vrozov commented on code in PR #52373:
URL: https://github.com/apache/spark/pull/52373#discussion_r2612605903


##########
core/src/test/scala/org/apache/spark/util/UninterruptibleThreadSuite.scala:
##########
@@ -80,130 +99,128 @@ class UninterruptibleThreadSuite extends SparkFunSuite {
     t.interrupt()
     interruptLatch.countDown()
     t.join()
-    assert(hasInterruptedException === false)
-    assert(interruptStatusBeforeExit)
+    assert(!hasInterruptedException, "runUninterruptibly should not be 
interrupted")
+    assert(interruptStatusBeforeExit, "interrupt flag should be set")
   }
 
   test("nested runUninterruptibly") {
     val enterRunUninterruptibly = new CountDownLatch(1)
     val interruptLatch = new CountDownLatch(1)
     @volatile var hasInterruptedException = false
     @volatile var interruptStatusBeforeExit = false
-    val t = new UninterruptibleThread("test") {
+    val t = new UninterruptibleThread("runUninterruptibly") {
       override def run(): Unit = {
         runUninterruptibly {
           enterRunUninterruptibly.countDown()
-          Uninterruptibles.awaitUninterruptibly(interruptLatch, 10, 
TimeUnit.SECONDS)
-          hasInterruptedException = sleep(1)
-          runUninterruptibly {
-            if (sleep(1)) {
-              hasInterruptedException = true
+          hasInterruptedException = await(interruptLatch)
+          if (!hasInterruptedException) {
+            runUninterruptibly {
+              hasInterruptedException = sleep(0)
             }
-          }
-          if (sleep(1)) {
-            hasInterruptedException = true
+            hasInterruptedException |= sleep(0)
           }
         }
         interruptStatusBeforeExit = Thread.interrupted()
       }
     }
     t.start()
-    assert(enterRunUninterruptibly.await(10, TimeUnit.SECONDS), "await 
timeout")
+    assert(!await(enterRunUninterruptibly), "await interrupted")
     t.interrupt()
     interruptLatch.countDown()
+    assert(!sleep(0), "sleep should not be interrupted")
+    t.interrupt()
+    assert(!sleep(0), "sleep should not be interrupted")
+    t.interrupt()
     t.join()
-    assert(hasInterruptedException === false)
-    assert(interruptStatusBeforeExit)
+    assert(!hasInterruptedException, "runUninterruptibly should not be 
interrupted")
+    assert(interruptStatusBeforeExit, "interrupt flag should be set")
   }
 
   test("no runUninterruptibly") {
     @volatile var hasInterruptedException = false
-    val t = new UninterruptibleThread("test") {
+    @volatile var interruptStatusBeforeExit = false
+    val t = new UninterruptibleThread("run") {
       override def run(): Unit = {
-        if (sleep(0)) {
-          hasInterruptedException = true
-        }
+        hasInterruptedException = sleep(0)
+        interruptStatusBeforeExit = Thread.interrupted()
       }
     }
     t.interrupt()

Review Comment:
   All other implementations follow the reference implementation as well. This 
is why "I doubt it will make any difference for Java 11 and above"



##########
core/src/test/scala/org/apache/spark/util/UninterruptibleThreadSuite.scala:
##########
@@ -28,50 +28,69 @@ import org.apache.spark.SparkFunSuite
 
 class UninterruptibleThreadSuite extends SparkFunSuite {
 
-  /** Sleep millis and return true if it's interrupted */
+  /* Sleep millis and return true if it's interrupted */
   private def sleep(millis: Long): Boolean = {
     try {
       Thread.sleep(millis)
       false
     } catch {
-      case _: InterruptedException =>
+      case e: InterruptedException =>
+        log.error("Thread interrupted during sleep", e)
+        true
+    }
+  }
+
+  /* Await latch and return true if it's interrupted */
+  private def await(
+      latch: CountDownLatch,
+      timeout: Long = 10,
+      timeUnit: TimeUnit = TimeUnit.SECONDS): Boolean = {
+    try {
+      if (!latch.await(timeout, timeUnit)) {
+        log.error("timeout while waiting for the latch")
+        fail("timeout while waiting for the latch")
+      }
+      false
+    } catch {
+      case e: InterruptedException =>
+        log.error("Thread interrupted during await", e)
         true
     }
   }
 
   test("interrupt when runUninterruptibly is running") {
     val enterRunUninterruptibly = new CountDownLatch(1)
+    val interruptLatch = new CountDownLatch(1)
     @volatile var hasInterruptedException = false
     @volatile var interruptStatusBeforeExit = false
-    val t = new UninterruptibleThread("test") {
+    val t = new UninterruptibleThread("runUninterruptibly") {
       override def run(): Unit = {
         runUninterruptibly {
           enterRunUninterruptibly.countDown()
-          hasInterruptedException = sleep(1000)
+          hasInterruptedException = await(interruptLatch)
         }
         interruptStatusBeforeExit = Thread.interrupted()
       }
     }
     t.start()
-    assert(enterRunUninterruptibly.await(10, TimeUnit.SECONDS), "await 
timeout")
+    assert(!await(enterRunUninterruptibly), "await interrupted")
     t.interrupt()
+    interruptLatch.countDown()

Review Comment:
   OK, there is a tiny possibility that 2 context switches happen at exactly 
those locations, but there is a similar tiny possibility that there is no 
context switch to the main thread during `sleep(1000)` as well. It can be fixed 
by
   ```
   hasInterruptedException = await(interruptLatch) || Thread.interrupted()
   ```



##########
core/src/test/scala/org/apache/spark/util/UninterruptibleThreadSuite.scala:
##########
@@ -28,50 +28,67 @@ import org.apache.spark.SparkFunSuite
 
 class UninterruptibleThreadSuite extends SparkFunSuite {
 
-  /** Sleep millis and return true if it's interrupted */
+  /* Sleep millis and return true if it's interrupted */
   private def sleep(millis: Long): Boolean = {
     try {
       Thread.sleep(millis)
       false
     } catch {
-      case _: InterruptedException =>
+      case e: InterruptedException =>
+        log.error("Thread interrupted during sleep", e)
+        true
+    }
+  }
+
+  /* Await latch and return true if it's interrupted */
+  private def await(latch: CountDownLatch, timeout: Long = 10,
+                    timeUnit: TimeUnit = TimeUnit.SECONDS): Boolean = {
+    try {
+      if (!latch.await(timeout, timeUnit)) {
+        log.error("timeout while waiting for the latch")

Review Comment:
   > I would suggest to make it explicit 
   
   It is not clear what can be made explicit. I am not aware of any explicit 
way to fail a test on a thread other than the main test thread. Throwing an 
exception not on the main test thread causes the thread to be terminated. Test 
framework does not track that thread, and it does not fail the test if the 
thread terminates normally or abnormally (on exception). It does not matter 
what exception is thrown. This is why there are other checks on the main test 
thread that validate test logic should the thread terminate on exception.
   



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to