Kafka pollable consumers lose ConsumerFactory listeners when max.poll.records is greater than one

Đang mở
  • Có gì không ổn?
#3,324 2 bình luận 0 reaction 0 người được giao Xem trên GitHub

Maintainer thường phản hồi trong vòng 1 ngày

Chưa có ai nhận issue này.

Đánh giá

Độ khó
4/5
Thời gian dự kiến
3-5 ngày
Mức phù hợp với người mới
55/100
Loại issue
Lỗi
Độ rõ ràng
Khá rõ ràng
Mức độ hoạt động
Sôi nổi
Công nghệ
java, kafka, spring
Lĩnh vực
backend, distributed-systems, observability

Hướng nghiên cứu

Start by reading Spring Integration Kafka's KafkaMessageSource, where the issue says a replacement DefaultKafkaConsumerFactory is created when max.poll.records is missing or greater than 1. Check how that replacement handles the original factory's listeners, then run the linked reproducer's SpringCloudStreamPollableKafkaMetricsTests to compare the two configurations. Done means the listener is preserved and native kafka.consumer.* meters register with either setting.

Do mô hình lập chỉ mục viết ra từ nội dung của issue.

Mô tả

Thanks for considering raising a Spring Cloud Stream issue.

Describe the issue
Kafka pollable consumers lose ConsumerFactory.Listener instances when max.poll.records is not set to 1.

This prevents Micrometer native Kafka metrics from being registered for the consumer created by KafkaMessageSource.

The issue is visible through Spring Cloud Stream, but the suspected root cause appears to be in Spring Integration Kafka's KafkaMessageSource.

To Reproduce
The issue can be reproduced with this standalone demo application:

https://github.com/ferblaca/demoStreamPollable

The reproducer uses:

  • Spring Boot 4.1.2;
  • Spring Cloud Stream 5.0.4;
  • the Spring Cloud Stream Kafka binder;
  • a real Kafka broker started with Testcontainers;
  • two Spring Cloud Stream pollable consumers.

Steps:

  1. Clone the repository:

    git clone https://github.com/ferblaca/demoStreamPollable.git
cd demoStreamPollable

  2. Run the reproducer test:

    ./mvnw -B -Dtest=SpringCloudStreamPollableKafkaMetricsTests test

  3. Review the test implementation:

    src/test/java/com/example/demoStreamPollable/SpringCloudStreamPollableKafkaMetricsTests.java

  4. The test creates two pollable bindings:

    pollableInput-in-0
pollableInputMaxOne-in-0

  5. The first binding is configured with:

    spring.cloud.stream.kafka.bindings.pollableInput-in-0.consumer.configuration.max.poll.records=10

    The test publishes and consumes a real Kafka message through PollableMessageSource, then verifies that no meter whose name starts with kafka.consumer is registered.

  6. The second binding is configured with:

    spring.cloud.stream.kafka.bindings.pollableInputMaxOne-in-0.consumer.configuration.max.poll.records=1

    The test publishes and consumes another real Kafka message through a second PollableMessageSource, then verifies that a meter whose name starts with kafka.consumer is registered.

The test passes and demonstrates the different behavior between both configurations.

Version of the framework

  • Spring Boot 4.1.2
  • Spring Cloud Stream 5.0.4
  • Java 25
  • Kafka broker: Testcontainers apache/kafka-native:3.8.0

Expected behavior
When KafkaMessageSource creates a replacement consumer factory, it should preserve the behavior of the supplied factory, including registered ConsumerFactory.Listener instances such as:

MicrometerConsumerListener

Native kafka.consumer.* metrics should be registered regardless of whether max.poll.records is explicitly set to 1.

Screenshots
Not applicable.

Additional context
Spring Cloud Stream's Kafka binder creates the consumer factory and invokes the registered ClientFactoryCustomizer before creating the pollable source.

The relevant flow is:

KafkaMessageChannelBinder
  -> createPolledConsumerResources(...)
  -> createKafkaConsumerFactory(...)
  -> ClientFactoryCustomizer.configure(consumerFactory)
  -> new KafkaMessageSource<>(consumerFactory, consumerProperties)

The customizer adds a MicrometerConsumerListener to the original DefaultKafkaConsumerFactory.

However, when max.poll.records is missing or greater than 1, KafkaMessageSource forces it to 1 by creating a new DefaultKafkaConsumerFactory.

The replacement factory copies the configuration and deserializers, but does not copy the listeners from the original factory. Consequently, the MicrometerConsumerListener is lost before the actual Kafka consumer is created.

The current behavior also logs:

max.poll.records' has been forced from 10 to 1, to avoid having to seek after each record

As a workaround, applications can configure max.poll.records=1 for each pollable Kafka binding. This prevents KafkaMessageSource from creating the replacement factory, but it exposes an internal implementation detail to applications.

Could you please confirm whether this issue should be fixed in:

  • Spring Integration Kafka, in KafkaMessageSource;
  • Spring Cloud Stream's Kafka binder;
  • or both?

The suspected fix is to copy the ConsumerFactory.Listener instances when KafkaMessageSource creates the replacement DefaultKafkaConsumerFactory. The reproducer is intentionally implemented through Spring Cloud Stream so that the behavior can be verified using a real PollableMessageSource and the Kafka binder, but the suspected defect appears to belong to Spring Integration Kafka.

Ngôn ngữ chính
Java
Star
1.1k
Fork
651
Merge trung bình
2 ngày 15 giờ
Pull request đã merge (30 ngày)
8

Chuẩn bị môi trường

Dự án này không cung cấp dev container, Dockerfile hay hướng dẫn đóng góp, nên bạn cần tự thiết lập môi trường: hãy bắt đầu từ README và xem hướng dẫn đóng góp lần đầu của chúng tôi để biết các bước chung.

Bắt đầu từ đâu

  1. Đọc hết issue, rồi đọc hướng dẫn đóng góp của dự án.
  2. Bình luận trên issue rằng bạn sẽ nhận — tránh hai người làm cùng một việc.
  3. Fork repository và làm thay đổi trên một nhánh.
  4. Mở pull request có tham chiếu số hiệu của issue.

Issue khác của spring-cloud/spring-cloud-stream

Tất cả issue của spring-cloud/spring-cloud-stream

Issue tương tự

Thêm issue về Java

Nhận issue mới trong hộp thư của bạn

Bản tóm tắt ngắn những issue GitHub phù hợp với người mới.