Fix the algorithm counting the package failures

Use the sliding window algorithm to detect if there exists a window
containing failures equal to or above the trigger threshold.

Bug: 140841942
Test: atest PackageWatchdogTest
Change-Id: I34a20e4d3b98a093dffa05fc7c7c026905834b53

2 files changed, 45 insertions, 17 deletions

diff --git a/services/core/java/com/android/server/PackageWatchdog.java b/services/core/java/com/android/server/PackageWatchdog.java
index 548665ba3a32..4403efcd58c3 100644
--- a/services/core/java/com/android/server/PackageWatchdog.java
+++ b/services/core/java/com/android/server/PackageWatchdog.java
@@ -34,6 +34,7 @@ import android.text.TextUtils;
 import android.util.ArrayMap;
 import android.util.ArraySet;
 import android.util.AtomicFile;
+import android.util.LongArrayQueue;
 import android.util.Slog;
 import android.util.Xml;
 
@@ -969,6 +970,9 @@ public class PackageWatchdog {
     class MonitoredPackage {
         //TODO(b/120598832): VersionedPackage?
         private final String mName;
+        // Times when package failures happen sorted in ascending order
+        @GuardedBy("mLock")
+        private final LongArrayQueue mFailureHistory = new LongArrayQueue();
         // One of STATE_[ACTIVE|INACTIVE|PASSED|FAILED]. Updated on construction and after
         // methods that could change the health check state: handleElapsedTimeLocked and
         // tryPassHealthCheckLocked
@@ -988,12 +992,6 @@ public class PackageWatchdog {
         // of the package, see #getHealthCheckStateLocked
         @GuardedBy("mLock")
         private long mHealthCheckDurationMs = Long.MAX_VALUE;
-        // System uptime of first package failure
-        @GuardedBy("mLock")
-        private long mUptimeStartMs;
-        // Number of failures since mUptimeStartMs
-        @GuardedBy("mLock")
-        private int mFailures;
 
         MonitoredPackage(String name, long durationMs, boolean hasPassedHealthCheck) {
             this(name, durationMs, Long.MAX_VALUE, hasPassedHealthCheck);
@@ -1028,20 +1026,17 @@ public class PackageWatchdog {
          */
         @GuardedBy("mLock")
         public boolean onFailureLocked() {
+            // Sliding window algorithm: find out if there exists a window containing failures >=
+            // mTriggerFailureCount.
             final long now = mSystemClock.uptimeMillis();
-            final long duration = now - mUptimeStartMs;
-            if (duration > mTriggerFailureDurationMs) {
-                // TODO(b/120598832): Reseting to 1 is not correct
-                // because there may be more than 1 failure in the last trigger window from now
-                // This is the RescueParty impl, will leave for now
-                mFailures = 1;
-                mUptimeStartMs = now;
-            } else {
-                mFailures++;
+            mFailureHistory.addLast(now);
+            while (now - mFailureHistory.peekFirst() > mTriggerFailureDurationMs) {
+                // Prune values falling out of the window
+                mFailureHistory.removeFirst();
             }
-            boolean failed = mFailures >= mTriggerFailureCount;
+            boolean failed = mFailureHistory.size() >= mTriggerFailureCount;
             if (failed) {
-                mFailures = 0;
+                mFailureHistory.clear();
             }
             return failed;
         }
diff --git a/tests/PackageWatchdog/src/com/android/server/PackageWatchdogTest.java b/tests/PackageWatchdog/src/com/android/server/PackageWatchdogTest.java
index ab31ed7389a3..ff4c9702bd80 100644
--- a/tests/PackageWatchdog/src/com/android/server/PackageWatchdogTest.java
+++ b/tests/PackageWatchdog/src/com/android/server/PackageWatchdogTest.java
@@ -765,6 +765,39 @@ public class PackageWatchdogTest {
         assertThat(observer.mHealthCheckFailedPackages).containsExactly(APP_B);
     }
 
+    /** Test we are notified when enough failures are triggered within any window. */
+    @Test
+    public void testFailureTriggerWindow() {
+        adoptShellPermissions(
+                Manifest.permission.WRITE_DEVICE_CONFIG,
+                Manifest.permission.READ_DEVICE_CONFIG);
+        DeviceConfig.setProperty(DeviceConfig.NAMESPACE_ROLLBACK,
+                PackageWatchdog.PROPERTY_WATCHDOG_TRIGGER_FAILURE_COUNT,
+                Integer.toString(3), /*makeDefault*/false);
+        DeviceConfig.setProperty(DeviceConfig.NAMESPACE_ROLLBACK,
+                PackageWatchdog.PROPERTY_WATCHDOG_TRIGGER_DURATION_MILLIS,
+                Integer.toString(1000), /*makeDefault*/false);
+        PackageWatchdog watchdog = createWatchdog();
+        TestObserver observer = new TestObserver(OBSERVER_NAME_1);
+
+        watchdog.startObservingHealth(observer, Arrays.asList(APP_A), Long.MAX_VALUE);
+        // Raise 2 failures at t=0 and t=900 respectively
+        watchdog.onPackageFailure(Arrays.asList(new VersionedPackage(APP_A, VERSION_CODE)));
+        mTestLooper.dispatchAll();
+        moveTimeForwardAndDispatch(900);
+        watchdog.onPackageFailure(Arrays.asList(new VersionedPackage(APP_A, VERSION_CODE)));
+        mTestLooper.dispatchAll();
+
+        // Raise 2 failures at t=1100
+        moveTimeForwardAndDispatch(200);
+        watchdog.onPackageFailure(Arrays.asList(new VersionedPackage(APP_A, VERSION_CODE)));
+        watchdog.onPackageFailure(Arrays.asList(new VersionedPackage(APP_A, VERSION_CODE)));
+        mTestLooper.dispatchAll();
+
+        // We should receive APP_A since there are 3 failures within 1000ms window
+        assertThat(observer.mHealthCheckFailedPackages).containsExactly(APP_A);
+    }
+
     private void adoptShellPermissions(String... permissions) {
         InstrumentationRegistry
                 .getInstrumentation()