[CASSANDRA-17054] v4+ protocol did not clean up client warnings, which caused leaking the state

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 4.1-alpha1, 4.1
• Component/s: Messaging/Client
• Bug Category: Correctness - Transient Incorrect Response
• Severity: Normal
• Complexity: Low Hanging Fruit
• Discovered By: User Report
• Platform: All
• Since Version: 4.0.0

Source Control Link

https://github.com/apache/cassandra/commit/d8fff9b2d0e79fa8d945a8cdc1295fda3deae77c

Test and Documentation Plan

jvm-dtest

• Assignee: David Capwell
• Reporter: David Capwell

Authors

David Capwell

Reviewers

Caleb Rackliffe, Jon Meredith, Sam Tunnicliffe

• Created: 21/Oct/21 18:15
• Updated: 27/May/22 19:24
• Resolved: 25/Oct/21 15:44

Description

If you perform a query in v5, this will cause the STARTUP message to be handled in the netty thread, but the way this is done is by calling an internal API to dispatcher which requires the caller to clean up; but we do not clean up; at this point the netty thread will have a ClientWarn state populated. If you now perform the same query again, but with the v3 protocol, this will pick up the state and try to serialize it, causing a client error (in java as java rejects the output from the server) saying that v3 may not have client warnings.

Issue Links

• links to: GH PR: trunk