Fix LlamaDemo Android instrumentation CI reliability (#153)

- Fix emulator configuration to use 16GB RAM properly with both `ram-size` and `-memory` flags
- Add robust `adb push` with timeout, retry, and file size verification

Author: kirklandsign

.github/workflows/llm-android.yml

@@ -42,8 +42,6 @@ jobs:
     env:
       API_LEVEL: 34
       ARCH: x86_64
-      EMULATOR_OPTIONS: -no-window -gpu swiftshader_indirect -noaudio -no-boot-anim -camera-back none
-      RAM_SIZE: 16384
 
     name: Instrumentation Test LlamaDemo
     steps:
@@ -83,35 +81,22 @@ jobs:
           path: |
             ~/.android/avd/*
             ~/.android/adb*
-          key: avd-${{ env.API_LEVEL }}-${{ env.ARCH }}-ram${{ env.RAM_SIZE }}-disk16G
+          key: avd-${{ env.API_LEVEL }}-${{ env.ARCH }}-ram16G-disk16G-v5
 
       - name: Create AVD and generate snapshot for caching
         if: steps.avd-cache.outputs.cache-hit != 'true'
         uses: reactivecircus/android-emulator-runner@v2
         with:
           api-level: ${{ env.API_LEVEL }}
           arch: ${{ env.ARCH }}
+          ram-size: 16384M
+          disk-size: 16384M
           force-avd-creation: true
-          emulator-options: ${{ env.EMULATOR_OPTIONS }}
+          emulator-options: -no-window -gpu swiftshader_indirect -noaudio -no-boot-anim -camera-back none -no-snapshot-save -memory 16384
           disable-animations: false
           working-directory: llm/android/LlamaDemo
           script: echo "Generated AVD snapshot for caching."
 
-      - name: Configure AVD RAM and disk
-        run: |
-          AVD_DIR="$HOME/.android/avd"
-          for config in "$AVD_DIR"/*.avd/config.ini; do
-            if [ -f "$config" ]; then
-              echo "Updating config in $config"
-              # Update RAM
-              sed -i 's/hw.ramSize=.*/hw.ramSize=${{ env.RAM_SIZE }}/' "$config" || true
-              grep -q "hw.ramSize" "$config" || echo "hw.ramSize=${{ env.RAM_SIZE }}" >> "$config"
-              # Update disk size to 16GB for large models
-              sed -i 's/disk.dataPartition.size=.*/disk.dataPartition.size=16G/' "$config" || true
-              grep -q "disk.dataPartition.size" "$config" || echo "disk.dataPartition.size=16G" >> "$config"
-            fi
-          done
-
       - name: Download model files
         env:
           MODEL_PRESET: ${{ inputs.model_preset || 'stories' }}
@@ -163,8 +148,10 @@ jobs:
         with:
           api-level: ${{ env.API_LEVEL }}
           arch: ${{ env.ARCH }}
-          force-avd-creation: false
-          emulator-options: -no-snapshot-save ${{ env.EMULATOR_OPTIONS }}
+          ram-size: 16384M
+          disk-size: 16384M
+          force-avd-creation: true
+          emulator-options: -no-window -gpu swiftshader_indirect -noaudio -no-boot-anim -camera-back none -no-snapshot-save -memory 16384
           disable-animations: true
           working-directory: llm/android/LlamaDemo
           script: bash ./scripts/run-ci-tests.sh "$MODEL_PRESET" "$MODEL_FILE" "$TOKENIZER_FILE"

llm/android/LlamaDemo/app/src/androidTest/java/com/example/executorchllamademo/UIWorkflowTest.java

@@ -199,8 +199,8 @@ public void testSendMessageAndReceiveResponse() throws Exception {
             onView(withId(R.id.sendButton)).perform(click());
 
             // --- Wait for response and validate ---
-            // Wait 5 seconds for model to generate response
-            Thread.sleep(5000);
+            // Wait 50 seconds for model to generate response
+            Thread.sleep(50000);
 
             // Extract all messages from the list
             AtomicInteger messageCount = new AtomicInteger(0);

llm/android/LlamaDemo/scripts/run-ci-tests.sh

@@ -25,17 +25,59 @@ echo "TOKENIZER_FILE: $TOKENIZER_FILE"
 echo "=== Emulator Memory Info ==="
 adb shell cat /proc/meminfo | head -5
 
+echo "=== Emulator Disk Space ==="
+adb shell df -h /data
+
 # Clean and prepare device directory
 adb shell rm -rf /data/local/tmp/llama
 adb shell mkdir -p /data/local/tmp/llama
 
-# Push pre-downloaded model files to device
+# Push pre-downloaded model files to device with timeout and retry
 echo "=== Pushing pre-downloaded model files to device ==="
 for file in /tmp/llama_models/*; do
-  echo "Pushing $(basename "$file")..."
-  adb push "$file" /data/local/tmp/llama/
+  filename=$(basename "$file")
+  filesize=$(stat -c%s "$file" 2>/dev/null || stat -f%z "$file" 2>/dev/null || echo "0")
+  # Calculate timeout: 30 seconds base + 1 second per 50MB
+  timeout_secs=$((30 + filesize / 50000000))
+  echo "Pushing $filename (size: $((filesize / 1024 / 1024))MB, timeout: ${timeout_secs}s)..."
+
+  max_retries=3
+  retry=0
+  success=false
+
+  while [ $retry -lt $max_retries ] && [ "$success" = "false" ]; do
+    # Run push (ignore exit code, verify by checking file on device)
+    timeout $timeout_secs adb push "$file" /data/local/tmp/llama/ || true
+
+    # Verify file was pushed by checking it exists and has correct size
+    device_size=$(adb shell "stat -c%s /data/local/tmp/llama/$filename 2>/dev/null || echo 0" | tr -d '\r')
+    if [ "$device_size" = "$filesize" ]; then
+      success=true
+      echo "Successfully pushed $filename (verified size: $device_size bytes)"
+    else
+      retry=$((retry + 1))
+      echo "Push failed or incomplete (attempt $retry/$max_retries, expected $filesize bytes, got $device_size bytes)"
+      if [ $retry -lt $max_retries ]; then
+        echo "Waiting 5 seconds before retry..."
+        sleep 5
+        echo "Checking if emulator is still responsive..."
+        adb shell getprop ro.build.version.sdk || echo "WARNING: Emulator may be unresponsive"
+      fi
+    fi
+  done
+
+  if [ "$success" = "false" ]; then
+    echo "ERROR: Failed to push $filename after $max_retries attempts"
+    exit 1
+  fi
+
+  echo "Checking emulator responsiveness..."
+  adb shell getprop ro.build.version.sdk || echo "WARNING: Emulator may be unresponsive"
 done
 
+echo "=== Syncing filesystem ==="
+adb shell sync
+
 echo "=== Model directory contents ==="
 adb shell ls -la /data/local/tmp/llama/