ops: add v20r1 live-monitor helper (prod/ops/v20r1_monitor.sh)

prod/ops/v20r1_monitor.sh

@@ -0,0 +1,69 @@
+#!/usr/bin/env bash
+# v20r1 monitor — sandbox ops helper (now repo-tracked: prod/ops/).
+# Run anytime to snapshot: process, lease, heartbeat, FORCE_ENGAGE cadence,
+# and independent HL testnet truth (unsigned /info — same source of truth the
+# kernel's HlVenueTruthSupplier uses). Testnet-only; reads only.
+#
+# Usage: /mnt/dolphinng5_predict/prod/ops/v20r1_monitor.sh   (or from anywhere)
+#
+# NOTE: v20r1-specific constants (signer, spool, PIDs) are hardcoded; when a new
+# campaign boots, bump ADDR/SPOOL and the PID pair in section [1].
+set -u
+ADDR="0x6af790574E5E9EA067FB4Ce7458B429039cE3f96"
+LOG="/root/dolphin_logs/uv_journal_runs/f13_terminal_slot_release_v20r1_20260826/runner.log"
+LEASE="/tmp/flight13_hl_testnet_single_writer.lock"
+PID_WRAPPER=790059   # `timeout --signal=TERM 8h launch_flight13_testnet.sh`
+PID_KERNEL=790060    # `/root/uv-wt/f13-hl/.f13_boot.py`
+
+echo "=== v20r1 @ $(date -u +%FT%TZ) ==="
+
+echo "[1] PROCESS (wrapper=${PID_WRAPPER} kernel=${PID_KERNEL})"
+ps -p "$PID_WRAPPER","$PID_KERNEL" -o pid,ppid,stat,etime,args 2>/dev/null || echo "  !! PIDs ${PID_WRAPPER}/${PID_KERNEL} NOT alive"
+
+echo "[2] EXCLUSIVE-WRITER LEASE (${LEASE})"
+ls -la "$LEASE" 2>/dev/null && cat "$LEASE" 2>/dev/null || echo "  !! lease file absent"
+
+echo "[3] HEARTBEAT (runner.log tail — TPSL ticker + latest BAR)"
+tail -n 12 "$LOG" 2>/dev/null | grep -vE "RuntimeWarning|stddev|warnings.warn|c /= stddev|numpy" || echo "  (log unreadable)"
+
+echo "[4] FORCE_ENGAGE cadence (synthetic SHORT firehose)"
+printf '  synthetic-enter log lines: '; grep -c "FORCE_ENGAGE: SYNTHETIC ENTER" "$LOG" 2>/dev/null || echo 0
+grep "in_flight=" "$LOG" 2>/dev/null | tail -n 1 | sed 's/^/  last in_flight: /'
+
+echo "[5] INDEPENDENT HL TESTNET TRUTH (unsigned /info) — signer ${ADDR}"
+# Probe captures curl exit+http code separately from the body so a DNS/TLS
+# outage (the current failure mode: api.hyperlight-testnet.xyz NXDOMAIN across
+# resolvers) is reported as UNREACHABLE — not a misleading JSON parse error.
+_hl_info() {  # $1 = HL /info type (openOrders | clearinghouseState | userFills | allMids)
+  local ef bf code rc
+  ef=$(mktemp); bf=$(mktemp)
+  code=$(curl -sS --max-time 12 -o "$bf" -w "%{http_code}" \
+          -X POST https://api.hyperlight-testnet.xyz/info \
+          -H 'Content-Type: application/json' \
+          -d "{\"type\":\"$1\",\"user\":\"$ADDR\"}" 2>"$ef"); rc=$?
+  if [ "$rc" -ne 0 ] || [ "${code:-000}" = "000" ]; then
+    printf '  %s: UNREACHABLE — venue truth blocked (curl rc=%s http=%s; DNS/venue outage)\n' "$1" "${rc:-0}" "${code:-000}"
+    printf '    detail: %s\n' "$(head -c 200 "$ef" 2>/dev/null)"
+    rm -f "$ef" "$bf"; return
+  fi
+  python3 - "$bf" "$1" <<'PY'
+import sys,json
+bf,typ=sys.argv[1],sys.argv[2]
+try:
+    d=json.load(open(bf))
+    if typ=='openOrders':
+        o=d; print('  openOrders:', len(o) if isinstance(o,list) else o)
+        for x in (o if isinstance(o,list) else []): print('    -',x.get('coin'),x.get('side'),'px',x.get('px'),'oid',x.get('oid'))
+    else:
+        m=d['marginSummary']; print('  accountValue=%s totalNtlPos=%s withdrawable=%s'%(m['accountValue'],m['totalNtlPos'],d['withdrawable']))
+        for p in d.get('assetPositions',[]):po=p['position']
+        if float(po['szi'])!=0: print('  POSITION',po['coin'],'szi',po['szi'],'entry',po['entryPx'],'ROE%',po['returnOnEquity'])
+except Exception as e: print('  '+typ+' parse failed:',e)
+PY
+  rm -f "$ef" "$bf"
+}
+_hl_info openOrders
+_hl_info clearinghouseState
+
+echo "[6] ALERTS (tail CRITICAL/ERROR/REFUSING/LEASE-stale-metadata)"
+grep -E "CRITICAL|ERROR|REFUSING|LEASE REFUSED|stale.*metadata" "$LOG" 2>/dev/null | tail -n 6 | sed 's/^/  /'