replicas-engine 0.1.162 → 0.1.164

package/dist/src/index.js

@@ -285,7 +285,7 @@ function parseReplicasConfigString(content, filename) {
 }
 
 // ../shared/src/engine/environment.ts
-var DAYTONA_SNAPSHOT_ID = "13-05-2026-royal-york-v8";
+var DAYTONA_SNAPSHOT_ID = "13-05-2026-royal-york-v10";
 
 // ../shared/src/engine/types.ts
 var DEFAULT_CHAT_TITLES = {

package/package.json

@@ -1,14 +1,16 @@
 {
   "name": "replicas-engine",
-  "version": "0.1.162",
+  "version": "0.1.164",
   "description": "Lightweight API server for Replicas workspaces",
   "type": "module",
   "main": "dist/src/index.js",
   "bin": {
-    "replicas-engine": "./dist/src/index.js"
+    "replicas-engine": "./dist/src/index.js",
+    "replicas-engine-watchdog": "./scripts/engine-watchdog.sh"
   },
   "files": [
-    "dist"
+    "dist",
+    "scripts/engine-watchdog.sh"
   ],
   "scripts": {
     "dev": "tsx watch src/index.ts",

package/scripts/engine-watchdog.sh

@@ -0,0 +1,78 @@
+#!/usr/bin/env bash
+# Watchdog for replicas-engine: monitors the process and restarts on crash.
+# Uses exponential backoff to avoid tight restart loops on persistent failures.
+# Resets backoff after the engine runs stably for STABLE_THRESHOLD seconds.
+set -u
+
+WARMING_FLAG="${1:-}"
+PIDFILE="/tmp/replicas-engine.pid"
+MAX_BACKOFF=60
+BACKOFF=1
+RESTART_COUNT=0
+MAX_RESTARTS=20
+STABLE_THRESHOLD=300
+
+log() {
+  echo "[$(date -Iseconds)] [engine-watchdog] $*"
+}
+
+cleanup() {
+  log "Watchdog shutting down"
+  if [ -f "$PIDFILE" ]; then
+    local pid
+    pid=$(cat "$PIDFILE")
+    if kill -0 "$pid" 2>/dev/null; then
+      kill "$pid" 2>/dev/null
+      wait "$pid" 2>/dev/null
+    fi
+  fi
+  rm -f "$PIDFILE"
+  exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+while true; do
+  if [ "$RESTART_COUNT" -ge "$MAX_RESTARTS" ]; then
+    log "Exceeded max restarts ($MAX_RESTARTS). Giving up."
+    exit 1
+  fi
+
+  BOOTSTRAP_LOG="/tmp/replicas-engine-bootstrap-$(date +%Y%m%d-%H%M%S).log"
+  log "Starting replicas-engine (attempt $((RESTART_COUNT + 1)))${WARMING_FLAG:+ [warming mode]}"
+
+  START_TIME=$(date +%s)
+  replicas-engine $WARMING_FLAG >> "$BOOTSTRAP_LOG" 2>&1 &
+  ENGINE_PID=$!
+  echo "$ENGINE_PID" > "$PIDFILE"
+
+  wait "$ENGINE_PID"
+  EXIT_CODE=$?
+  END_TIME=$(date +%s)
+  UPTIME=$((END_TIME - START_TIME))
+
+  rm -f "$PIDFILE"
+
+  # Exit code 0 = clean shutdown, 143 = SIGTERM (sandbox stopping)
+  if [ $EXIT_CODE -eq 0 ] || [ $EXIT_CODE -eq 143 ]; then
+    log "Engine exited (code $EXIT_CODE). Not restarting."
+    exit 0
+  fi
+
+  RESTART_COUNT=$((RESTART_COUNT + 1))
+
+  if [ "$UPTIME" -ge "$STABLE_THRESHOLD" ]; then
+    BACKOFF=1
+    RESTART_COUNT=1
+    log "Engine crashed (exit code $EXIT_CODE, uptime ${UPTIME}s) after stable run. Resetting backoff."
+  else
+    log "Engine crashed (exit code $EXIT_CODE, uptime ${UPTIME}s). Restart #$RESTART_COUNT in ${BACKOFF}s..."
+    sleep "$BACKOFF"
+    if [ "$BACKOFF" -lt "$MAX_BACKOFF" ]; then
+      BACKOFF=$((BACKOFF * 2))
+      if [ "$BACKOFF" -gt "$MAX_BACKOFF" ]; then
+        BACKOFF=$MAX_BACKOFF
+      fi
+    fi
+  fi
+done