Is it possible to skip mapping configuration for some env vars?

[ppatierno]

Hi all,
I have the following KafkaConfig class in our Strimzi HTTP bridge (based on Quarkus) https://github.com/ppatierno/strimzi-kafka-bridge/blob/road-to-quarkus/src/main/java/io/strimzi/kafka/bridge/config/KafkaConfig.java

With this class we are able to get configuration from an application.properties file like this https://github.com/ppatierno/strimzi-kafka-bridge/blob/road-to-quarkus/config/application.properties

You can see we are mapping kafka.producer., kafka.consumer. and kafka.admin. prefixed properties to corresponding maps into the KafkaConfig. Also everything under just kafka. (which doesn't match above prefixes) goes to common map.

The problem I am facing is that the environment where this application runs sets some env vars (used for a different purpose) starting with KAFKA_BRIDGE_ so the Quarkus configuration is trying to map them into the KafkaConfig.common map ... which doesn't make any sense for our application.
Furthermore, some of these env vars could be empty and the Quarkus application doesn't start at all because of error like this (taking KAFKA_BRIDGE_PRODUCER_CONFIG env var as an example):

ERROR: Failed to start application (with profile [prod])
java.util.NoSuchElementException: SRCFG00040: The config property kafka.bridge.producer.config is defined as the empty String ("") which the following Converter considered to be null: io.smallrye.config.Converters$BuiltInConverter

So my question is ... is there any way to avoid some prefixed env vars to be mapped (i.e. all starting with KAFKA_BRIDGE_) or disabling the env var config source at all (because we don't need that)?
I tried to skip KAFKA_BRIDGE_ prefixed env vars by using a config relocator returning just null when it's queried for properties this way:

@Override
public ConfigValue getValue(ConfigSourceInterceptorContext context, String name) {
...
...
   if (name.startsWith("kafka.bridge.")) {
            return null;
   } 
}

I also tried a really BAD hack by adding trash "bucket" (actually a Map) into the KafkaConfig class to get all these env vars, but it doesn't work as well.
Any suggestions? (other then renaming the env vars in our environment to avoid them starting with KAFKA_BRIDGE_ prefix ;-))

[radcortez]

Currently, there is no way to do it.

When mapping properties, the Config system doesn't care from where the config property is coming from.

I've suggested the interceptor and remove the properties. Still, unfortunately, that doesn't work because we always apply another interceptor on top (no matter the priority), to convert environment variables to their dotted format.

The bucket solution should work, but remember to quote the kafka.bridge in the interceptor. Also, you need to add a converter to handle empty values, because we always expect to have a value so you can do something like:

    Map<String, @WithConverter(IgnoreConverter.class) String> bridge();

    class IgnoreConverter implements Converter<String> {
        @Override
        public String convert(final String s) throws IllegalArgumentException, NullPointerException {
            return "ignored";
        }
    }

[ppatierno]

Thanks Roberto, even if I don't think this is a really great way to fix this :-(
Would it make sense to fill an issue about this use case, allowing something like an "exclusion" support where you can specify what you don't want to be mapped during configuration process?

[radcortez]

That is already supported at the builder level, but the API is unavailable for the user (at least in Quarkus). I may think of a way to expose it, but it will take some time because it involves changes in both SmallRye Config and Quarkus. Unfortunately, I also have a few other priorities.

[ppatierno]

It's not a higher priority thing. I was suggesting to open an issue to track it as an enhancement which seems to make sense.

[radcortez]

Sure, go ahead. Thank you!

[ppatierno]

FYI #33592, Thanks!