Kafka topic not readable caused issues for login

[mooreds]

Kafka topic not readable caused issues for login

Description

A customer had kafka set up to listen to webhooks. They had it misconfigured and FusionAuth couldn't write to the kafka topic.

FusionAuth failed with a 502 error.

Affects versions

Saw it on 1.47.1.

Steps to reproduce

From a customer report.

1. Set up a kafka topic as a webhook receiver, but configure permissions so that FusionAuth can not write to that topic.
2. Go to the 'send test message' screen in FusionAuth.
3. Remove the 'request.timeout.ms' field from the producer configuration (this was an accident)
4. inspect the 'send the test event' button and save the request as curl
5. send the curl request via postman

That caused the issue. (I'm not sure that steps 4 and 5 are required but listing for completeness.)

Expected behavior

Failures of Kakfa topics should not affect the ability of users to log in.

Related

• Incorrect Kafka configuration can cause service degradation #2483

Community guidelines

All issues filed in this repository must abide by the FusionAuth community guidelines.

Additional context

Internal:

• https://inversoft.slack.com/archives/C03G6DMC0SV/p1694098416173359
• https://inversoft.slack.com/archives/C051S8N8E/p1694813777409519

[robotdan]

I am assuming this is all working as designed.

If you configure the login event to be transactional, and we can't publish an event to the Kafka topic, I would expect this to fail the login request.

If you have configured the login event to always succeed (No webhooks required to succeed) and the Kafka failure affects login, then this would be a bug.

Can you confirm how the login event is configured?

[mooreds]

@robotdan from the customer:

As far as I remember, we have marked the user.login.success to be non-transactional and have configure an non-writeable kafka topic (Kafka user was missing the permissions to write on the topic).

So this appears to fall into the second category mentioned above (the one where it is a bug).

[robotdan]

Internal:
We should never fail due to a Kafka failure - unless we are waiting to understand the result due to a TX configuration. If there is an edge case where we could fail a request as implied above, we need to fix.