Copilot commented on code in PR #13449:
URL: https://github.com/apache/gluten/pull/13449#discussion_r4513729743


##########
gluten-ut/native/KafkaPollTimeoutTest.cc:
##########
@@ -0,0 +1,301 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <algorithm>
+#include <chrono>
+#include <cstdint>
+#include <deque>
+#include <limits>
+#include <memory>
+#include <string>
+#include <unordered_map>
+#include <utility>
+#include <vector>
+
+#include <gtest/gtest.h>
+#include <librdkafka/rdkafka.h>
+#include <librdkafka/rdkafka_mock.h>
+#include <librdkafka/rdkafkacpp.h>
+
+#include "operators/reader/KafkaConnector.h"
+#include "operators/reader/KafkaDataSource.h"
+#include "velox/common/base/tests/GTestUtils.h"
+#include "velox/vector/tests/utils/VectorTestBase.h"
+
+using namespace facebook::velox;
+
+namespace gluten {
+namespace {
+
+constexpr char kConnectorId[] = "kafka-poll-timeout-test";
+constexpr char kTopic[] = "kafka-poll-timeout";
+constexpr int64_t kPollTimeoutMs = 10;
+
+} // namespace
+
+class KafkaPollTimeoutTest : public ::testing::Test,
+                             public test::VectorTestBase {
+protected:
+  static void SetUpTestCase() {
+    
memory::MemoryManager::testingSetInstance(memory::MemoryManager::Options{});
+  }
+
+  void SetUp() override {
+    auto *conf = rd_kafka_conf_new();
+    char error[512];
+    ASSERT_EQ(rd_kafka_conf_set(conf, "test.mock.num.brokers", "1", error,
+                                sizeof(error)),
+              RD_KAFKA_CONF_OK);
+    producer_ = rd_kafka_new(RD_KAFKA_PRODUCER, conf, error, sizeof(error));
+    ASSERT_NE(producer_, nullptr) << error;
+    cluster_ = rd_kafka_handle_mock_cluster(producer_);
+    ASSERT_NE(cluster_, nullptr);
+    ASSERT_EQ(rd_kafka_mock_topic_create(cluster_, kTopic, 1, 1),
+              RD_KAFKA_RESP_ERR_NO_ERROR);
+    // Keep broker responses outside both the configured timeout and the old 
100
+    // ms consume wait.
+    ASSERT_EQ(rd_kafka_mock_broker_set_rtt(cluster_, 1, 1'000),
+              RD_KAFKA_RESP_ERR_NO_ERROR);
+  }
+
+  void TearDown() override {
+    if (producer_ != nullptr) {
+      rd_kafka_destroy(producer_);
+    }
+  }
+
+  std::unique_ptr<KafkaDataSource> source() {
+    auto type = ROW({"offset"}, {BIGINT()});
+    connector::ColumnHandleMap columns{
+        {"offset", std::make_shared<KafkaColumnHandle>("offset", BIGINT())}};
+    return std::make_unique<KafkaDataSource>(type, columns, pool());
+  }
+
+  std::shared_ptr<KafkaConnectorSplit> split(int64_t endOffset) {
+    std::unordered_map<std::string, std::string> params{
+        {"bootstrap.servers", rd_kafka_mock_cluster_bootstraps(cluster_)}};
+    return std::make_shared<KafkaConnectorSplit>(kConnectorId, kTopic, 0, 0,
+                                                 endOffset, kPollTimeoutMs,
+                                                 true, std::move(params));
+  }
+
+  rd_kafka_t *producer_{nullptr};
+  rd_kafka_mock_cluster_t *cluster_{nullptr};
+};
+
+TEST_F(KafkaPollTimeoutTest, shortPollTimeoutBoundsConsumerWait) {
+  int64_t minimumElapsedMs = std::numeric_limits<int64_t>::max();
+  // Use the fastest of three independent readers so an isolated scheduler 
delay
+  // cannot fail the test. Every old-code attempt waits at least 100 ms; the
+  // configured timeout is only 10 ms.
+  for (int attempt = 0; attempt < 3; ++attempt) {
+    auto reader = source();
+    reader->addSplit(split(1));
+    ContinueFuture future = ContinueFuture::makeEmpty();
+    const auto start = std::chrono::steady_clock::now();
+    VELOX_ASSERT_THROW(reader->next(1, future),
+                       "Timed out after 10 ms fetching offset");
+    const int64_t elapsedMs =
+        std::chrono::duration_cast<std::chrono::milliseconds>(
+            std::chrono::steady_clock::now() - start)
+            .count();
+    minimumElapsedMs = std::min(minimumElapsedMs, elapsedMs);
+    EXPECT_EQ(reader->getCompletedRows(), 0);
+    EXPECT_EQ(reader->getCompletedBytes(), 0);
+    // Consumer teardown is deliberately outside the measured poll interval.
+  }
+  RecordProperty("minimum_poll_elapsed_ms", std::to_string(minimumElapsedMs));
+  EXPECT_LT(minimumElapsedMs, 80)
+      << "A 10 ms poll must not use the fixed 100 ms consumer wait";

Review Comment:
   This test relies on wall-clock timing with a hard threshold (`< 80ms`). Even 
with the 3-attempt minimum, it can still be flaky under CI load / scheduling 
stalls. Since the PR already introduces fully deterministic scripted tests that 
assert exact consume timeout arguments, consider removing this wall-clock 
assertion or rewriting it to validate behavior without elapsed-time thresholds 
(e.g., by asserting the actual consume timeout argument via a controllable 
consumer hook).



##########
cpp/velox/operators/reader/KafkaDataSource.cc:
##########
@@ -194,14 +194,42 @@ std::optional<RowVectorPtr> 
KafkaDataSource::next(uint64_t size, ContinueFuture&
   return nullptr;
 }
 
+std::chrono::steady_clock::time_point KafkaDataSource::pollTime() const {
+  return std::chrono::steady_clock::now();
+}
+
+RdKafka::Message* KafkaDataSource::consume(int timeoutMs) {
+  return consumer_->consume(timeoutMs);
+}
+
 std::vector<std::unique_ptr<RdKafka::Message>> KafkaDataSource::poll(uint64_t 
maxRows) {
   std::vector<std::unique_ptr<RdKafka::Message>> messages;
   const auto pollTimeoutMs = split_->pollTimeoutMs > 0 ? split_->pollTimeoutMs 
: kDefaultPollTimeoutMs;
-  const auto deadline = std::chrono::steady_clock::now() + 
std::chrono::milliseconds(pollTimeoutMs);
+  const auto deadline = pollTime() + std::chrono::milliseconds(pollTimeoutMs);
+  const auto failOnTimeout = [&]() {
+    VELOX_FAIL(
+        "Timed out after {} ms fetching offset {} of {}-{}",
+        pollTimeoutMs,
+        nextOffset_,
+        split_->topic,
+        split_->partition);
+  };
+  // Bound record-consumption waits by the time left in this poll. Metadata 
and recovery
+  // operations retain their separate timeout allowances.
+  const auto remainingTimeoutMs = [this, &deadline]() {
+    return static_cast<int>(std::clamp<int64_t>(
+        std::chrono::ceil<std::chrono::milliseconds>(deadline - 
pollTime()).count(),

Review Comment:
   `std::chrono::ceil` can round *up* the remaining duration, producing a 
`timeoutMs` that is larger than the true remaining budget (by up to ~0.999ms). 
That undermines the goal of strictly capping consume waits by remaining poll 
time. Consider using truncation/flooring (e.g., 
`duration_cast<std::chrono::milliseconds>`) so the computed timeout never 
exceeds the remaining time; if that yields 0 for sub-millisecond remainder, 
it’s preferable to exit early rather than potentially blocking past the 
deadline.



##########
cpp/velox/operators/reader/KafkaDataSource.h:
##########
@@ -64,6 +65,12 @@ class KafkaDataSource : public 
facebook::velox::connector::DataSource {
 
  private:
   friend class KafkaConnectorTest;
+  friend class KafkaPollTimeoutArgumentTest;
+
+  // Keep time and record consumption replaceable for deterministic timeout 
tests.
+  virtual std::chrono::steady_clock::time_point pollTime() const;
+
+  virtual RdKafka::Message* consume(int timeoutMs);
 

Review Comment:
   Adding a test friend plus private virtual hooks increases coupling between 
tests and KafkaDataSource internals (tests can reach private `poll()` / state). 
To reduce refactor friction, consider making only the minimal test hooks 
`protected` (e.g., `pollTime()`/`consume()` and/or small `protected` accessors 
for `nextOffset_`/`splitDone_`) and avoid granting broad friendship to the 
entire test fixture where possible.



-- 
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