[
https://issues.apache.org/jira/browse/FLINK-10838?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Piotr Nowojski updated FLINK-10838:
-----------------------------------
Description:
Currently we have many tests that in order to test for example `at-least-once`,
they fail Kafka brokers, or brake network connection to them. It seems that
those tests are more testing Kafka brokers then our own code and in the process
(because of bugs in Kafka) cause lots of false positive results.
We could try to rewrite those tests to spawn one of the Flink's Task Managers
in separate process and instead of failing network/kafka broker we could
SIGKILL Task Manager, without touching/affecting Zookeeper/Kafka brokers. It
should improve stability of our tests.
CC [~till.rohrmann] [~aljoscha]
was:
Currently we have many tests that in order to test for example `at-least-once`,
they fail Kafka brokers, or brake network connection to them. It seems that
those tests are more testing Kafka brokers then our own code and in the process
(because of bugs in Kafka) cause lots of false positive results.
We could try to rewrite those tests to spawn one of the Flink's Task Managers
in separate process and instead of failing network/kafka broker we could
SIGKILL Task Manager, without touching/affecting Zookeeper/Kafka brokers. It
should improve stability of our tests.
> Rewrite Kafka tests that fail Kafka brokers
> -------------------------------------------
>
> Key: FLINK-10838
> URL: https://issues.apache.org/jira/browse/FLINK-10838
> Project: Flink
> Issue Type: Improvement
> Components: Kafka Connector
> Affects Versions: 1.7.0
> Reporter: Piotr Nowojski
> Priority: Major
>
> Currently we have many tests that in order to test for example
> `at-least-once`, they fail Kafka brokers, or brake network connection to
> them. It seems that those tests are more testing Kafka brokers then our own
> code and in the process (because of bugs in Kafka) cause lots of false
> positive results.
> We could try to rewrite those tests to spawn one of the Flink's Task Managers
> in separate process and instead of failing network/kafka broker we could
> SIGKILL Task Manager, without touching/affecting Zookeeper/Kafka brokers. It
> should improve stability of our tests.
> CC [~till.rohrmann] [~aljoscha]
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
