ops: add plant-vision node1 service and update monitor/prober scripts

ops/monitor_notify_sofiia.sh

@@ -7,6 +7,7 @@ ROUTER_URL="${ROUTER_URL:-http://127.0.0.1:9102}"
 REPORT_ENABLED="${SOFIIA_REPORTS_ENABLED:-true}"
 REPORT_MODE="${SOFIIA_REPORT_MODE:-fail_only}"   # fail_only | always
 REPORT_TIMEOUT="${SOFIIA_REPORT_TIMEOUT:-180}"
+REPORT_MAX_TOKENS="${SOFIIA_REPORT_MAX_TOKENS:-900}"
 REPORT_CHAT_ID="${SOFIIA_REPORT_CHAT_ID:-ops-monitor-sofiia}"
 REPORT_USER_ID="${SOFIIA_REPORT_USER_ID:-ops-monitor-agent}"
 REPORT_USERNAME="${SOFIIA_REPORT_USERNAME:-monitor-agent}"
@@ -23,7 +24,7 @@ if [[ ! -f "$STATUS_JSON" ]]; then
   exit 0
 fi
 
-python3 - "$STATUS_JSON" "$ROOT" "$ROUTER_URL" "$REPORT_MODE" "$REPORT_TIMEOUT" "$REPORT_CHAT_ID" "$REPORT_USER_ID" "$REPORT_USERNAME" "$REPORT_TELEGRAM_CHAT_ID" "$SOFIIA_BOT_TOKEN" <<'PY'
+python3 - "$STATUS_JSON" "$ROOT" "$ROUTER_URL" "$REPORT_MODE" "$REPORT_TIMEOUT" "$REPORT_MAX_TOKENS" "$REPORT_CHAT_ID" "$REPORT_USER_ID" "$REPORT_USERNAME" "$REPORT_TELEGRAM_CHAT_ID" "$SOFIIA_BOT_TOKEN" <<'PY'
 import json
 import sys
 from pathlib import Path
@@ -35,11 +36,12 @@ root = Path(sys.argv[2])
 router_url = sys.argv[3].rstrip('/')
 report_mode = sys.argv[4]
 timeout_s = int(sys.argv[5])
-chat_id = sys.argv[6]
-user_id = sys.argv[7]
-username = sys.argv[8]
-tg_chat_id = sys.argv[9].strip()
-tg_token = sys.argv[10].strip()
+max_tokens = int(sys.argv[6])
+chat_id = sys.argv[7]
+user_id = sys.argv[8]
+username = sys.argv[9]
+tg_chat_id = sys.argv[10].strip()
+tg_token = sys.argv[11].strip()
 
 payload = json.loads(status_json.read_text(encoding='utf-8'))
 status = str(payload.get('status', 'unknown')).lower()
@@ -70,7 +72,7 @@ prompt = (
 
 body = {
     'prompt': prompt,
-    'max_tokens': 400,
+    'max_tokens': max_tokens,
     'temperature': 0.1,
     'metadata': {
         'source': 'ops-monitor-canary',
@@ -99,26 +101,42 @@ try:
         print(f"[OK] sofiia report sent: backend={data.get('backend')} model={data.get('model')} preview={short!r}")
 
         if tg_chat_id and tg_token and text:
-            msg = (
+            def chunk_text(value: str, limit: int = 3500):
+                chunks = []
+                remaining = value
+                while remaining:
+                    if len(remaining) <= limit:
+                        chunks.append(remaining)
+                        break
+                    split_at = remaining.rfind('\n', 0, limit)
+                    if split_at < max(1, limit // 2):
+                        split_at = limit
+                    chunks.append(remaining[:split_at].rstrip())
+                    remaining = remaining[split_at:].lstrip()
+                return chunks or [value]
+
+            header = (
                 "[NODE1 Monitor]\n"
                 f"status={payload.get('status')} exit_code={payload.get('exit_code')}\n\n"
-                f"{text[:3500]}"
             )
-            tg_req = urlreq.Request(
-                url=f"https://api.telegram.org/bot{tg_token}/sendMessage",
-                data=json.dumps({"chat_id": tg_chat_id, "text": msg}).encode('utf-8'),
-                headers={'Content-Type': 'application/json'},
-                method='POST',
-            )
-            try:
+            parts = chunk_text(text, 3500 - len("(99/99)\n"))
+            total = len(parts)
+            delivered = 0
+            for idx, part in enumerate(parts, start=1):
+                prefix = f"({idx}/{total})\n" if total > 1 else ""
+                msg = f"{header}{prefix}{part}" if idx == 1 else f"{prefix}{part}"
+                tg_req = urlreq.Request(
+                    url=f"https://api.telegram.org/bot{tg_token}/sendMessage",
+                    data=json.dumps({"chat_id": tg_chat_id, "text": msg}).encode('utf-8'),
+                    headers={'Content-Type': 'application/json'},
+                    method='POST',
+                )
                 with urlreq.urlopen(tg_req, timeout=20) as tg_resp:
                     tg_data = json.loads(tg_resp.read().decode('utf-8', errors='ignore'))
-                    if tg_data.get('ok'):
-                        print(f"[OK] telegram report delivered: chat_id={tg_chat_id}")
-                    else:
-                        print(f"[WARN] telegram send not ok: {tg_data}")
-            except Exception as tg_e:
-                print(f"[WARN] telegram send failed: {tg_e}")
+                    if not tg_data.get('ok'):
+                        raise RuntimeError(f"telegram send not ok: {tg_data}")
+                delivered += 1
+            print(f"[OK] telegram report delivered: chat_id={tg_chat_id} parts={delivered}")
         else:
             print('[INFO] telegram delivery skipped (missing SOFIIA_REPORT_TELEGRAM_CHAT_ID or token or empty text)')
 except HTTPError as e: