[Phase 1] Fix Flaky Integration Tests + Migrate some tests to Test Containers

sdk-tests/src/test/java/io/dapr/it/DaprRun.java

@@ -172,10 +172,13 @@ public void stop() throws InterruptedException, IOException {
     System.out.println("Stopping dapr application ...");
     try {
       this.stopCommand.run();
-
       System.out.println("Dapr application stopped.");
     } catch (RuntimeException e) {
-      System.out.println("Could not stop app " + this.appName + ": " + e.getMessage());
+      if (e.getMessage() != null && e.getMessage().contains("Could not find success criteria")) {
+        System.out.println("App " + this.appName + " already stopped or not found (ignored).");
+      } else {
+        System.out.println("Could not stop app " + this.appName + ": " + e.getMessage());
+      }
     }
   }
 
@@ -219,8 +222,7 @@ public void waitForAppHealth(int maxWaitMilliseconds) throws InterruptedExceptio
         while (System.currentTimeMillis() <= maxWait) {
           try {
             stub.healthCheck(Empty.getDefaultInstance());
-            // artursouza: workaround due to race condition with runtime's probe on app's health.
-            Thread.sleep(5000);
+            Thread.sleep(2000);
             return;
           } catch (Exception e) {
             Thread.sleep(1000);
@@ -232,29 +234,31 @@ public void waitForAppHealth(int maxWaitMilliseconds) throws InterruptedExceptio
         channel.shutdown();
       }
     } else {
-      Duration waitDuration = Duration.ofMillis(maxWaitMilliseconds);
+      long maxWait = System.currentTimeMillis() + maxWaitMilliseconds;
       HttpClient client = HttpClient.newBuilder()
           .version(HttpClient.Version.HTTP_1_1)
-          .connectTimeout(waitDuration)
+          .connectTimeout(Duration.ofSeconds(5))
           .build();
       String url = "http://127.0.0.1:" + this.getAppPort() + "/health";
       HttpRequest request = HttpRequest.newBuilder()
           .GET()
           .uri(URI.create(url))
           .build();
 
-      try {
-        HttpResponse<String> response = client.send(request, HttpResponse.BodyHandlers.ofString());
-
-        if (response.statusCode() != 200) {
-          throw new RuntimeException("error: HTTP service is not healthy.");
+      while (System.currentTimeMillis() <= maxWait) {
+        try {
+          HttpResponse<String> response = client.send(request, HttpResponse.BodyHandlers.ofString());
+          if (response.statusCode() == 200) {
+            Thread.sleep(2000);
+            return;
+          }
+        } catch (IOException e) {
+          // not ready yet
         }
-      } catch (IOException e) {
-          throw new RuntimeException("exception: HTTP service is not healthy.");
+        Thread.sleep(1000);
       }
 
-      // artursouza: workaround due to race condition with runtime's probe on app's health.
-      Thread.sleep(5000);
+      throw new RuntimeException("timeout: HTTP service is not healthy.");
     }
   }
 

sdk-tests/src/test/java/io/dapr/it/actors/ActorReminderRecoveryIT.java

@@ -129,24 +129,19 @@ public void reminderRecoveryTest(
   ) throws Exception {
     setup(actorType);
 
-    logger.debug("Pausing 3 seconds to let gRPC connection get ready");
-    Thread.sleep(3000);
-
     logger.debug("Invoking actor method 'startReminder' which will register a reminder");
     proxy.invokeMethod("setReminderData", reminderDataParam).block();
 
     proxy.invokeMethod("startReminder",  reminderName).block();
 
-    logger.debug("Pausing 7 seconds to allow reminder to fire");
-    Thread.sleep(7000);
-
+    logger.debug("Waiting for reminder to fire at least 3 times");
     final List<MethodEntryTracker> logs = new ArrayList<>();
     callWithRetry(() -> {
       logs.clear();
       logs.addAll(fetchMethodCallLogs(proxy));
       validateMethodCalls(logs, METHOD_NAME, 3);
       validateMessageContent(logs, METHOD_NAME, expectedReminderStateText);
-    }, 5000);
+    }, 30000);
 
     // Restarts runtime only.
     logger.info("Stopping Dapr sidecar");

sdk-tests/src/test/java/io/dapr/it/actors/ActorTimerRecoveryIT.java

@@ -54,7 +54,6 @@ public void timerRecoveryTest() throws Exception {
       true,
       60000);
 
-    Thread.sleep(3000);
     String actorType="MyActorTest";
     logger.debug("Creating proxy builder");
 
@@ -68,16 +67,14 @@ public void timerRecoveryTest() throws Exception {
     logger.debug("Invoking actor method 'startTimer' which will register a timer");
     proxy.invokeMethod("startTimer", "myTimer").block();
 
-    logger.debug("Pausing 7 seconds to allow timer to fire");
-    Thread.sleep(7000);
-
+    logger.debug("Waiting for timer to fire at least 3 times");
     final List<MethodEntryTracker> logs = new ArrayList<>();
     callWithRetry(() -> {
       logs.clear();
       logs.addAll(fetchMethodCallLogs(proxy));
       validateMethodCalls(logs, METHOD_NAME, 3);
       validateMessageContent(logs, METHOD_NAME, "ping!");
-    }, 5000);
+    }, 30000);
 
     // Restarts app only.
     runs.left.stop();
@@ -91,7 +88,7 @@ public void timerRecoveryTest() throws Exception {
       newLogs.clear();
       newLogs.addAll(fetchMethodCallLogs(proxy));
       validateMethodCalls(newLogs, METHOD_NAME, 3);
-    }, 10000);
+    }, 30000);
 
     // Check that the restart actually happened by confirming the old logs are not in the new logs.
     for (MethodEntryTracker oldLog: logs) {