Chris Egerton created KAFKA-9560:
------------------------------------

             Summary: Connector::validate is utilized concurrently by the 
framework, but not documented as thread-safe
                 Key: KAFKA-9560
                 URL: https://issues.apache.org/jira/browse/KAFKA-9560
             Project: Kafka
          Issue Type: Bug
          Components: KafkaConnect
            Reporter: Chris Egerton


Requests to the {{PUT /connector-plugins/\{connectorType}/config/validate}} 
endpoint are [delegated to the 
herder|https://github.com/apache/kafka/blob/16ee326755e3f13914a0ed446c34c84e65fc0bc4/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/rest/resources/ConnectorPluginsResource.java#L81],
 which [caches connector 
instances|https://github.com/apache/kafka/blob/16ee326755e3f13914a0ed446c34c84e65fc0bc4/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/AbstractHerder.java#L536-L544]
 that are used [during config 
validation|https://github.com/apache/kafka/blob/16ee326755e3f13914a0ed446c34c84e65fc0bc4/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/AbstractHerder.java#L310].
 This has the effect that, should concurrent requests to that endpoint occur 
for the same connector type, the same connector instance may be responsible for 
[validating those 
configurations|https://github.com/apache/kafka/blob/16ee326755e3f13914a0ed446c34c84e65fc0bc4/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/AbstractHerder.java#L334]
 concurrently _(may_ instead of _will_ because there is also a race condition 
in the {{AbstractHerder::getConnector}} method that potentially fails to detect 
that an instance of the connector has already been created and, as a result, 
creates a second instance).

This is slightly problematic because the 
[Connector::validate|https://github.com/apache/kafka/blob/16ee326755e3f13914a0ed446c34c84e65fc0bc4/connect/api/src/main/java/org/apache/kafka/connect/connector/Connector.java#L122-L127]
 method is not marked as thread-safe. However, because a lot of connectors out 
there tend to use the default implementation for that method, it's probably not 
super urgent that we patch this immediately.

A couple of options are:
 # Update the docs for that method to specify that it must be thread-safe
 # Rewrite the connector validation logic in the framework to avoid 
concurrently invoking {{Connector::validate}} on the same instance.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to