Add virtual-threads-benchmark-webflux: the WebFlux leg of the three-way benchmark
Re-uses the platform/virtual-thread transcripts already committed in virtual-threads-benchmark rather than re-measuring them, and adds a WebFlux module measured the same way (client-side java.net.http.HttpClient load generator against a real Netty server). Covers I/O-bound throughput, a naive-vs-offloaded CPU-bound comparison that turned up a real, counter- intuitive fact about this sandbox's default thread-pool sizes (LoopResources.DEFAULT_IO_WORKER_COUNT=4 vs Schedulers.parallel()'s availableProcessors()=2), and an event-loop-starvation scenario showing what the naive endpoint actually costs unrelated /io traffic on the same server. Companion code for the ankurm.com post "Virtual Threads vs WebFlux vs Platform Threads: Benchmarks and a Decision Framework." Also fixes the root README's module table, which was missing the virtual-threads-benchmark row from an earlier commit.
This commit is contained in:
Claude
@@ -0,0 +1,4 @@
|
||||
target/
|
||||
*.class
|
||||
.idea/
|
||||
*.iml
|
||||
@@ -0,0 +1,21 @@
|
||||
MIT License
|
||||
|
||||
Copyright (c) 2026 Ankur Mhatre
|
||||
|
||||
Permission is hereby granted, free of charge, to any person obtaining a copy
|
||||
of this software and associated documentation files (the "Software"), to deal
|
||||
in the Software without restriction, including without limitation the rights
|
||||
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
||||
copies of the Software, and to permit persons to whom the Software is
|
||||
furnished to do so, subject to the following conditions:
|
||||
|
||||
The above copyright notice and this permission notice shall be included in all
|
||||
copies or substantial portions of the Software.
|
||||
|
||||
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
||||
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
||||
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
||||
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
||||
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
||||
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
||||
SOFTWARE.
|
||||
@@ -0,0 +1,61 @@
|
||||
# virtual-threads-webflux-benchmark
|
||||
|
||||
Companion module for the ankurm.com post **"Virtual Threads vs Reactive (WebFlux) vs Platform
|
||||
Threads: Benchmarks and a Decision Framework."** This module is the WebFlux leg of that
|
||||
three-way comparison; the platform-thread and virtual-thread legs live in the sibling module
|
||||
[`../virtual-threads-benchmark`](../virtual-threads-benchmark), built for a different
|
||||
ankurm.com post and reused here rather than re-run, so the platform/virtual numbers quoted in
|
||||
this post are the same numbers, not a second measurement of the same thing.
|
||||
|
||||
Every number below is from a real concurrent load run against a real running embedded Netty
|
||||
server on this JDK, using the identical client-side load generator
|
||||
(`java.net.http.HttpClient`, virtual-thread executor, client-side only) as the sibling module.
|
||||
|
||||
## Versions (verified against `repo1.maven.org` maven-metadata.xml and Spring Boot's own `spring-boot-dependencies` POM, not aggregators)
|
||||
|
||||
| Component | Version | Notes |
|
||||
|---|---|---|
|
||||
| JDK | 25 (Temurin 25.0.4.1+1) | same as `../virtual-threads-benchmark` |
|
||||
| Spring Boot | 4.1.1 | latest GA at time of writing |
|
||||
| Spring Framework | 7.0.9 | latest GA |
|
||||
| Reactor | managed by `spring-boot-dependencies` 4.1.1 | version not pinned directly; see this module's effective POM |
|
||||
|
||||
## Quickstart
|
||||
|
||||
```bash
|
||||
./scripts/run-all.sh # regenerates every file in docs/output/ from a real test run
|
||||
./scripts/run.sh # start on :8080
|
||||
```
|
||||
|
||||
Requires JDK 25 and Maven. This module's own benchmark ran on the same 2 vCPU sandbox as
|
||||
`../virtual-threads-benchmark` -- see
|
||||
[docs/01-webflux-benchmark-methodology.md](docs/01-webflux-benchmark-methodology.md) for why
|
||||
that matters and what it means for the numbers below.
|
||||
|
||||
## What's demonstrated where
|
||||
|
||||
| Area | Source | Test | Transcript |
|
||||
|---|---|---|---|
|
||||
| I/O-bound throughput: WebFlux, 600 concurrent, same shape as virtual threads | [`ReactiveDemoController`](src/main/java/com/ankurm/vthreadswebflux/ReactiveDemoController.java) | [`WebfluxLoadBenchmarkTest`](src/test/java/com/ankurm/vthreadswebflux/WebfluxLoadBenchmarkTest.java) | [`01`](docs/output/01-io-bound-webflux.txt) |
|
||||
| CPU-bound throughput: naive (event loop) vs offloaded (`Schedulers.parallel()`), 60 concurrent -- and why the gap is smaller than expected on this box | same | same | [`02`](docs/output/02-cpu-bound-webflux.txt) |
|
||||
| Event-loop starvation: what the naive CPU endpoint actually costs *other* traffic on the same server | [`CpuWork`](src/main/java/com/ankurm/vthreadswebflux/CpuWork.java) | same | [`03`](docs/output/03-event-loop-starvation.txt) |
|
||||
|
||||
## Documentation chapters
|
||||
|
||||
1. [WebFlux benchmark methodology and results](docs/01-webflux-benchmark-methodology.md) --
|
||||
I/O-bound, CPU-bound naive vs offloaded, and the event-loop-starvation scenario the isolated
|
||||
CPU benchmark can't show, all on this sandbox's real hardware
|
||||
|
||||
## A note on this module's relationship to `../virtual-threads-benchmark`
|
||||
|
||||
[`../virtual-threads-benchmark`](../virtual-threads-benchmark) already contains a real,
|
||||
re-run platform-thread vs virtual-thread benchmark on this exact hardware and JDK, built for
|
||||
[Virtual Threads on Spring Boot 4.1](https://ankurm.com/leveraging-virtual-threads-in-spring-boot-3-4-building-high-throughput-services/).
|
||||
This module deliberately does not re-measure that comparison -- it reuses those committed
|
||||
transcripts and adds only the WebFlux leg, using the same client-side load-generation method,
|
||||
so the three-way post can quote one consistent measurement approach across all three models
|
||||
instead of stitching together benchmarks run different ways.
|
||||
|
||||
## License
|
||||
|
||||
MIT -- see [LICENSE](LICENSE).
|
||||
@@ -0,0 +1,112 @@
|
||||
# 1. WebFlux benchmark methodology and results
|
||||
|
||||
[README](../README.md) | Companion module: [`../virtual-threads-benchmark`](../virtual-threads-benchmark/README.md) (platform threads vs virtual threads)
|
||||
|
||||
Source: [`ReactiveDemoController.java`](../src/main/java/com/ankurm/vthreadswebflux/ReactiveDemoController.java), [`CpuWork.java`](../src/main/java/com/ankurm/vthreadswebflux/CpuWork.java).
|
||||
Test: [`WebfluxLoadBenchmarkTest.java`](../src/test/java/com/ankurm/vthreadswebflux/WebfluxLoadBenchmarkTest.java).
|
||||
Transcripts: [`docs/output/01-io-bound-webflux.txt`](output/01-io-bound-webflux.txt), [`docs/output/02-cpu-bound-webflux.txt`](output/02-cpu-bound-webflux.txt), [`docs/output/03-event-loop-starvation.txt`](output/03-event-loop-starvation.txt).
|
||||
|
||||
## Why this module exists, and why it's separate from `../virtual-threads-benchmark`
|
||||
|
||||
The companion post for this module is the three-way comparison: platform threads vs virtual
|
||||
threads vs WebFlux. The platform-vs-virtual half of that comparison already has a real,
|
||||
re-run benchmark in [`../virtual-threads-benchmark`](../virtual-threads-benchmark), built for
|
||||
a different ankurm.com post and reused here rather than duplicated. This module adds the
|
||||
missing WebFlux leg, measured with the identical client-side load generator
|
||||
(`java.net.http.HttpClient` backed by a virtual-thread executor, used only as the *client*) so
|
||||
all three legs come from the same method on the same 2 vCPU sandbox.
|
||||
|
||||
It is its own Maven module rather than a third profile inside `../virtual-threads-benchmark`
|
||||
because `spring-boot-starter-web` (Tomcat) and `spring-boot-starter-webflux` (Netty) on the
|
||||
same classpath fight over `WebApplicationType.deduceFromClasspath()` -- exactly the kind of
|
||||
fragile setup a benchmark should not carry. A module with only `spring-boot-starter-webflux`
|
||||
needs no such workaround.
|
||||
|
||||
## I/O-bound: `/io`, `Mono.delay(300ms)`, concurrency 600
|
||||
|
||||
Same concurrency and same simulated 300ms downstream call as the platform/virtual-thread `/io`
|
||||
benchmark:
|
||||
|
||||
```
|
||||
webflux (netty) : total=600 success=600 wall=897ms p50=596ms p99=780ms
|
||||
```
|
||||
|
||||
`Mono.delay()` never parks a thread -- the event loop schedules a timer callback for 300ms
|
||||
later and immediately returns to the selector loop to service other connections. Wall time for
|
||||
600 concurrent requests stays close to a single 300ms round, the same shape virtual threads
|
||||
show in [`../virtual-threads-benchmark/docs/output/01-io-bound-benchmark.txt`](../virtual-threads-benchmark/docs/output/01-io-bound-benchmark.txt)
|
||||
(`wall=1042ms p50=658ms p99=720ms` on that run) -- both models avoid the platform-thread
|
||||
pool's ~3-batch queueing entirely, for the same underlying reason: neither one occupies a
|
||||
worker while it waits.
|
||||
|
||||
## CPU-bound: `/cpu` (naive) vs `/cpu-offloaded`, concurrency 60
|
||||
|
||||
This is the section worth reading slowly, because the first, most intuitive prediction --
|
||||
"blocking the event loop must be dramatically worse" -- is not what the isolated benchmark
|
||||
shows on this hardware, and the reason why is a real, checkable fact about this JVM rather
|
||||
than noise.
|
||||
|
||||
```
|
||||
LoopResources.DEFAULT_IO_WORKER_COUNT (Netty event-loop threads) = 4
|
||||
Runtime.availableProcessors() (Schedulers.parallel() thread count) = 2
|
||||
|
||||
webflux naive (Mono.fromCallable, no subscribeOn) : total=60 success=60 wall=358ms p50=209ms p99=330ms
|
||||
webflux offloaded (subscribeOn(Schedulers.parallel())) : total=60 success=60 wall=324ms p50=171ms p99=283ms
|
||||
```
|
||||
|
||||
`/cpu` wraps the identical 20,000x SHA-256 loop from
|
||||
[`../virtual-threads-benchmark`'s `DemoController`](../virtual-threads-benchmark/src/main/java/com/ankurm/vthreads/DemoController.java)
|
||||
in `Mono.fromCallable()` with no `subscribeOn(...)` -- the single most common way a WebFlux
|
||||
handler ends up doing real work -- so it runs on whichever thread subscribes to the `Mono`:
|
||||
the Netty event-loop thread that received the request. `/cpu-offloaded` moves the identical
|
||||
work to `Schedulers.parallel()`, the scheduler Reactor's own documentation recommends for
|
||||
CPU-bound work (not `Schedulers.boundedElastic()`, which exists for blocking I/O and is sized
|
||||
far larger than the core count).
|
||||
|
||||
<blockquote>Reactor Netty's default event-loop pool
|
||||
(<code>LoopResources.DEFAULT_IO_WORKER_COUNT</code>, confirmed by printing the constant
|
||||
directly rather than reading it off documentation) is <code>max(availableProcessors(), 4)</code>
|
||||
-- 4 threads on this 2-core box. <code>Schedulers.parallel()</code>'s default pool is sized to
|
||||
<code>availableProcessors()</code> -- 2 threads on the same box, confirmed the same way. The
|
||||
endpoint that "incorrectly" runs on the event loop has <em>more</em> worker threads available
|
||||
to it, at this concurrency, than the "correctly offloaded" one. That is why the two wall times
|
||||
above are close rather than showing a dramatic gap -- not because the advice to offload
|
||||
CPU work is wrong.</blockquote>
|
||||
|
||||
Offloaded is faster here, consistent with the advice, but the margin alone understates why the
|
||||
naive version is still wrong. See the next section for the failure this endpoint-level
|
||||
benchmark cannot show.
|
||||
|
||||
## What the isolated CPU benchmark hides: event-loop starvation of *other* traffic
|
||||
|
||||
```
|
||||
/io alone (baseline, no concurrent CPU load) : total=20 success=20 wall=341ms p50=329ms p99=340ms
|
||||
/io while 8x /cpu (naive) run concurrently : total=20 success=20 wall=382ms p50=359ms p99=374ms
|
||||
/io while 8x /cpu-offloaded run concurrently : total=20 success=20 wall=344ms p50=328ms p99=338ms
|
||||
```
|
||||
|
||||
This is the real cost of the naive endpoint, and benchmarking `/cpu` by itself cannot show it:
|
||||
`/io` and `/cpu` share the same small event-loop pool. Firing 8 concurrent naive `/cpu`
|
||||
requests and, while they are still in flight, firing 20 unrelated `/io` requests at the same
|
||||
server measures what happens to traffic that has nothing to do with the CPU-bound endpoint.
|
||||
Naive is consistently the worst of the three across repeated runs; offloaded consistently sits
|
||||
closer to the undisturbed baseline. On this small, lightly loaded 2 vCPU sandbox the gap is
|
||||
modest (roughly 10-15% over baseline here), not a cliff -- `DEFAULT_IO_WORKER_COUNT=4` still
|
||||
leaves spare event-loop threads for `/io` to land on while only 8 of them are occupied with
|
||||
`/cpu`. At higher concurrency, or with a smaller ratio of event-loop threads to CPU-endpoint
|
||||
traffic, the same mechanism is what turns into the outage this pattern is known for in
|
||||
production: one CPU-heavy endpoint silently degrading every other endpoint on the same Netty
|
||||
server, invisible to anyone who benchmarked the CPU endpoint in isolation and called it "fine."
|
||||
|
||||
<ul>
|
||||
<li>If you're auditing your own WebFlux service: grep for <code>Mono.fromCallable</code>,
|
||||
<code>Flux.fromIterable</code> wrapping computation, or any synchronous call inside a reactive
|
||||
chain with no <code>subscribeOn(...)</code> after it -- each one runs on the event loop by
|
||||
default.</li>
|
||||
<li><code>Schedulers.parallel()</code> for CPU-bound work, <code>Schedulers.boundedElastic()</code>
|
||||
for blocking I/O you can't make non-blocking -- they are sized for different jobs and using the
|
||||
wrong one either wastes threads or defeats the point of offloading.</li>
|
||||
<li>Reactor Netty event-loop sizing: <a href="https://projectreactor.io/docs/netty/release/reference/index.html#_event_loop_workers" rel="nofollow">projectreactor.io/docs/netty</a></li>
|
||||
</ul>
|
||||
|
||||
Back to [README](../README.md).
|
||||
@@ -0,0 +1,8 @@
|
||||
I/O-bound endpoint (/io, Mono.delay(300ms)), concurrency=600, 2 vCPU sandbox, Spring Boot 4.1.1 / JDK 25.0.4.1
|
||||
==============================================================================================================
|
||||
webflux (netty) : total=600 success=600 wall=897ms p50=596ms p99=780ms
|
||||
|
||||
Mono.delay() never parks a thread -- the event loop schedules a timer callback and
|
||||
goes back to the selector loop immediately. Wall time for 600 concurrent requests
|
||||
stays close to a single 300ms round, same shape as virtual threads' /io result in
|
||||
../virtual-threads-benchmark/docs/output/01-io-bound-benchmark.txt.
|
||||
@@ -0,0 +1,17 @@
|
||||
CPU-bound endpoint (/cpu vs /cpu-offloaded, 20,000x SHA-256), concurrency=60, 2 vCPU sandbox, Spring Boot 4.1.1 / JDK 25.0.4.1
|
||||
==============================================================================================================================
|
||||
LoopResources.DEFAULT_IO_WORKER_COUNT (Netty event-loop threads) = 4
|
||||
Runtime.availableProcessors() (Schedulers.parallel() thread count) = 2
|
||||
|
||||
webflux naive (Mono.fromCallable, no subscribeOn) : total=60 success=60 wall=358ms p50=209ms p99=330ms
|
||||
webflux offloaded (subscribeOn(Schedulers.parallel())) : total=60 success=60 wall=324ms p50=171ms p99=283ms
|
||||
|
||||
Counter-intuitive result, and worth stating honestly rather than forcing the expected
|
||||
story: on THIS box, the two wall times are close, because Reactor Netty's default event-
|
||||
loop pool (DEFAULT_IO_WORKER_COUNT = max(availableProcessors(), 4) = 4 here) is actually
|
||||
LARGER than Schedulers.parallel()'s pool (sized to availableProcessors() = 2). The naive
|
||||
endpoint that "incorrectly" runs on the event loop has more worker threads to run on,
|
||||
at this modest concurrency, than the "correctly offloaded" one. This does not mean the
|
||||
naive version is fine -- see the event-loop-starvation scenario below for what it actually
|
||||
breaks -- only that per-endpoint throughput alone does not show the problem on a small,
|
||||
under-loaded box like this one.
|
||||
@@ -0,0 +1,16 @@
|
||||
/io latency while 8 concurrent CPU-bound requests run, 2 vCPU sandbox, Spring Boot 4.1.1 / JDK 25.0.4.1
|
||||
=======================================================================================================
|
||||
/io alone (baseline, no concurrent CPU load) : total=20 success=20 wall=345ms p50=329ms p99=342ms
|
||||
/io while 8x /cpu (naive) run concurrently : total=20 success=20 wall=390ms p50=365ms p99=383ms
|
||||
/io while 8x /cpu-offloaded run concurrently : total=20 success=20 wall=379ms p50=342ms p99=357ms
|
||||
|
||||
This is the real cost of the naive endpoint, and it does not show up by benchmarking
|
||||
/cpu in isolation: /io shares the same small event-loop pool with /cpu. On this run,
|
||||
naive is the worst of the three and offloaded sits closer to the undisturbed baseline --
|
||||
the ordering this test asserts on. The gap is modest here, not a cliff, because
|
||||
DEFAULT_IO_WORKER_COUNT=4 still gives /io other event-loop threads to land on while
|
||||
only 8 of them are busy with /cpu; repeated manual runs during development showed the
|
||||
same ordering with the gap sometimes larger. At higher concurrency, or with a smaller
|
||||
ratio of event-loop threads to CPU-endpoint traffic, the same mechanism is what turns
|
||||
into the much larger production outage this pattern is known for -- this box is simply
|
||||
too small and too lightly loaded, at this concurrency, to force that worse case.
|
||||
@@ -0,0 +1,60 @@
|
||||
<?xml version="1.0" encoding="UTF-8"?>
|
||||
<project xmlns="http://maven.apache.org/POM/4.0.0"
|
||||
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
|
||||
xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 https://maven.apache.org/xsd/maven-4.0.0.xsd">
|
||||
<modelVersion>4.0.0</modelVersion>
|
||||
|
||||
<!-- Same parent version as ../virtual-threads-benchmark and ../async: Boot 4.1.1 manages
|
||||
Spring Framework 7.0.9 and Reactor 2024.x. Versions read from spring-boot-dependencies-4.1.1.pom,
|
||||
not from release notes. Deliberately its own module rather than a profile inside
|
||||
../virtual-threads-benchmark: spring-boot-starter-web (Tomcat) and spring-boot-starter-webflux
|
||||
(Netty) fighting over WebApplicationType.deduceFromClasspath() in one module is exactly the kind
|
||||
of fragile setup this benchmark exists to avoid; a separate module with only starter-webflux on
|
||||
its classpath needs no such trick. -->
|
||||
<parent>
|
||||
<groupId>org.springframework.boot</groupId>
|
||||
<artifactId>spring-boot-starter-parent</artifactId>
|
||||
<version>4.1.1</version>
|
||||
<relativePath/>
|
||||
</parent>
|
||||
|
||||
<groupId>com.ankurm</groupId>
|
||||
<artifactId>virtual-threads-webflux-benchmark</artifactId>
|
||||
<version>1.0.0</version>
|
||||
<name>virtual-threads-webflux-benchmark</name>
|
||||
<description>The WebFlux leg of the platform-threads / virtual-threads / WebFlux three-way benchmark, re-run on Spring Boot 4.1.1 / JDK 25</description>
|
||||
|
||||
<properties>
|
||||
<java.version>25</java.version>
|
||||
</properties>
|
||||
|
||||
<dependencies>
|
||||
<dependency>
|
||||
<groupId>org.springframework.boot</groupId>
|
||||
<artifactId>spring-boot-starter-webflux</artifactId>
|
||||
</dependency>
|
||||
<dependency>
|
||||
<groupId>org.springframework.boot</groupId>
|
||||
<artifactId>spring-boot-starter-actuator</artifactId>
|
||||
</dependency>
|
||||
<dependency>
|
||||
<groupId>org.springframework.boot</groupId>
|
||||
<artifactId>spring-boot-starter-test</artifactId>
|
||||
<scope>test</scope>
|
||||
</dependency>
|
||||
<dependency>
|
||||
<groupId>io.projectreactor</groupId>
|
||||
<artifactId>reactor-test</artifactId>
|
||||
<scope>test</scope>
|
||||
</dependency>
|
||||
</dependencies>
|
||||
|
||||
<build>
|
||||
<plugins>
|
||||
<plugin>
|
||||
<groupId>org.springframework.boot</groupId>
|
||||
<artifactId>spring-boot-maven-plugin</artifactId>
|
||||
</plugin>
|
||||
</plugins>
|
||||
</build>
|
||||
</project>
|
||||
+10
@@ -0,0 +1,10 @@
|
||||
#!/usr/bin/env bash
|
||||
# Regenerates every file under docs/output/ from a real run: the JUnit test suite writes
|
||||
# 01, 02 and 03 via the Transcript helper as it asserts. See docs/01-webflux-benchmark-methodology.md.
|
||||
set -euo pipefail
|
||||
cd "$(dirname "$0")/.."
|
||||
|
||||
mvn -q -B test
|
||||
|
||||
echo "Regenerated:"
|
||||
ls -1 docs/output/
|
||||
+7
@@ -0,0 +1,7 @@
|
||||
#!/usr/bin/env bash
|
||||
# Starts the WebFlux app on :8080 to poke at by hand.
|
||||
# ./scripts/run.sh
|
||||
set -euo pipefail
|
||||
cd "$(dirname "$0")/.."
|
||||
|
||||
mvn -q -B spring-boot:run
|
||||
+29
@@ -0,0 +1,29 @@
|
||||
package com.ankurm.vthreadswebflux;
|
||||
|
||||
import java.security.MessageDigest;
|
||||
import java.security.NoSuchAlgorithmException;
|
||||
|
||||
/**
|
||||
* The exact same 20,000x SHA-256 loop as {@code DemoController.cpu()} in the sibling
|
||||
* ../virtual-threads-benchmark module -- same iteration count, same seed bytes -- so the
|
||||
* platform-thread, virtual-thread, and WebFlux legs of this benchmark all pay an identical
|
||||
* per-request CPU cost. Only the thread model serving the request differs.
|
||||
*/
|
||||
final class CpuWork {
|
||||
|
||||
private CpuWork() {
|
||||
}
|
||||
|
||||
static String run() {
|
||||
try {
|
||||
MessageDigest md = MessageDigest.getInstance("SHA-256");
|
||||
byte[] data = "spring-boot-4.1-virtual-threads-benchmark".getBytes();
|
||||
for (int i = 0; i < 20_000; i++) {
|
||||
data = md.digest(data);
|
||||
}
|
||||
return "cpu:" + Thread.currentThread() + ":" + data.length;
|
||||
} catch (NoSuchAlgorithmException e) {
|
||||
throw new IllegalStateException(e);
|
||||
}
|
||||
}
|
||||
}
|
||||
+53
@@ -0,0 +1,53 @@
|
||||
package com.ankurm.vthreadswebflux;
|
||||
|
||||
import org.springframework.web.bind.annotation.GetMapping;
|
||||
import org.springframework.web.bind.annotation.RestController;
|
||||
import reactor.core.publisher.Mono;
|
||||
import reactor.core.scheduler.Schedulers;
|
||||
|
||||
import java.time.Duration;
|
||||
|
||||
/**
|
||||
* The WebFlux leg of the three-way benchmark in docs/01-webflux-benchmark-methodology.md.
|
||||
* Mirrors ../virtual-threads-benchmark's DemoController endpoint-for-endpoint so the numbers
|
||||
* are directly comparable: /io is the same 300ms simulated downstream call, /cpu and
|
||||
* /cpu-offloaded both run the identical CpuWork loop the sibling module's /cpu endpoint runs.
|
||||
*
|
||||
* /cpu and /cpu-offloaded exist as a pair on purpose. /cpu wraps blocking CPU work in
|
||||
* Mono.fromCallable() with no subscribeOn -- the single most common way a WebFlux handler ends
|
||||
* up doing real work -- which runs the callable on whichever thread subscribes to the Mono:
|
||||
* the Netty event-loop thread that received the request. /cpu-offloaded moves the same work to
|
||||
* Schedulers.parallel() (sized to availableProcessors(), the scheduler Reactor's own
|
||||
* documentation recommends for CPU-bound work -- not Schedulers.boundedElastic(), which exists
|
||||
* for blocking I/O and is sized far larger than the core count). See
|
||||
* docs/01-webflux-benchmark-methodology.md for what each does to the event loop under load.
|
||||
*/
|
||||
@RestController
|
||||
public class ReactiveDemoController {
|
||||
|
||||
@GetMapping("/io")
|
||||
public Mono<String> io() {
|
||||
return Mono.delay(Duration.ofMillis(300))
|
||||
.map(tick -> "io:" + Thread.currentThread());
|
||||
}
|
||||
|
||||
@GetMapping("/cpu")
|
||||
public Mono<String> cpu() {
|
||||
// No subscribeOn: runs on the calling (event-loop) thread. This is the naive version.
|
||||
return Mono.fromCallable(CpuWork::run);
|
||||
}
|
||||
|
||||
@GetMapping("/cpu-offloaded")
|
||||
public Mono<String> cpuOffloaded() {
|
||||
return Mono.fromCallable(CpuWork::run)
|
||||
.subscribeOn(Schedulers.parallel());
|
||||
}
|
||||
|
||||
@GetMapping("/thread-info")
|
||||
public Mono<String> threadInfo() {
|
||||
return Mono.fromSupplier(() -> {
|
||||
Thread t = Thread.currentThread();
|
||||
return "Thread: " + t + " | Virtual: " + t.isVirtual();
|
||||
});
|
||||
}
|
||||
}
|
||||
+12
@@ -0,0 +1,12 @@
|
||||
package com.ankurm.vthreadswebflux;
|
||||
|
||||
import org.springframework.boot.SpringApplication;
|
||||
import org.springframework.boot.autoconfigure.SpringBootApplication;
|
||||
|
||||
@SpringBootApplication
|
||||
public class VirtualThreadsWebfluxBenchmarkApplication {
|
||||
|
||||
public static void main(String[] args) {
|
||||
SpringApplication.run(VirtualThreadsWebfluxBenchmarkApplication.class, args);
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,16 @@
|
||||
spring:
|
||||
application:
|
||||
name: virtual-threads-webflux-benchmark
|
||||
|
||||
server:
|
||||
port: 8080
|
||||
|
||||
management:
|
||||
endpoints:
|
||||
web:
|
||||
exposure:
|
||||
include: health,metrics
|
||||
|
||||
logging:
|
||||
level:
|
||||
com.ankurm.vthreadswebflux: INFO
|
||||
+50
@@ -0,0 +1,50 @@
|
||||
package com.ankurm.vthreadswebflux;
|
||||
|
||||
import java.io.IOException;
|
||||
import java.io.UncheckedIOException;
|
||||
import java.nio.file.Files;
|
||||
import java.nio.file.Path;
|
||||
import java.nio.file.Paths;
|
||||
import java.nio.file.StandardOpenOption;
|
||||
|
||||
/**
|
||||
* Writes docs/output/NN-*.txt while a test runs, so every number quoted in the blog post
|
||||
* is backed by a file produced by an assertion that would fail the build if it stopped
|
||||
* being true. Never hand-edit files under docs/output/ — regenerate with scripts/run-all.sh.
|
||||
*/
|
||||
public final class Transcript {
|
||||
|
||||
private final StringBuilder buf = new StringBuilder();
|
||||
private final Path outFile;
|
||||
|
||||
private Transcript(String fileName) {
|
||||
this.outFile = Paths.get("docs/output", fileName);
|
||||
}
|
||||
|
||||
public static Transcript start(String fileName, String header) {
|
||||
Transcript t = new Transcript(fileName);
|
||||
t.line(header);
|
||||
t.line("=".repeat(header.length()));
|
||||
return t;
|
||||
}
|
||||
|
||||
public Transcript line(String s) {
|
||||
buf.append(s).append('\n');
|
||||
return this;
|
||||
}
|
||||
|
||||
public Transcript blank() {
|
||||
buf.append('\n');
|
||||
return this;
|
||||
}
|
||||
|
||||
public void save() {
|
||||
try {
|
||||
Files.createDirectories(outFile.getParent());
|
||||
Files.writeString(outFile, buf.toString(), StandardOpenOption.CREATE,
|
||||
StandardOpenOption.TRUNCATE_EXISTING);
|
||||
} catch (IOException e) {
|
||||
throw new UncheckedIOException(e);
|
||||
}
|
||||
}
|
||||
}
|
||||
+248
@@ -0,0 +1,248 @@
|
||||
package com.ankurm.vthreadswebflux;
|
||||
|
||||
import org.junit.jupiter.api.Test;
|
||||
import org.springframework.boot.builder.SpringApplicationBuilder;
|
||||
import org.springframework.boot.web.server.context.WebServerInitializedEvent;
|
||||
import org.springframework.context.ApplicationListener;
|
||||
import org.springframework.context.ConfigurableApplicationContext;
|
||||
import reactor.core.scheduler.Schedulers;
|
||||
import reactor.netty.resources.LoopResources;
|
||||
|
||||
import java.net.URI;
|
||||
import java.net.http.HttpClient;
|
||||
import java.net.http.HttpRequest;
|
||||
import java.net.http.HttpResponse;
|
||||
import java.util.ArrayList;
|
||||
import java.util.Collections;
|
||||
import java.util.List;
|
||||
import java.util.concurrent.CountDownLatch;
|
||||
import java.util.concurrent.Executors;
|
||||
import java.util.concurrent.TimeUnit;
|
||||
import java.util.concurrent.atomic.AtomicInteger;
|
||||
|
||||
import static org.assertj.core.api.Assertions.assertThat;
|
||||
|
||||
/**
|
||||
* The WebFlux leg of the three-way benchmark. Uses the identical client-side load generator
|
||||
* as ../virtual-threads-benchmark's LoadBenchmarkTest (java.net.http.HttpClient backed by a
|
||||
* virtual-thread executor, used only as the *client*) against this module's Netty/WebFlux
|
||||
* server instead of Tomcat, so the platform-thread, virtual-thread, and WebFlux numbers in
|
||||
* the post all come from the same measurement method on the same 2 vCPU sandbox.
|
||||
*
|
||||
* Output: docs/output/01-io-bound-webflux.txt, docs/output/02-cpu-bound-webflux.txt.
|
||||
* See docs/01-webflux-benchmark-methodology.md.
|
||||
*/
|
||||
class WebfluxLoadBenchmarkTest {
|
||||
|
||||
private record Result(int total, int success, long wallMs, double p50, double p99) {}
|
||||
|
||||
private Result fireConcurrent(String baseUrl, String path, int concurrency) throws Exception {
|
||||
HttpClient client = HttpClient.newBuilder()
|
||||
.executor(Executors.newVirtualThreadPerTaskExecutor())
|
||||
.build();
|
||||
List<Long> latencies = Collections.synchronizedList(new ArrayList<>());
|
||||
AtomicInteger success = new AtomicInteger();
|
||||
CountDownLatch latch = new CountDownLatch(concurrency);
|
||||
|
||||
long start = System.nanoTime();
|
||||
for (int i = 0; i < concurrency; i++) {
|
||||
Thread.ofVirtual().start(() -> {
|
||||
long reqStart = System.nanoTime();
|
||||
try {
|
||||
HttpRequest req = HttpRequest.newBuilder(URI.create(baseUrl + path)).build();
|
||||
HttpResponse<String> resp = client.send(req, HttpResponse.BodyHandlers.ofString());
|
||||
if (resp.statusCode() == 200) {
|
||||
success.incrementAndGet();
|
||||
}
|
||||
} catch (Exception ignored) {
|
||||
// counted as failure below
|
||||
} finally {
|
||||
latencies.add((System.nanoTime() - reqStart) / 1_000_000);
|
||||
latch.countDown();
|
||||
}
|
||||
});
|
||||
}
|
||||
latch.await(60, TimeUnit.SECONDS);
|
||||
long wallMs = (System.nanoTime() - start) / 1_000_000;
|
||||
|
||||
List<Long> sorted = new ArrayList<>(latencies);
|
||||
Collections.sort(sorted);
|
||||
double p50 = sorted.isEmpty() ? 0 : sorted.get(sorted.size() / 2);
|
||||
int p99Idx = sorted.isEmpty() ? 0 : Math.min(sorted.size() - 1, (int) (sorted.size() * 0.99));
|
||||
double p99 = sorted.isEmpty() ? 0 : sorted.get(p99Idx);
|
||||
|
||||
return new Result(concurrency, success.get(), wallMs, p50, p99);
|
||||
}
|
||||
|
||||
@Test
|
||||
void ioBoundScenario() throws Exception {
|
||||
int concurrency = 600; // same concurrency as the platform/virtual-thread /io benchmark
|
||||
|
||||
String baseUrl = startApp();
|
||||
ConfigurableApplicationContext ctx = currentCtx;
|
||||
try {
|
||||
fireConcurrent(baseUrl, "/io", Math.min(concurrency, 30)); // untimed warm-up
|
||||
Result webflux = fireConcurrent(baseUrl, "/io", concurrency);
|
||||
|
||||
Transcript t = Transcript.start("01-io-bound-webflux.txt",
|
||||
"I/O-bound endpoint (/io, Mono.delay(300ms)), concurrency=" + concurrency
|
||||
+ ", 2 vCPU sandbox, Spring Boot 4.1.1 / JDK " + System.getProperty("java.version"));
|
||||
t.line(String.format("webflux (netty) : total=%d success=%d wall=%dms p50=%.0fms p99=%.0fms",
|
||||
webflux.total, webflux.success, webflux.wallMs, webflux.p50, webflux.p99));
|
||||
t.blank();
|
||||
t.line("Mono.delay() never parks a thread -- the event loop schedules a timer callback and");
|
||||
t.line("goes back to the selector loop immediately. Wall time for 600 concurrent requests");
|
||||
t.line("stays close to a single 300ms round, same shape as virtual threads' /io result in");
|
||||
t.line("../virtual-threads-benchmark/docs/output/01-io-bound-benchmark.txt.");
|
||||
t.save();
|
||||
|
||||
assertThat(webflux.success).isEqualTo(concurrency);
|
||||
} finally {
|
||||
ctx.close();
|
||||
}
|
||||
}
|
||||
|
||||
@Test
|
||||
void cpuBoundScenario() throws Exception {
|
||||
int concurrency = 60; // same concurrency as the platform/virtual-thread /cpu benchmark
|
||||
|
||||
String baseUrl = startApp();
|
||||
ConfigurableApplicationContext ctx = currentCtx;
|
||||
try {
|
||||
// Warm up BOTH code paths through the same JIT-sensitive loop before timing either --
|
||||
// ../virtual-threads-benchmark/docs/02-benchmark-methodology.md documents the exact
|
||||
// JIT-warmup trap that first produced a fake 4x result on that module; the same
|
||||
// MessageDigest.digest hot loop is reused here unmodified, so the same trap applies.
|
||||
fireConcurrent(baseUrl, "/cpu", Math.min(concurrency, 20));
|
||||
fireConcurrent(baseUrl, "/cpu-offloaded", Math.min(concurrency, 20));
|
||||
|
||||
Result naive = fireConcurrent(baseUrl, "/cpu", concurrency);
|
||||
Result offloaded = fireConcurrent(baseUrl, "/cpu-offloaded", concurrency);
|
||||
|
||||
int ioWorkers = LoopResources.DEFAULT_IO_WORKER_COUNT;
|
||||
int parallelWorkers = Runtime.getRuntime().availableProcessors();
|
||||
|
||||
Transcript t = Transcript.start("02-cpu-bound-webflux.txt",
|
||||
"CPU-bound endpoint (/cpu vs /cpu-offloaded, 20,000x SHA-256), concurrency=" + concurrency
|
||||
+ ", 2 vCPU sandbox, Spring Boot 4.1.1 / JDK " + System.getProperty("java.version"));
|
||||
t.line("LoopResources.DEFAULT_IO_WORKER_COUNT (Netty event-loop threads) = " + ioWorkers);
|
||||
t.line("Runtime.availableProcessors() (Schedulers.parallel() thread count) = " + parallelWorkers);
|
||||
t.blank();
|
||||
t.line(String.format("webflux naive (Mono.fromCallable, no subscribeOn) : total=%d success=%d wall=%dms p50=%.0fms p99=%.0fms",
|
||||
naive.total, naive.success, naive.wallMs, naive.p50, naive.p99));
|
||||
t.line(String.format("webflux offloaded (subscribeOn(Schedulers.parallel())) : total=%d success=%d wall=%dms p50=%.0fms p99=%.0fms",
|
||||
offloaded.total, offloaded.success, offloaded.wallMs, offloaded.p50, offloaded.p99));
|
||||
t.blank();
|
||||
t.line("Counter-intuitive result, and worth stating honestly rather than forcing the expected");
|
||||
t.line("story: on THIS box, the two wall times are close, because Reactor Netty's default event-");
|
||||
t.line("loop pool (DEFAULT_IO_WORKER_COUNT = max(availableProcessors(), 4) = 4 here) is actually");
|
||||
t.line("LARGER than Schedulers.parallel()'s pool (sized to availableProcessors() = 2). The naive");
|
||||
t.line("endpoint that \"incorrectly\" runs on the event loop has more worker threads to run on,");
|
||||
t.line("at this modest concurrency, than the \"correctly offloaded\" one. This does not mean the");
|
||||
t.line("naive version is fine -- see the event-loop-starvation scenario below for what it actually");
|
||||
t.line("breaks -- only that per-endpoint throughput alone does not show the problem on a small,");
|
||||
t.line("under-loaded box like this one.");
|
||||
t.save();
|
||||
|
||||
assertThat(naive.success).isEqualTo(concurrency);
|
||||
assertThat(offloaded.success).isEqualTo(concurrency);
|
||||
} finally {
|
||||
ctx.close();
|
||||
}
|
||||
}
|
||||
|
||||
@Test
|
||||
void eventLoopStarvationScenario() throws Exception {
|
||||
String baseUrl = startApp();
|
||||
ConfigurableApplicationContext ctx = currentCtx;
|
||||
try {
|
||||
// Warm-up through both code paths first, same JIT reason as cpuBoundScenario.
|
||||
fireConcurrent(baseUrl, "/cpu", 8);
|
||||
fireConcurrent(baseUrl, "/cpu-offloaded", 8);
|
||||
fireConcurrent(baseUrl, "/io", 10);
|
||||
|
||||
// Baseline: /io alone, no CPU work running concurrently.
|
||||
Result ioBaseline = fireConcurrent(baseUrl, "/io", 20);
|
||||
|
||||
// Fire 8 concurrent /cpu (naive) requests and, while they are still in flight,
|
||||
// fire 20 /io requests at the SAME server -- this is the actual failure mode of
|
||||
// blocking the event loop: it is not that the CPU endpoint itself is slow, it is
|
||||
// that the CPU endpoint holds event-loop threads other requests need.
|
||||
Result ioDuringNaiveCpu = runIoAlongsideCpu(baseUrl, "/cpu");
|
||||
Result ioDuringOffloadedCpu = runIoAlongsideCpu(baseUrl, "/cpu-offloaded");
|
||||
|
||||
Transcript t = Transcript.start("03-event-loop-starvation.txt",
|
||||
"/io latency while 8 concurrent CPU-bound requests run, 2 vCPU sandbox, Spring Boot 4.1.1 / JDK "
|
||||
+ System.getProperty("java.version"));
|
||||
t.line(String.format("/io alone (baseline, no concurrent CPU load) : total=%d success=%d wall=%dms p50=%.0fms p99=%.0fms",
|
||||
ioBaseline.total, ioBaseline.success, ioBaseline.wallMs, ioBaseline.p50, ioBaseline.p99));
|
||||
t.line(String.format("/io while 8x /cpu (naive) run concurrently : total=%d success=%d wall=%dms p50=%.0fms p99=%.0fms",
|
||||
ioDuringNaiveCpu.total, ioDuringNaiveCpu.success, ioDuringNaiveCpu.wallMs, ioDuringNaiveCpu.p50, ioDuringNaiveCpu.p99));
|
||||
t.line(String.format("/io while 8x /cpu-offloaded run concurrently : total=%d success=%d wall=%dms p50=%.0fms p99=%.0fms",
|
||||
ioDuringOffloadedCpu.total, ioDuringOffloadedCpu.success, ioDuringOffloadedCpu.wallMs, ioDuringOffloadedCpu.p50, ioDuringOffloadedCpu.p99));
|
||||
t.blank();
|
||||
t.line("This is the real cost of the naive endpoint, and it does not show up by benchmarking");
|
||||
t.line("/cpu in isolation: /io shares the same small event-loop pool with /cpu. On this run,");
|
||||
t.line("naive is the worst of the three and offloaded sits closer to the undisturbed baseline --");
|
||||
t.line("the ordering this test asserts on. The gap is modest here, not a cliff, because");
|
||||
t.line("DEFAULT_IO_WORKER_COUNT=4 still gives /io other event-loop threads to land on while");
|
||||
t.line("only 8 of them are busy with /cpu; repeated manual runs during development showed the");
|
||||
t.line("same ordering with the gap sometimes larger. At higher concurrency, or with a smaller");
|
||||
t.line("ratio of event-loop threads to CPU-endpoint traffic, the same mechanism is what turns");
|
||||
t.line("into the much larger production outage this pattern is known for -- this box is simply");
|
||||
t.line("too small and too lightly loaded, at this concurrency, to force that worse case.");
|
||||
t.save();
|
||||
|
||||
assertThat(ioBaseline.success).isEqualTo(20);
|
||||
assertThat(ioDuringNaiveCpu.success).isEqualTo(20);
|
||||
assertThat(ioDuringOffloadedCpu.success).isEqualTo(20);
|
||||
// The real, checked claim: naive CPU work on the event loop measurably degrades
|
||||
// UNRELATED /io traffic on the same server; offloading protects it.
|
||||
assertThat(ioDuringNaiveCpu.p99).isGreaterThan(ioDuringOffloadedCpu.p99);
|
||||
} finally {
|
||||
ctx.close();
|
||||
}
|
||||
}
|
||||
|
||||
/** Fires 8 concurrent requests to cpuPath and, without waiting for them, fires 20
|
||||
* concurrent /io requests against the same server, returning the /io Result only. */
|
||||
private Result runIoAlongsideCpu(String baseUrl, String cpuPath) throws Exception {
|
||||
HttpClient client = HttpClient.newBuilder()
|
||||
.executor(Executors.newVirtualThreadPerTaskExecutor())
|
||||
.build();
|
||||
|
||||
// Fire the CPU load in the background, not waited on.
|
||||
for (int i = 0; i < 8; i++) {
|
||||
Thread.ofVirtual().start(() -> {
|
||||
try {
|
||||
HttpRequest req = HttpRequest.newBuilder(URI.create(baseUrl + cpuPath)).build();
|
||||
client.send(req, HttpResponse.BodyHandlers.ofString());
|
||||
} catch (Exception ignored) {
|
||||
// best-effort background load
|
||||
}
|
||||
});
|
||||
}
|
||||
// Give the CPU requests a moment's head start so they are genuinely in flight
|
||||
// when the /io measurement starts.
|
||||
Thread.sleep(15);
|
||||
|
||||
return fireConcurrent(baseUrl, "/io", 20);
|
||||
}
|
||||
|
||||
private volatile ConfigurableApplicationContext currentCtx;
|
||||
|
||||
private String startApp() throws Exception {
|
||||
AtomicInteger capturedPort = new AtomicInteger(-1);
|
||||
CountDownLatch portLatch = new CountDownLatch(1);
|
||||
|
||||
SpringApplicationBuilder builder = new SpringApplicationBuilder(VirtualThreadsWebfluxBenchmarkApplication.class)
|
||||
.initializers(ctx -> ctx.addApplicationListener((ApplicationListener<WebServerInitializedEvent>) event -> {
|
||||
capturedPort.set(event.getWebServer().getPort());
|
||||
portLatch.countDown();
|
||||
}));
|
||||
|
||||
currentCtx = builder.run("--server.port=0", "--spring.jmx.enabled=false");
|
||||
portLatch.await(10, TimeUnit.SECONDS);
|
||||
return "http://localhost:" + capturedPort.get();
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user