ops: harden v20r1 monitor (correct hyperliquid-testnet.xyz host, DNS-aware venue-truth probe, heartbeat-tail dedup)

prod/ops/v20r1_monitor.sh

@@ -1,19 +1,24 @@
 #!/usr/bin/env bash
-# v20r1 monitor — sandbox ops helper (now repo-tracked: prod/ops/).
+# v20r1 monitor — sandbox ops helper (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
+# and independent HL testnet truth (unsigned /info — the same host the kernel's
-# kernel's HlVenueTruthSupplier uses). Testnet-only; reads only.
+# HlVenueTruthSupplier reads: hl/venue_truth.py + hl/instrument_provider.py).
+# Testnet-only; reads only. No system code/data is touched by this helper.
 #
 # Usage: /mnt/dolphinng5_predict/prod/ops/v20r1_monitor.sh   (or from anywhere)
 #
-# NOTE: v20r1-specific constants (signer, spool, PIDs) are hardcoded; when a new
+# NOTE: v20r1-specific constants (signer, spool, PIDs, HL host) are hardcoded;
-# campaign boots, bump ADDR/SPOOL and the PID pair in section [1].
+# when a new campaign boots, bump ADDR/LOG/PID_*. The HL host is the canonical
+# testnet hostname from prod/hl/config.py resolve_urls() => 'api.hyperliquid-testnet.xyz'
+# (liquid, NOT 'hyperlight'). A typo'd host resolves as NXDOMAIN and would
+# silently block venue truth — see section [5] error handling below.
 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`
+HL_API="https://api.hyperliquid-testnet.xyz/info"   # canonical testnet host (liquid, not hyperlight)
 
 echo "=== v20r1 @ $(date -u +%FT%TZ) ==="
 
@@ -23,22 +28,22 @@ ps -p "$PID_WRAPPER","$PID_KERNEL" -o pid,ppid,stat,etime,args 2>/dev/null || ec
 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)"
+echo "[3] HEARTBEAT (runner.log tail — TPSL ticker + latest BAR; stale-metadata noise excluded here, surfaced in [6])"
-tail -n 12 "$LOG" 2>/dev/null | grep -vE "RuntimeWarning|stddev|warnings.warn|c /= stddev|numpy" || echo "  (log unreadable)"
+tail -n 12 "$LOG" 2>/dev/null | grep -vE "RuntimeWarning|stddev|warnings.warn|c /= stddev|numpy|stale.*metadata" || 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}"
+echo "[5] INDEPENDENT HL TESTNET TRUTH (unsigned /info) — signer ${ADDR} @ ${HL_API}"
-# Probe captures curl exit+http code separately from the body so a DNS/TLS
+# Capture curl exit + HTTP code SEPARATELY from the body so a DNS/TLS failure
-# outage (the current failure mode: api.hyperlight-testnet.xyz NXDOMAIN across
+# prints UNREACHABLE — not a misleading JSON "Expecting value" parse error
-# resolvers) is reported as UNREACHABLE — not a misleading JSON parse error.
+# (which previously masked a bad hostname as a parse quirk).
 _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 "$bf" -w "%{http_code}" \
-          -X POST https://api.hyperlight-testnet.xyz/info \
+          -X POST "$HL_API" \
           -H 'Content-Type: application/json' \
           -d "{\"type\":\"$1\",\"user\":\"$ADDR\"}" 2>"$ef"); rc=$?
   if [ "$rc" -ne 0 ] || [ "${code:-000}" = "000" ]; then