voonhous commented on code in PR #19485:
URL: https://github.com/apache/hudi/pull/19485#discussion_r3956277289
##########
hudi-utilities/src/test/java/org/apache/hudi/utilities/deltastreamer/HoodieDeltaStreamerTestBase.java:
##########
@@ -763,34 +776,110 @@ static HoodieInstant
assertCommitMetadataForIncrSource(String expected, String t
return lastInstant;
}
+ /**
+ * Polls {@code condition} until it holds, the deltastreamer future
finishes, or the timeout expires.
+ *
+ * <p>On timeout the last error the condition threw is attached to the
failure, so the report names the
+ * assertion that never held rather than only this method.
+ */
static void waitTillCondition(Function<Boolean, Boolean> condition, Future
dsFuture, long timeoutInSecs) throws Exception {
- Future<Boolean> res = Executors.newSingleThreadExecutor().submit(() -> {
- boolean ret = false;
- while (!ret && !dsFuture.isDone()) {
- try {
- Thread.sleep(2000);
- ret = condition.apply(true);
- log.info("Condition completed successfully");
- } catch (Throwable error) {
- log.debug("Got error waiting for condition", error);
- ret = false;
+ AtomicReference<Throwable> lastError = new AtomicReference<>();
+ AtomicInteger completedEvaluations = new AtomicInteger();
+ ExecutorService executor = Executors.newSingleThreadExecutor();
+ try {
+ Future<Boolean> res = executor.submit(() -> {
+ boolean ret = false;
+ // The executor check matters as well as the interrupt flag: the
interrupt from shutdownNow is
+ // delivered once, and a condition that swallows it would otherwise
leave the flag clear and keep
+ // this thread polling for the lifetime of the JVM.
+ while (!ret && !dsFuture.isDone() &&
!Thread.currentThread().isInterrupted() && !executor.isShutdown()) {
+ try {
+ Thread.sleep(POLL_INTERVAL_MS);
+ ret = condition.apply(true);
+ completedEvaluations.incrementAndGet();
+ if (ret) {
+ log.info("Condition completed successfully");
+ }
+ } catch (InterruptedException interrupted) {
+ // Thread.sleep clears the interrupt flag when it throws, so
catching this with everything
+ // else would re-enter the loop. Restore the flag and stop; this
is not a condition failure,
+ // so it is deliberately not recorded as one.
+ Thread.currentThread().interrupt();
+ break;
+ } catch (Throwable error) {
+ log.debug("Got error waiting for condition", error);
+ lastError.set(error);
+ completedEvaluations.incrementAndGet();
+ ret = false;
+ }
+ }
+ return ret;
+ });
+ try {
+ res.get(timeoutInSecs, TimeUnit.SECONDS);
Review Comment:
Fixed in `9f3cd75`. The result of `res.get` is captured, and the
`dsFuture.isDone()` exit logs the same detail at warn through a new
`describeProgress`, which `describeTimeout` also calls, so all three existing
messages are byte-for-byte unchanged.
##########
hudi-utilities/src/test/java/org/apache/hudi/utilities/deltastreamer/HoodieDeltaStreamerTestBase.java:
##########
@@ -763,34 +776,110 @@ static HoodieInstant
assertCommitMetadataForIncrSource(String expected, String t
return lastInstant;
}
+ /**
+ * Polls {@code condition} until it holds, the deltastreamer future
finishes, or the timeout expires.
+ *
+ * <p>On timeout the last error the condition threw is attached to the
failure, so the report names the
+ * assertion that never held rather than only this method.
+ */
static void waitTillCondition(Function<Boolean, Boolean> condition, Future
dsFuture, long timeoutInSecs) throws Exception {
- Future<Boolean> res = Executors.newSingleThreadExecutor().submit(() -> {
- boolean ret = false;
- while (!ret && !dsFuture.isDone()) {
- try {
- Thread.sleep(2000);
- ret = condition.apply(true);
- log.info("Condition completed successfully");
- } catch (Throwable error) {
- log.debug("Got error waiting for condition", error);
- ret = false;
+ AtomicReference<Throwable> lastError = new AtomicReference<>();
+ AtomicInteger completedEvaluations = new AtomicInteger();
+ ExecutorService executor = Executors.newSingleThreadExecutor();
+ try {
+ Future<Boolean> res = executor.submit(() -> {
+ boolean ret = false;
+ // The executor check matters as well as the interrupt flag: the
interrupt from shutdownNow is
+ // delivered once, and a condition that swallows it would otherwise
leave the flag clear and keep
+ // this thread polling for the lifetime of the JVM.
+ while (!ret && !dsFuture.isDone() &&
!Thread.currentThread().isInterrupted() && !executor.isShutdown()) {
+ try {
+ Thread.sleep(POLL_INTERVAL_MS);
+ ret = condition.apply(true);
+ completedEvaluations.incrementAndGet();
+ if (ret) {
+ log.info("Condition completed successfully");
+ }
+ } catch (InterruptedException interrupted) {
+ // Thread.sleep clears the interrupt flag when it throws, so
catching this with everything
+ // else would re-enter the loop. Restore the flag and stop; this
is not a condition failure,
+ // so it is deliberately not recorded as one.
+ Thread.currentThread().interrupt();
+ break;
+ } catch (Throwable error) {
+ log.debug("Got error waiting for condition", error);
+ lastError.set(error);
+ completedEvaluations.incrementAndGet();
+ ret = false;
+ }
+ }
+ return ret;
+ });
+ try {
+ res.get(timeoutInSecs, TimeUnit.SECONDS);
+ } catch (TimeoutException e) {
+ Throwable last = lastError.get();
+ Throwable cause = last == null ? e : last;
+ AssertionError failure = new AssertionError(
+ describeTimeout(last, completedEvaluations.get(),
timeoutInSecs), cause);
Review Comment:
Fixed in `9f3cd75`, the counter is read first. The comment now records why:
the skew then lands in the `last != null` branch, reporting a real error with a
possibly stale count, rather than in the branch that would deny an error that
did happen.
--
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]
