Save additional copies on byte[] and String REST resources

[franz1981]

Similarly to #46936 this is saving multiple copies before sending an HTTP response to the wire

e.g.

before this PR (for String):

• Buffer::buffer(String) encode the String::getBytes into a new byte[] (see https://github.com/eclipse-vertx/vert.x/blob/4.5.14/src/main/java/io/vertx/core/buffer/impl/BufferImpl.java#L75-L81)
• BufferImpl(byte[]) allocates a new byte[]-backed buffer, thrashing the just created byte[] (see https://github.com/eclipse-vertx/vert.x/blob/4.5.14/src/main/java/io/vertx/core/buffer/impl/BufferImpl.java#L75-L81)
• the vertx Buffer is saved into a HTTP response's data via https://github.com/eclipse-vertx/vert.x/blob/4.5.14/src/main/java/io/vertx/core/http/impl/Http1xServerResponse.java#L433 without copying the underlying Netty ByteBuf (which is sadly backed by a byte[])

If Http 1.1:

• encoding a full http response in Netty (non-chunked) would go into https://github.com/netty/netty/blob/netty-4.1.119.Final/codec-http/src/main/java/io/netty/handler/codec/http/HttpObjectEncoder.java#L347-L361: it can either just copy the byte[]-based Netty Buffer into the allocated one for the full encoded response by Netty (which is uncommon - unless for benchmarks which can predict correctly the encoded expected size) or passing it straight (while retaining it) to the Netty pipeline as it is
• In the latter case, since native layers cannot use byte Buffers, Netty will copy it again into an off-heap one (see https://github.com/netty/netty/blob/620794f458549b0d228715ebf4b359fdabe46140/transport/src/main/java/io/netty/channel/nio/AbstractNioByteChannel.java#L272-L280)

after this PR (for String):

• only one copy happens, but requires 3 times the available String size, since UTF-8 max allocated space is required; usually this is not a problem since Netty pooled buffers have size-based caching, which means that is likely not allocating anything new

I can modify as well the String allocation to perform exact length check - or by using reflection, I can ask to the String itself (using the coder field into it) if is a latin one, in O(1) - but I don't want to push this too far.

For the byte[] case instead, there's a similar story (since Buffer.buffer(byte[]) perform another copy in a vertx heap based buffer, but there's no drawback when it comes with preallocated capacity.

[franz1981]

@tsegismont This is for you as well bud; both because it involves some inner Netty behaviour and because it likely show some existing gap in Vertx Buffer's impl itself

[quarkus-bot]

Status for workflow Quarkus CI

This is the status report for running Quarkus CI on commit 7590321.

✅ 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 Integration Tests - JDK 17

📦 integration-tests/injectmock

✖ io.quarkus.it.mockbean.PerClassSpyTest.testWithoutSpy - History

• Mock accessed after inline mocks were cleared - org.mockito.exceptions.misusing.DisabledMockException

org.mockito.exceptions.misusing.DisabledMockException: Mock accessed after inline mocks were cleared
	at io.quarkus.it.mockbean.PerClassSpyTest$IdentityService.call(PerClassSpyTest.java:43)
	at io.quarkus.it.mockbean.PerClassSpyTest.testWithoutSpy(PerClassSpyTest.java:36)
	at java.base/java.lang.reflect.Method.invoke(Method.java:569)
	at io.quarkus.test.junit.QuarkusTestExtension.runExtensionMethod(QuarkusTestExtension.java:1026)
	at io.quarkus.test.junit.QuarkusTestExtension.interceptTestMethod(QuarkusTestExtension.java:873)
	at java.base/java.util.ArrayList.forEach(ArrayList.java:1511)
	at java.base/java.util.ArrayList.forEach(ArrayList.java:1511)

---

⚙️ JVM Integration Tests - JDK 17 Windows

📦 integration-tests/injectmock

✖ io.quarkus.it.mockbean.PerClassSpyTest.testWithoutSpy - History

• Mock accessed after inline mocks were cleared - org.mockito.exceptions.misusing.DisabledMockException

org.mockito.exceptions.misusing.DisabledMockException: Mock accessed after inline mocks were cleared
	at io.quarkus.it.mockbean.PerClassSpyTest$IdentityService.call(PerClassSpyTest.java:43)
	at io.quarkus.it.mockbean.PerClassSpyTest.testWithoutSpy(PerClassSpyTest.java:36)
	at java.base/java.lang.reflect.Method.invoke(Method.java:569)
	at io.quarkus.test.junit.QuarkusTestExtension.runExtensionMethod(QuarkusTestExtension.java:1026)
	at io.quarkus.test.junit.QuarkusTestExtension.interceptTestMethod(QuarkusTestExtension.java:873)
	at java.base/java.util.ArrayList.forEach(ArrayList.java:1511)
	at java.base/java.util.ArrayList.forEach(ArrayList.java:1511)

---

⚙️ JVM Integration Tests - JDK 21

📦 integration-tests/injectmock

✖ io.quarkus.it.mockbean.PerClassSpyTest.testWithoutSpy - History

• Mock accessed after inline mocks were cleared - org.mockito.exceptions.misusing.DisabledMockException

org.mockito.exceptions.misusing.DisabledMockException: Mock accessed after inline mocks were cleared
	at io.quarkus.it.mockbean.PerClassSpyTest$IdentityService.call(PerClassSpyTest.java:43)
	at io.quarkus.it.mockbean.PerClassSpyTest.testWithoutSpy(PerClassSpyTest.java:36)
	at java.base/java.lang.reflect.Method.invoke(Method.java:580)
	at io.quarkus.test.junit.QuarkusTestExtension.runExtensionMethod(QuarkusTestExtension.java:1026)
	at io.quarkus.test.junit.QuarkusTestExtension.interceptTestMethod(QuarkusTestExtension.java:873)
	at java.base/java.util.ArrayList.forEach(ArrayList.java:1596)
	at java.base/java.util.ArrayList.forEach(ArrayList.java:1596)

📦 integration-tests/virtual-threads/grpc-virtual-threads

✖ io.quarkus.grpc.example.streaming.VertxVirtualThreadTest.testGrpcClient - History

• 1 expectation failed. Expected status code <200> but was <500>. - java.lang.AssertionError

java.lang.AssertionError: 
1 expectation failed.
Expected status code <200> but was <500>.

	at java.base/java.lang.reflect.Constructor.newInstanceWithCaller(Constructor.java:502)
	at java.base/java.lang.reflect.Constructor.newInstance(Constructor.java:486)
	at org.codehaus.groovy.reflection.CachedConstructor.invoke(CachedConstructor.java:73)
	at org.codehaus.groovy.runtime.callsite.ConstructorSite$ConstructorSiteNoUnwrapNoCoerce.callConstructor(ConstructorSite.java:108)

[tsegismont]

@franz1981 this looks interesting, have you got a significant improvement, in plain text benchmark?

[franz1981]

this looks interesting, have you got a significant improvement, in plain text benchmark?

I checked that there was no regression; but in our perf tests we tends to use the blocking worker pool and not running on the I/O thread - so it is less evident what improvement you got since the main bottleneck there is the thread handoff.

We should add some more test which uses byte[] in the I/O thread instead: IDK if @geoand got some idea for the byte[] case since I have noticed that is fairly tested in our Quarkus tests so maybe is used way more than I was expecting (for common user scenarios) - and that means can be a nice additional test to have too.