You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@kafka.apache.org by "Nikita Shupletsov (Jira)" <ji...@apache.org> on 2026/01/13 00:39:00 UTC

[jira] [Created] (KAFKA-20064) A race condition in admin client can lead to the result future never being completed

Nikita Shupletsov created KAFKA-20064:
-----------------------------------------

             Summary: A race condition in admin client can lead to the result future never being completed
                 Key: KAFKA-20064
                 URL: https://issues.apache.org/jira/browse/KAFKA-20064
             Project: Kafka
          Issue Type: Bug
          Components: clients
    Affects Versions: 4.0.0
            Reporter: Nikita Shupletsov
            Assignee: Nikita Shupletsov
         Attachments: A_test_that_reproduces_the_issue.patch

Here we make two independent calls that check the cache: [https://github.com/apache/kafka/blob/trunk/clients/src/main/java/org/apache/kafka/clients/admin/internals/AdminApiDriver.java#L118-L119]

if the admin client is shared between multiple threads, there may be a situation when one request is initializing the AdminApiDriver, and the other one invalidates some values in the cache.

E.g.:
There are two partitions in the cache: 1,2. and two requests requesting offsets for both partitions.
Thread1: calls uncachedLookupKeys, which returns an empty list, as both requested values are in the cache.
Thread2: faced an issue with connecting to a node and invalidates the cache for partition 2.
Thread1: calls cachedKeyBrokerIdMapping, which returns one 1.
The future for partition 2 will never be completed, as we never created a request for it.

An integration test that stages all these conditions and reproduces the issue is in the attachment



--
This message was sent by Atlassian Jira
(v8.20.10#820010)