chia7712 commented on code in PR #20334:
URL: https://github.com/apache/kafka/pull/20334#discussion_r2651526623
##########
docs/upgrade.html:
##########
@@ -113,6 +113,35 @@ <h5><a id="upgrade_420_notable"
href="#upgrade_420_notable">Notable changes in 4
<li>
The <code>num.replica.fetchers</code> config has a new lower bound of
1.
</li>
+ <li>
+ Improvements have been made to the validation rules and default values
of LIST-type configurations
+ (<a href="https://cwiki.apache.org/confluence/x/HArXF";>KIP-1161</a>).
+ <ul>
+ <li>
+ LIST-type configurations now enforce stricter validation:
+ <ul>
+ <li>Null values are no longer accepted for most LIST-type
configurations, except those that explicitly
+ allow a null default value or where a null value has a
well-defined semantic meaning.</li>
+ <li>Duplicate entries within the same list are no longer
permitted.</li>
Review Comment:
The scenario is more complex than I initially thought
If `log.segment.bytes` is configured to a new value and is stored in the
snapshot file, a broker with stricter validation will fail. This is because
`DynamicBrokerConfig.readDynamicBrokerConfigsFromSnapshot` loads the dynamic
configs before starting the log components.
By contrast, the broker can start up successfully if that small value is not
yet in the snapshot file, since the log components won't see the latest
`log.segment.bytes` during startup
In short, the handling of dynamic configs is already inconsistent :cry:
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
