redpanda startup fails to parse config after setting "redpanda.delete_retention_ms" to "null" using rpk

[larsfroelich]

Version & Environment

• Redpanda version: v21.11.11 (rev ace82c7)
• Operating System: Ubuntu 20.04.4 LTS

What went wrong?

After setting "redpanda.delete_retention_ms" to "null" using rpk, redpanda fails to startup with a parsing error application.cc:203 - Failure during startup: YAML::TypedBadConversion<std::__1::chrono::duration<long long, std::__1::ratio<1l, 1000l> > > (yaml-cpp: error std::__1::chrono::duration<long long, std::__1::ratio<1l, 1000l> > > (yaml-cpp: error at line 10, column 24: bad conversion)

What should have happened instead?

The redpanda.delete_retention_ms in /etc/redpanda/redpanda.yaml should be modified to value null (without quotation marks)

How to reproduce the issue?

1. Install redpanda as per docs
2. run sudo rpk config set redpanda.delete_retention_ms null
3. run systemctl restart redpanda

JIRA Link: CORE-878

[jcsp]

delete_retention_ms is not nullable, so the bug here is that it's allowed to be set to null at all.

In 22.1, there is a new configuration system (this would be set with "rpk cluster config set delete_retention_ms", which will properly prevent setting a value of the wrong type. However, this ticket makes me think we should also be improving rpk's "config set" command, which currently lets users set absolutely any value they like in the redpanda.yaml file (#4176)

[larsfroelich]

retention_bytes has a default value of null according to docs, so I wonder if the same parsing-error occurs there as well after setting it to null using sudo rpk config set redpanda.retention_bytes null (unable to test this at the moment)

[jcsp]

retention_bytes is nullable, delete_retention_ms is not. So null should work for retention_bytes