@testdriverai/runner 7.8.0-canary.13 → 7.8.0-canary.15

package/lib/ably-service.js

@@ -275,6 +275,9 @@ class AblyService extends EventEmitter {
 
       this.emit('log', `Command received: ${type} (requestId=${requestId})`);
 
+      // Stop re-publishing runner.ready once we get the first command
+      this._stopReadySignal();
+
       // Per-command timeout: use message.timeout if provided, else default 120s
       // Prevents hanging forever if screenshot capture or S3 upload stalls
       const commandTimeout = (message.timeout && message.timeout > 0)
@@ -415,8 +418,8 @@ class AblyService extends EventEmitter {
 
         // Detect discontinuity: channel re-attached but message continuity was lost.
         // Use historyBeforeSubscribe() on each subscription to recover missed messages.
-        if (current === 'attached' && stateChange.resumed === false && previous) {
-          this.emit('log', `Ably channel [session]: DISCONTINUITY (resumed=false)${reasonMsg ? ' — ' + reasonMsg : ''}`);
+        if (current === 'attached' && stateChange.resumed === false && previous === 'attached') {
+          this.emit('log', `Ably channel [session]: DISCONTINUITY (resumed=false)${reasonMsg ? ' — ' + reasonMsg : ''}`);    
 
           Sentry.withScope((scope) => {
             scope.setTag('ably.client', 'runner');
@@ -459,7 +462,7 @@ class AblyService extends EventEmitter {
     // Signal readiness to SDK — commands sent before this would be lost
     const readyPayload = {
       type: 'runner.ready',
-      os: 'windows',
+      os: process.platform === 'win32' ? 'windows' : 'linux',
       sandboxId: this._sandboxId,
       runnerVersion: getLocalVersion() || 'unknown',
       timestamp: Date.now(),
@@ -473,6 +476,39 @@ class AblyService extends EventEmitter {
     }
     await this._sessionChannel.publish('control', readyPayload);
     this.emit('log', 'Published runner.ready signal');
+
+    // Re-publish runner.ready every 3s for up to 60s.
+    // The SDK may connect after the first publish (race condition),
+    // and Ably channel history may not be enabled. Repeating ensures
+    // the SDK catches at least one live runner.ready message.
+    this._readyInterval = setInterval(async () => {
+      try {
+        readyPayload.timestamp = Date.now();
+        await this._sessionChannel.publish('control', readyPayload);
+        this.emit('log', 'Re-published runner.ready signal');
+      } catch (err) {
+        this.emit('log', `Failed to re-publish runner.ready: ${err.message}`);
+      }
+    }, 3000);
+
+    // Stop after 60s regardless
+    this._readyTimeout = setTimeout(() => {
+      this._stopReadySignal();
+    }, 60000);
+  }
+
+  /**
+   * Stop the repeated runner.ready signal (called on first command or after timeout).
+   */
+  _stopReadySignal() {
+    if (this._readyInterval) {
+      clearInterval(this._readyInterval);
+      this._readyInterval = null;
+    }
+    if (this._readyTimeout) {
+      clearTimeout(this._readyTimeout);
+      this._readyTimeout = null;
+    }
   }
 
   /**
@@ -617,6 +653,8 @@ class AblyService extends EventEmitter {
   async close() {
     this.emit('log', 'Closing Ably service...');
 
+    this._stopReadySignal();
+
     if (this._statsInterval) {
       clearInterval(this._statsInterval);
       this._statsInterval = null;

package/lib/automation.js

@@ -45,8 +45,10 @@ const API_KEY = process.env.TD_API_KEY;
 // shell injection and escaping issues.
 
 const PYTHON = IS_WINDOWS ? 'python' : 'python3';
+// On Linux, ensure DISPLAY is set (use env var or fallback to :0)
+// The os.environ.get() preserves the parent's DISPLAY setting for E2B's :1 display
 const PY_IMPORT = IS_LINUX
-  ? "import os; os.environ['DISPLAY'] = ':0'; import pyautogui, sys; pyautogui.FAILSAFE = False; "
+  ? "import os; os.environ.setdefault('DISPLAY', ':0'); import pyautogui, sys; pyautogui.FAILSAFE = False; "
   : 'import pyautogui, sys; pyautogui.FAILSAFE = False; ';
 
 /**
@@ -525,11 +527,14 @@ class Automation extends EventEmitter {
         const timeout = Math.ceil((data.timeout || 300000) / 1000); // ms to seconds
         const requestId = data.requestId;
 
-        // Buffer stdout chunks to ~16KB before emitting over Ably.
+        // Buffer stdout chunks to ~32KB before emitting over Ably.
         // This reduces message count while keeping each message well under
-        // Ably's 64KB limit. The SDK accumulates these chunks and reconstructs
-        // the full stdout — the final response only carries returncode + stderr.
-        const CHUNK_FLUSH_SIZE = 16 * 1024; // 16KB
+        // Ably's 64KB limit. 32KB leaves headroom for the JSON envelope +
+        // string escaping while halving the message count vs the previous
+        // 16KB size, helping avoid Ably's per-channel rate limit on verbose
+        // commands. The SDK accumulates these chunks and reconstructs the
+        // full stdout — the final response only carries returncode + stderr.
+        const CHUNK_FLUSH_SIZE = 32 * 1024; // 32KB
         let chunkBuffer = '';
         const flushChunkBuffer = () => {
           if (chunkBuffer.length > 0) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@testdriverai/runner",
-  "version": "7.8.0-canary.13",
+  "version": "7.8.0-canary.15",
   "description": "TestDriver Runner - Ably-based remote automation agent with Node.js automation",
   "main": "index.js",
   "bin": {

package/scripts-desktop/start-agent.sh

@@ -0,0 +1,105 @@
+#!/bin/bash
+# ─── TestDriver Sandbox Agent Startup ────────────────────────────────────────
+# Starts the sandbox-agent.js (Ably-based automation agent) inside the E2B
+# sandbox. This script is called by the API after writing the config file
+# to /tmp/testdriver-agent.json.
+#
+# This matches the Windows runner pattern: the agent runs locally on the
+# sandbox and executes commands via pyautogui (instead of @e2b/desktop RPC).
+#
+# Usage: bash /opt/testdriver-runner/scripts-desktop/start-agent.sh [&]
+#
+# Prerequisites:
+#   - Desktop environment running (start-desktop.sh completed)
+#   - Config file at /tmp/testdriver-agent.json with Ably credentials
+#   - Node.js installed
+#   - Runner installed at /opt/testdriver-runner
+
+set -e
+
+export DISPLAY="${DISPLAY:-:0}"
+export XAUTHORITY="${XAUTHORITY:-${HOME}/.Xauthority}"
+
+RUNNER_DIR="/opt/testdriver-runner"
+CONFIG_PATH="/tmp/testdriver-agent.json"
+LOG_FILE="/tmp/sandbox-agent.log"
+PID_FILE="/tmp/sandbox-agent.pid"
+
+log() {
+  echo "[$(date -Iseconds)] [start-agent] $1" | tee -a "$LOG_FILE"
+}
+
+# ─── Check if already running ─────────────────────────────────────────────────
+if [ -f "$PID_FILE" ]; then
+  existing_pid=$(cat "$PID_FILE")
+  if kill -0 "$existing_pid" 2>/dev/null; then
+    log "Agent already running (PID: $existing_pid), exiting"
+    exit 0
+  else
+    log "Stale PID file found, removing"
+    rm -f "$PID_FILE"
+  fi
+fi
+
+# ─── Verify prerequisites ─────────────────────────────────────────────────────
+if [ ! -d "$RUNNER_DIR" ]; then
+  log "ERROR: Runner not found at $RUNNER_DIR"
+  exit 1
+fi
+
+if [ ! -f "$RUNNER_DIR/sandbox-agent.js" ]; then
+  log "ERROR: sandbox-agent.js not found in $RUNNER_DIR"
+  exit 1
+fi
+
+if ! command -v node &> /dev/null; then
+  log "ERROR: Node.js not installed"
+  exit 1
+fi
+
+# ─── Wait for config file (with timeout) ─────────────────────────────────────
+# The API writes the config file before calling this script, but we add a
+# brief wait just in case there's any race condition.
+WAIT_TIMEOUT=30
+WAIT_INTERVAL=1
+elapsed=0
+
+log "Waiting for config file: $CONFIG_PATH"
+while [ ! -f "$CONFIG_PATH" ] && [ $elapsed -lt $WAIT_TIMEOUT ]; do
+  sleep $WAIT_INTERVAL
+  elapsed=$((elapsed + WAIT_INTERVAL))
+done
+
+if [ ! -f "$CONFIG_PATH" ]; then
+  log "ERROR: Config file not found after ${WAIT_TIMEOUT}s: $CONFIG_PATH"
+  exit 1
+fi
+
+log "Config file found"
+
+# ─── Start the agent ──────────────────────────────────────────────────────────
+log "Starting sandbox-agent.js..."
+log "DISPLAY=$DISPLAY, RUNNER_DIR=$RUNNER_DIR"
+
+# Run in background, redirect output to log file
+cd "$RUNNER_DIR"
+nohup node sandbox-agent.js >> "$LOG_FILE" 2>&1 &
+AGENT_PID=$!
+
+# Write PID file for process management
+echo "$AGENT_PID" > "$PID_FILE"
+
+log "Agent started (PID: $AGENT_PID)"
+log "Log file: $LOG_FILE"
+
+# Brief pause to catch any immediate startup errors
+sleep 2
+
+if kill -0 "$AGENT_PID" 2>/dev/null; then
+  log "Agent running successfully"
+  exit 0
+else
+  log "ERROR: Agent exited unexpectedly. Check $LOG_FILE for details"
+  tail -20 "$LOG_FILE" | while read line; do log "  $line"; done
+  exit 1
+fi