Fix the iOS Maestro E2E suite on Linux runners against remote simulators

.github/actions/maestro-ios/action.yml

@@ -45,6 +45,14 @@ runs:
           "https://github.com/software-mansion/sim-remote-releases/releases/download/softu/sim-remote-${TARGET}"
         chmod +x "$HOME/.local/bin/sim-remote"
         echo "$HOME/.local/bin" >> "$GITHUB_PATH"
+
+        # `softu` is a rolling tag and has been rebuilt without the shims before.
+        if ! "$HOME/.local/bin/sim-remote" install-shims --help >/dev/null 2>&1; then
+          echo "The sim-remote build at the 'softu' tag has no 'install-shims' command," >&2
+          echo "so Maestro cannot drive a remote simulator. Rebuild the tag from a tree" >&2
+          echo "that includes the Maestro shims (software-mansion/radon-cloud#143)." >&2
+          exit 1
+        fi
     - name: Acquire remote simulator
       shell: bash
       # Credentials come from the calling job's env:
@@ -53,7 +61,21 @@ runs:
         # An unset SIM_ROUTER_URL secret arrives as an empty env var, which
         # would override the router URL baked into the binary — drop it.
         [ -n "${SIM_ROUTER_URL:-}" ] || unset SIM_ROUTER_URL
-        sim-remote login --timeout 900
+
+        sim-remote login --no-acquire
+
+        # The router clamps a single acquire wait to 300s, so retry rather than
+        # asking for a longer timeout that is silently truncated.
+        for attempt in $(seq 1 8); do
+          if sim-remote acquire --timeout 300; then
+            exit 0
+          fi
+          echo "Attempt $attempt: no fleet machine available yet, retrying..."
+          sleep 30
+        done
+
+        echo "Could not acquire a fleet machine" >&2
+        exit 1
     - name: Set up JDK 17
       uses: actions/setup-java@v5
       with:

.github/workflow-scripts/maestro-ios.js

@@ -61,11 +61,7 @@ function installAppOnSimulator(appPath, udid) {
   childProcess.execSync(`xcrun simctl install "${udid}" "${appPath}"`);
 }
 
-// When running against a remote simulator through the sim-remote Maestro
-// shims (`sim-remote install-shims` exports SIM_REMOTE_BIN), localhost inside
-// the simulator is the remote host, not this machine — so the app cannot
-// reach the local Metro server on its own. A reverse tunnel exposes the local
-// Metro port to the simulator via localhost.
+// A remote simulator's localhost is the remote host, so Metro needs a tunnel.
 function exposeMetroToSimulator(udid) {
   const simRemote = process.env.SIM_REMOTE_BIN;
   if (!simRemote) {
@@ -79,6 +75,10 @@ function exposeMetroToSimulator(udid) {
 }
 
 function bringSimulatorInForeground() {
+  // Nothing to foreground on a remote simulator.
+  if (process.env.SIM_REMOTE_BIN) {
+    return;
+  }
   console.log('Bringing simulator in foreground');
   childProcess.execSync('open -a simulator');
 }
@@ -116,18 +116,52 @@ function startVideoRecording(udid, currentAttempt) {
   return recordingProcess;
 }
 
+// The movie is only written after SIGINT, so returning early truncates it.
+const RECORDING_SHUTDOWN_TIMEOUT_MS = 30 * 1000;
+
 function stopVideoRecording(recordingProcess) {
   if (!recordingProcess) {
     console.log("Passed a null recording process. Can't kill it");
-    return;
+    return Promise.resolve();
   }
 
   console.log(`Stop video record using pid: ${recordingProcess.pid}`);
 
-  recordingProcess.kill('SIGINT');
+  if (
+    recordingProcess.exitCode != null ||
+    recordingProcess.signalCode != null
+  ) {
+    return Promise.resolve();
+  }
+
+  // Awaiting the exit is also what reaps the child: the flows run in a
+  // synchronous loop, so nothing else turns the event loop.
+  return new Promise(resolve => {
+    const done = () => {
+      clearTimeout(timer);
+      resolve();
+    };
+    const timer = setTimeout(() => {
+      console.log(
+        `Recorder ${recordingProcess.pid} did not exit in time, killing it`,
+      );
+      recordingProcess.kill('SIGKILL');
+    }, RECORDING_SHUTDOWN_TIMEOUT_MS);
+    timer.unref?.();
+
+    recordingProcess.once('exit', done);
+    recordingProcess.once('error', done);
+    recordingProcess.kill('SIGINT');
+  });
 }
 
-function executeFlowWithRetries(appId, udid, flow, jsengine, currentAttempt) {
+async function executeFlowWithRetries(
+  appId,
+  udid,
+  flow,
+  jsengine,
+  currentAttempt,
+) {
   const recProcess = startVideoRecording(udid, currentAttempt);
   try {
     const timeout = 1000 * 60 * 10; // 10 minutes
@@ -139,13 +173,19 @@ function executeFlowWithRetries(appId, udid, flow, jsengine, currentAttempt) {
       timeout,
     });
 
-    stopVideoRecording(recProcess);
+    await stopVideoRecording(recProcess);
   } catch (error) {
-    stopVideoRecording(recProcess);
+    await stopVideoRecording(recProcess);
 
     if (currentAttempt < MAX_ATTEMPTS) {
       console.info(`Retrying flow: ${flow}`);
-      executeFlowWithRetries(appId, udid, flow, jsengine, currentAttempt + 1);
+      await executeFlowWithRetries(
+        appId,
+        udid,
+        flow,
+        jsengine,
+        currentAttempt + 1,
+      );
     } else {
       console.error(
         `Failed to execute flow ${flow} after ${MAX_ATTEMPTS} attempts.`,
@@ -155,18 +195,23 @@ function executeFlowWithRetries(appId, udid, flow, jsengine, currentAttempt) {
   }
 }
 
-function executeFlows(appId, udid, maestroFlow, jsengine) {
+async function executeFlows(appId, udid, maestroFlow, jsengine) {
   if (!fs.existsSync(maestroFlow) || !fs.lstatSync(maestroFlow).isDirectory()) {
-    executeFlowWithRetries(appId, udid, maestroFlow, jsengine, 1);
+    await executeFlowWithRetries(appId, udid, maestroFlow, jsengine, 1);
     return;
   }
 
   for (const file of fs.readdirSync(maestroFlow).sort()) {
     const filePath = `${maestroFlow.replace(/\/$/, '')}/${file}`;
     if (fs.lstatSync(filePath).isDirectory()) {
-      executeFlows(appId, udid, filePath, jsengine);
+      // Fragments pulled in via `runFlow`; they have no `launchApp` of their
+      // own and fail when run standalone.
+      if (file === 'helpers') {
+        continue;
+      }
+      await executeFlows(appId, udid, filePath, jsengine);
     } else if (file.endsWith('.yml') || file.endsWith('.yaml')) {
-      executeFlowWithRetries(appId, udid, filePath, jsengine, 1);
+      await executeFlowWithRetries(appId, udid, filePath, jsengine, 1);
     }
   }
 }
@@ -201,7 +246,7 @@ async function main(args = process.argv.slice(2)) {
     exposeMetroToSimulator(simulator.udid);
   }
   await launchAppOnSimulator(appId, simulator.udid, isDebug);
-  executeFlows(appId, simulator.udid, maestroFlow, jsengine);
+  await executeFlows(appId, simulator.udid, maestroFlow, jsengine);
   console.log('Test finished');
 }
 

.github/workflows/e2e-maestro.yml

@@ -69,6 +69,8 @@ jobs:
       fail-on-error: true
     secrets: inherit
 
+  # Every flavor of both suites runs at once: a dispatch wants four fleet
+  # machines, and waits on the acquire retry when they are not free.
   test_e2e_ios_templateapp:
     needs:
       [

packages/rn-tester/.maestro/scrollview-minindex-maintainvisible.yml

@@ -14,6 +14,9 @@ appId: ${APP_ID}
       id: 'ScrollViewMaintainVisibleContentPositionExample'
     direction: DOWN
     speed: 80
+    # Each scroll step ships the whole accessibility tree over the wire
+    # against a remote simulator, which the default timeout cannot absorb.
+    timeout: 90000
 - tapOn:
     id: 'ScrollViewMaintainVisibleContentPositionExample'
 - waitForAnimationToEnd:

packages/rn-tester/.maestro/scrollview-threshold-maintainvisible.yml

@@ -14,6 +14,9 @@ appId: ${APP_ID}
       id: 'ScrollViewMaintainVisibleContentPositionExample'
     direction: DOWN
     speed: 80
+    # Each scroll step ships the whole accessibility tree over the wire
+    # against a remote simulator, which the default timeout cannot absorb.
+    timeout: 90000
 - tapOn:
     id: 'ScrollViewMaintainVisibleContentPositionExample'
 - waitForAnimationToEnd: