then
+        log "level=warn event=lock_busy"
+        exit 0
+      fi
+      check_once
+      ;;
+    *)
+      usage >&2
+      exit 2
+      ;;
+  esac
+}
+
+main "$@"