Reduce allocation in matching of unknown properties

[radcortez]

Use io.smallrye.config.PropertyName instead of io.smallrye.config.KeyMap. io.smallrye.config.PropertyName requires less allocations and it is faster when matching property names, because it does not need to split the String.

• Fixes Increased build-time RSS usage in micrometer-prometheus integration test after moving Kubernetes to @ConfigMapping #44216

[gsmet]

I'm a bit worried about the test failures. But unfortunately there's nothing in the log about the 500 errors that the client encounters.

[gsmet]

I asked for a rebase.

[radcortez]

I'm a bit worried about the test failures. But unfortunately there's nothing in the log about the 500 errors that the client encounters.

Yes, I've executed this locally and got the same error. Apparently, they are related. I'll have a look.

[radcortez]

The issue was caused by the runtime property quarkus.rest-client.logging.scope could also match the build time property quarkus.rest-client.[name].scope.

I did change the logic in the recording to skip check in other type phases once a match is found, but we need to check all phases.

[gsmet]

I think the native test failing is probably still a problem.

[radcortez]

Didn't notice. Let me have a look.

[radcortez]

The issue was caused by * properties not having max priority when inserted into the Set. We do this in SmallRyeConfig, but it is an internal API, and I forgot to add it. For now, it stays here, but I'll provide a proper API for it in SmallRyeConfig.

[quarkus-bot]

Status for workflow Quarkus CI

This is the status report for running Quarkus CI on commit a04c136.

✅ The latest workflow run for the pull request has completed successfully.

It should be safe to merge provided you have a look at the other checks in the summary.

You can consult the Develocity build scans.

---

Flaky tests - Develocity

⚙️ JVM Tests - JDK 17 Windows

📦 integration-tests/grpc-hibernate

✖ com.example.grpc.hibernate.BlockingRawTest.shouldAdd - History

• Condition with Lambda expression in com.example.grpc.hibernate.BlockingRawTestBase was not fulfilled within 30 seconds. - org.awaitility.core.ConditionTimeoutException

org.awaitility.core.ConditionTimeoutException: Condition with Lambda expression in com.example.grpc.hibernate.BlockingRawTestBase was not fulfilled within 30 seconds.
	at org.awaitility.core.ConditionAwaiter.await(ConditionAwaiter.java:167)
	at org.awaitility.core.CallableCondition.await(CallableCondition.java:78)
	at org.awaitility.core.CallableCondition.await(CallableCondition.java:26)
	at org.awaitility.core.ConditionFactory.until(ConditionFactory.java:1006)
	at org.awaitility.core.ConditionFactory.until(ConditionFactory.java:975)
	at com.example.grpc.hibernate.BlockingRawTestBase.shouldAdd(BlockingRawTestBase.java:59)
	at java.base/java.lang.reflect.Method.invoke(Method.java:569)

---

⚙️ Native Tests - HTTP

📦 integration-tests/rest-client-reactive

✖ io.quarkus.it.rest.client.BasicTestIT.shouldCreateClientSpans - History

• expected: <1> but was: <2> - org.opentest4j.AssertionFailedError

org.opentest4j.AssertionFailedError: expected: <1> but was: <2>
	at io.quarkus.it.rest.client.BasicTest.shouldCreateClientSpans(BasicTest.java:216)
	at java.base/java.lang.reflect.Method.invoke(Method.java:569)
	at io.quarkus.test.junit.QuarkusTestExtension.interceptTestMethod(QuarkusTestExtension.java:805)
	at java.base/java.util.ArrayList.forEach(ArrayList.java:1511)
	at java.base/java.util.ArrayList.forEach(ArrayList.java:1511)

[radcortez]

Failing tests are just flaky now.

[franz1981]

What if...PropertyName become a mutable class (which can wrap different names) and we allocate a single instance of it - if it is often NOT added?

[radcortez]

We only need this at start to validate names. Adding is only done once. My plan was to write a proper API for this in SR Config.

[franz1981]

Depending what's the actual common case you want to optimize for and the number of names, you can just use remove checking for null, to save containsKey + remove to happen.
If this one is not the common scenario, instead, you can just perform a propertyNames.putIfAbsent and after checking that's not null, handling the subsequent remove/change/addition logic

[radcortez]

It is not exactly to optimize, but because different String may be equals (for instance, when they match *), then we need to give priority to the * name when inserting it into the collection.

[franz1981]

I know that sounds paranoid (it is a bit I admit), but beware keySet if the lifecycle of the String used as values is to go away, the key set within HashMap retain a strong reference to the original map (including the entries and values) preventing the Strings to be GCd - see https://github.com/openjdk/jdk/blob/master/src/java.base/share/classes/java/util/HashMap.java#L989 - the hidden this due to be an inner class

[radcortez]

We can rewrite it, but I'm not very concerned. This is used only during build time, and the method calls are all local when validating / generating the configuration, so everything should be GC by the end of the call.

[gsmet]

Let's get this in, we can still improve later.

[gsmet]

I think we will have to backport it to include it in 3.17 as it's fixing a regression.