Files
hermes-config/bin/loop-watchdog.sh
Alexander Whitestone 64a8ffa329 fix: watchdog single-instance guard, stop killing worker processes
- Pidfile guard prevents duplicate watchdog instances
- Removed check_zombies — was killing legitimate git push/clone from
  active workers (caused mass exit 143 failures)
- Replaced with check_disk that cleans clone dirs >1hr old
- Disabled gemini loop check (no API key configured)
- Workers use fresh clone per issue (no shared worktree contention)
- Simplified cleanup to rm -rf (no git worktree bookkeeping)
- Tested file_issue end-to-end — confirmed working

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-22 22:03:58 -04:00

179 lines
5.2 KiB
Bash

#!/usr/bin/env bash
# loop-watchdog.sh — Self-healing monitor for agent loops
# Single instance enforced via pidfile. Checks every 2 minutes.
# Restarts dead loops. Files Gitea issues for persistent failures.
# Does NOT kill git processes — workers handle their own cleanup.
set -uo pipefail
LOG_DIR="$HOME/.hermes/logs"
LOG="$LOG_DIR/watchdog.log"
PIDFILE="$LOG_DIR/watchdog.pid"
ISSUE_LOG="$LOG_DIR/watchdog-issues.json"
GITEA_URL="http://143.198.27.163:3000"
ADMIN_TOKEN=$(cat "$HOME/.config/gitea/token" 2>/dev/null)
ISSUE_REPO="rockachopa/hermes-agent"
CHECK_INTERVAL=120
mkdir -p "$LOG_DIR"
[ -f "$ISSUE_LOG" ] || echo '{}' > "$ISSUE_LOG"
# === Single instance guard ===
if [ -f "$PIDFILE" ]; then
old_pid=$(cat "$PIDFILE")
if kill -0 "$old_pid" 2>/dev/null; then
echo "Watchdog already running (PID $old_pid)" >&2
exit 0
fi
fi
echo $$ > "$PIDFILE"
trap 'rm -f "$PIDFILE"' EXIT
log() {
echo "[$(date '+%Y-%m-%d %H:%M:%S')] WATCHDOG: $*" >> "$LOG"
}
# File a Gitea issue. Deduplicates within 6 hours.
file_issue() {
local issue_key="$1"
local title="$2"
local body="$3"
local assignee="${4:-claude}"
local should_file
should_file=$(python3 -c "
import json, time
try:
with open('$ISSUE_LOG') as f: filed = json.load(f)
except: filed = {}
entry = filed.get('$issue_key', {})
if entry and entry.get('until', 0) > time.time():
print('no')
else:
filed['$issue_key'] = {'until': time.time() + 21600}
with open('$ISSUE_LOG', 'w') as f: json.dump(filed, f, indent=2)
print('yes')
" 2>/dev/null)
[ "$should_file" != "yes" ] && return 0
log "FILING ISSUE: $title"
local tmpfile="/tmp/watchdog-issue-$$.json"
python3 -c "
import json, sys
with open('$tmpfile', 'w') as f:
json.dump({
'title': '[watchdog] ' + sys.argv[1],
'body': sys.argv[2] + '\n\n---\n*Auto-filed by loop-watchdog*',
'assignees': [sys.argv[3]],
}, f)
" "$title" "$body" "$assignee" 2>/dev/null
if [ -f "$tmpfile" ]; then
curl -sf -X POST "${GITEA_URL}/api/v1/repos/${ISSUE_REPO}/issues" \
-H "Authorization: token ${ADMIN_TOKEN}" \
-H "Content-Type: application/json" \
-d @"$tmpfile" >/dev/null 2>&1 || log "WARN: Failed to file issue: $title"
rm -f "$tmpfile"
fi
}
# Check if a loop is alive. Restart if dead.
check_loop() {
local name="$1"
local grep_pat="$2"
local wake_cmd="$3"
local log_file="$4"
local pid
pid=$(pgrep -f "$grep_pat" 2>/dev/null | head -1)
if [ -z "$pid" ]; then
log "$name loop DOWN — restarting..."
eval "$wake_cmd"
sleep 5
pid=$(pgrep -f "$grep_pat" 2>/dev/null | head -1)
if [ -z "$pid" ]; then
file_issue "${name}-loop-dead" \
"${name} loop won't start" \
"The ${name} agent loop failed to start.\nCommand: ${wake_cmd}\nLast log: $(tail -10 "$log_file" 2>/dev/null)" \
"claude"
else
log "$name loop restarted (PID $pid)"
fi
return
fi
# Check for stalls (no log activity > 30 min, and not idle)
if [ -f "$log_file" ]; then
local last_mod now stale
last_mod=$(stat -f %m "$log_file" 2>/dev/null || stat -c %Y "$log_file" 2>/dev/null || echo 0)
now=$(date +%s)
stale=$(( now - last_mod ))
if [ "$stale" -gt 1800 ]; then
local last_line
last_line=$(tail -1 "$log_file" 2>/dev/null)
# Idle is fine, truly stuck is not
if ! echo "$last_line" | grep -q "Queue empty\|Waiting\|idle"; then
log "$name loop stale (${stale}s) — restarting"
pkill -f "$grep_pat" 2>/dev/null
sleep 2
eval "$wake_cmd"
fi
fi
# Check for crash loops (10+ failures in last 50 lines)
local recent_failures
recent_failures=$(tail -50 "$log_file" 2>/dev/null | grep -c "FAILED:\|ERROR:" || true)
if [ "$recent_failures" -ge 10 ]; then
local errors
errors=$(tail -50 "$log_file" 2>/dev/null | grep "FAILED:\|ERROR:" | tail -5)
file_issue "${name}-crash-loop" \
"${name} in crash loop (${recent_failures} failures in 50 lines)" \
"Recent errors:\n${errors}" \
"claude"
fi
fi
}
check_gitea() {
if ! curl -sf --max-time 5 "${GITEA_URL}/api/v1/version" >/dev/null 2>&1; then
log "Gitea UNREACHABLE"
file_issue "gitea-down" "Gitea unreachable" \
"Gitea at ${GITEA_URL} not responding. Check VPS 143.198.27.163." "claude"
fi
}
check_disk() {
local dir_count
dir_count=$(ls -1d "$HOME/worktrees"/claude-w* 2>/dev/null | wc -l | tr -d ' ')
if [ "${dir_count:-0}" -gt 40 ]; then
log "WARN: $dir_count clone dirs — cleaning old ones"
# Delete clone dirs not modified in the last hour
find "$HOME/worktrees" -maxdepth 1 -name "claude-w*" -type d -mmin +60 -exec rm -rf {} \; 2>/dev/null
fi
}
# === MAIN ===
log "=== Watchdog Started (PID $$) ==="
while true; do
check_gitea
check_loop "claude" "claude-loop.sh" \
"nohup bash ~/.hermes/bin/claude-loop.sh 10 >> ~/.hermes/logs/claude-loop.log 2>&1 &" \
"$LOG_DIR/claude-loop.log"
check_loop "kimi" "kimi-loop.sh" \
"nohup bash ~/.hermes/bin/kimi-loop.sh >> ~/.hermes/logs/kimi-loop.log 2>&1 &" \
"$LOG_DIR/kimi-loop.log"
# Gemini disabled — no API key configured
# check_loop "gemini" ...
check_disk
sleep "$CHECK_INTERVAL"
done