[CASSANDRA-15677] Topology events are not sent to clients if the nodes use the same network interface

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 4.0-beta1, 4.0
• Component/s: Messaging/Client
• Labels: pull-request-available
• Bug Category: Correctness
• Severity: Normal
• Complexity: Normal
• Discovered By: User Report
• Platform: All
• Since Version: 4.0-alpha1

Source Control Link

https://github.com/apache/cassandra/commit/6317fefec80c4c35aa588a6b610c25630df656f6

Test and Documentation Plan

Added a new in-jvm-dtests

Updated existing python dTests

 

• Assignee: Bryn Cooke
• Reporter: Alan Boudreault

Authors

Bryn Cooke

Reviewers

Brandon Williams

• Created: 30/Mar/20 18:26
• Updated: 08/Jun/21 00:58
• Resolved: 22/Jun/20 14:19

Description

This bug only happens when the cassandra nodes are configured to use a single network interface (ip) but different ports.  See CASSANDRA-7544.

Issue: The topology events aren't sent to clients. The problem is that the port is not taken into account when determining if we send it or not:
https://github.com/apache/cassandra/blob/trunk/src/java/org/apache/cassandra/transport/Server.java#L624

To reproduce:

# I think the cassandra-test branch is required to get the -S option (USE_SINGLE_INTERFACE)
ccm create -n4 local40 -v 4.0-alpha2 -S

 
Then run this small python driver script:

import time
from cassandra.cluster import Cluster

cluster = Cluster()
session = cluster.connect()

while True:
    print(cluster.metadata.all_hosts())
    print([h.is_up for h in cluster.metadata.all_hosts()])
    time.sleep(5)

Then decommission a node:

ccm node2 nodetool disablebinary
ccm node2 nodetool decommission

 
You should see that the node is never removed from the client cluster metadata and the reconnector started.

Issue Links

• causes: CASSANDRA-16234 Update NetBeans project file for dependency changes since 11th Feb 2020
• is duplicated by: CASSANDRA-15886 pushed_notifications_test.TestPushedNotifications.test_restart_node_localhost failure
• is related to: CASSANDRA-15912 Occasional NEW_NODE event sent instead of NODE_UP in dTest
• links to: DTest GitHub Pull Request #79
• links to: GitHub Pull Request #617
• links to: GitHub Pull Request #647
• links to: Java8 CI Run
• links to: Java 11 CI Run