"Unable to get Token Metadata" returned by Spark Connector 3.1.0 when a stale entry in system.peers is chosen

Description

After an upgrade to Spark Connector 3.1.0 was necessary for Cassandra 4.x, if a stale/invalid entry was left in the system.peers table and was used, the following would be returned and the connection would fail:

The Driver stacktrace:

Walmart regularly has toplogy changes that leave behind stale system.peers entries, they need the connection to still succeed in this scenario.
This worked as expected in C* 3.11 with the 2.3.2 Connector

In case it matters, they set the following:
listen_address: <IP>

rpc_address: 0.0.0.0

broadcast_rpc_address: <IP>

Environment

Cassandra 4.0.1
Spark Connector 3.0.1

Pull Requests

None

Activity

Show:

Details

Assignee

Reporter

Labels

Affects versions

Priority

Created March 29, 2022 at 6:55 PM
Updated March 29, 2022 at 9:27 PM