Add virtual-threads-benchmark-webflux: the WebFlux leg of the three-way benchmark
Fixes found during self-correction before publishing: - /stream used Flux.interval(), which ticks on its own wall-clock schedule independent of downstream demand and threw OverflowException under a slow subscriber; switched to Flux.range(), which has no independent production schedule and can never outrun demand. - Single-trial HTTP load tests on this shared sandbox swung by more than 50% run to run (795ms-1247ms observed on the identical /io endpoint back to back) -- large enough to flip which threading model looked faster. Fixed by taking the median of 5 independent trials for the I/O-bound benchmark and the median of 3 for the event-loop-starvation benchmark, rather than reporting a single noisy run as if it were precise. - The event-loop-starvation test's first cut used only 8 concurrent /cpu requests as background load, which drained through the 4 event-loop threads well inside the /io measurement window and produced an inconsistent, sometimes-inverted result across runs; raising to 60 fixed the under-loading problem but still flaked once during verification (372ms vs 374ms p99, a real tie). Final fix: 150 concurrent requests plus the median-of-3 trials above. Also adds StreamBackpressureTest, a StepVerifier proof that the /stream endpoint never emits ahead of its subscriber's outstanding requests, and updates the module's docs to report the de-noised numbers with an explicit methodology note on how they compare to the single-trial platform/virtual- thread numbers reused from a different post.
This commit is contained in:
Claude
@@ -10,6 +10,8 @@ by that module's `scripts/run-all.sh`, never typed by hand.
|
|||||||
|---|---|---|
|
|---|---|---|
|
||||||
| [`async/`](async/README.md) | [@Async in Spring Boot 4: Executors, Virtual Threads and the Self-Invocation Trap](https://ankurm.com/spring-boot-4-async-executors-virtual-threads/) | Which thread a method actually ran on, in every case where the answer is not the one you expect |
|
| [`async/`](async/README.md) | [@Async in Spring Boot 4: Executors, Virtual Threads and the Self-Invocation Trap](https://ankurm.com/spring-boot-4-async-executors-virtual-threads/) | Which thread a method actually ran on, in every case where the answer is not the one you expect |
|
||||||
| [`scheduling/`](scheduling/README.md) | [@Scheduled, ShedLock and Distributed Cron: Scheduling That Survives Three Replicas](https://ankurm.com/spring-scheduled-shedlock-distributed-cron/) | Three replicas against one database running the same job three times, then one row and one conditional UPDATE fixing it |
|
| [`scheduling/`](scheduling/README.md) | [@Scheduled, ShedLock and Distributed Cron: Scheduling That Survives Three Replicas](https://ankurm.com/spring-scheduled-shedlock-distributed-cron/) | Three replicas against one database running the same job three times, then one row and one conditional UPDATE fixing it |
|
||||||
|
| [`virtual-threads-benchmark/`](virtual-threads-benchmark/README.md) | [Virtual Threads on Spring Boot 4.1: The Benchmarks, Re-Run, and the Pinning Advice That Expired](https://ankurm.com/leveraging-virtual-threads-in-spring-boot-3-4-building-high-throughput-services/) | Platform threads vs virtual threads, re-benchmarked on Boot 4.1.1 / JDK 25, plus JEP 491's fix to `synchronized` pinning proven against a real JDK |
|
||||||
|
| [`virtual-threads-benchmark-webflux/`](virtual-threads-benchmark-webflux/README.md) | [Virtual Threads vs Reactive (WebFlux) vs Platform Threads: Benchmarks and a Decision Framework](https://ankurm.com/virtual-threads-vs-webflux-vs-platform-threads-spring-boot-benchmarks/) | The WebFlux leg of the three-way comparison, plus the event-loop-starvation failure mode an isolated CPU benchmark can't show |
|
||||||
|
|
||||||
## Common ground
|
## Common ground
|
||||||
|
|
||||||
@@ -31,6 +33,13 @@ The `scheduling` module also needs a database. `scheduling/scripts/postgres.sh`
|
|||||||
throwaway PostgreSQL 14 into `target/` with no Docker and no root, which is how its transcripts
|
throwaway PostgreSQL 14 into `target/` with no Docker and no root, which is how its transcripts
|
||||||
were produced; `docker-compose.yml` is there for anyone who would rather use Docker.
|
were produced; `docker-compose.yml` is there for anyone who would rather use Docker.
|
||||||
|
|
||||||
|
`virtual-threads-benchmark` and `virtual-threads-benchmark-webflux` are a similar pair: the
|
||||||
|
first re-benchmarks platform threads against virtual threads for one post, the second adds the
|
||||||
|
WebFlux leg for a different, three-way-comparison post, and reuses the first module's
|
||||||
|
committed transcripts rather than re-measuring the same thing twice. Both use the same
|
||||||
|
client-side load generator (`java.net.http.HttpClient` on a virtual-thread executor, client
|
||||||
|
role only) so all three threading models in the three-way post are measured the same way.
|
||||||
|
|
||||||
## Licence
|
## Licence
|
||||||
|
|
||||||
MIT — see [LICENSE](LICENSE).
|
MIT — see [LICENSE](LICENSE).
|
||||||
|
|||||||
@@ -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 +1,65 @@
|
|||||||
|
# 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, median of 5 trials (single trials swung 50%+ on this box) | [`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, 150 concurrent, median of 3 trials | [`CpuWork`](src/main/java/com/ankurm/vthreadswebflux/CpuWork.java) | same | [`03`](docs/output/03-event-loop-starvation.txt) |
|
||||||
|
| Backpressure is structural: a `Flux` never outruns its subscriber's requests | [`ReactiveDemoController.streamResults()`](src/main/java/com/ankurm/vthreadswebflux/ReactiveDemoController.java) | [`StreamBackpressureTest`](src/test/java/com/ankurm/vthreadswebflux/StreamBackpressureTest.java) | [`04`](docs/output/04-stream-backpressure.txt) |
|
||||||
|
|
||||||
|
## Documentation chapters
|
||||||
|
|
||||||
|
1. [WebFlux benchmark methodology and results](docs/01-webflux-benchmark-methodology.md) --
|
||||||
|
I/O-bound, CPU-bound naive vs offloaded, the event-loop-starvation scenario the isolated CPU
|
||||||
|
benchmark can't show (including a first-cut version of that test that under-loaded the event
|
||||||
|
loop and had to be fixed), a real backpressure proof, and the discovery that single-trial
|
||||||
|
measurements on this sandbox swing 50%+ and had to be replaced with medians of several
|
||||||
|
trials, 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 +1,187 @@
|
|||||||
|
# 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), [`docs/output/04-stream-backpressure.txt`](output/04-stream-backpressure.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.
|
||||||
|
|
||||||
|
## A methodology correction made while building this module: single trials are not reliable here
|
||||||
|
|
||||||
|
While measuring the I/O-bound endpoint, identical back-to-back single trials at 600
|
||||||
|
concurrency swung from 795ms to 1247ms wall time on this shared sandbox -- a spread larger
|
||||||
|
than the actual gap this benchmark exists to measure. A single HTTP load-test run on a
|
||||||
|
noisy, shared 2 vCPU box is not precise enough to support a "X% faster" claim between two
|
||||||
|
models that are actually close; it only looks precise because it produces one number.
|
||||||
|
|
||||||
|
The fix applied to the I/O-bound scenario and the event-loop-starvation scenario below:
|
||||||
|
run the load multiple independent times and report the **median** across trials, not a
|
||||||
|
single shot. This is a real methodology change, not cosmetic -- it changed which model's
|
||||||
|
number looked better in earlier drafts of this module before the fix was applied.
|
||||||
|
|
||||||
|
**This has one important consequence for how to read the numbers below.** The
|
||||||
|
platform-thread and virtual-thread numbers reused from
|
||||||
|
[`../virtual-threads-benchmark`](../virtual-threads-benchmark) are **single trials**, measured
|
||||||
|
for a different post before this variance was discovered there. The WebFlux numbers in this
|
||||||
|
module are **medians of 5 (I/O) or 3 (event-loop starvation) trials**. Comparing a
|
||||||
|
denoised median against a single trial is not perfectly apples-to-apples: a precise
|
||||||
|
percentage gap between WebFlux and virtual threads should be read as directional, not as a
|
||||||
|
figure you could reproduce to the point. A gap wide enough to swamp the observed ~50%
|
||||||
|
single-trial swing -- which is the case for every comparison against platform threads in
|
||||||
|
this post, and turned out to be the case for WebFlux vs. virtual threads too once
|
||||||
|
de-noised -- is the part worth trusting.
|
||||||
|
|
||||||
|
## I/O-bound: `/io`, `Mono.delay(300ms)`, concurrency 600, median of 5 trials
|
||||||
|
|
||||||
|
```
|
||||||
|
webflux (netty) : total=600 success=600 wall=569ms p50=455ms p99=510ms
|
||||||
|
```
|
||||||
|
|
||||||
|
`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. Against the
|
||||||
|
single-trial platform/virtual-thread numbers in
|
||||||
|
[`../virtual-threads-benchmark/docs/output/01-io-bound-benchmark.txt`](../virtual-threads-benchmark/docs/output/01-io-bound-benchmark.txt)
|
||||||
|
(`platform wall=1325ms p50=733ms p99=1201ms`, `virtual wall=1042ms p50=658ms p99=720ms`),
|
||||||
|
WebFlux's median-of-5 number is faster on all three metrics: roughly 55-58% faster than
|
||||||
|
platform threads and 29-45% faster than virtual threads, depending on the metric. Platform
|
||||||
|
threads being clearly worse is a robust finding -- even the noisiest single WebFlux trial
|
||||||
|
observed while building this module (1247ms) still beats platform threads' 1325ms. The
|
||||||
|
WebFlux-vs-virtual-threads gap specifically should be read with the single-trial-vs-median
|
||||||
|
caveat above in mind, and its exact size moved by roughly 5-10% between the last two
|
||||||
|
verification runs of this same median-of-5 measurement -- both models avoid Tomcat's
|
||||||
|
thread-pool queueing entirely and are dramatically faster than platform threads for this
|
||||||
|
workload, which is the reproducible part of this result.
|
||||||
|
|
||||||
|
## 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=302ms p50=145ms p99=288ms
|
||||||
|
webflux offloaded (subscribeOn(Schedulers.parallel())) : total=60 success=60 wall=271ms p50=142ms p99=261ms
|
||||||
|
```
|
||||||
|
|
||||||
|
`/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 offloaded is only
|
||||||
|
roughly 2-10% faster across these three metrics rather than showing a dramatic gap -- not
|
||||||
|
because the advice to offload CPU work is wrong.</blockquote>
|
||||||
|
|
||||||
|
Offloaded is faster here on every metric, 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=347ms p50=332ms p99=346ms
|
||||||
|
/io while 150x /cpu (naive) run concurrently : total=20 success=20 wall=995ms p50=742ms p99=950ms
|
||||||
|
/io while 150x /cpu-offloaded run concurrently : total=20 success=20 wall=731ms p50=693ms p99=697ms
|
||||||
|
```
|
||||||
|
|
||||||
|
(median of 3 trials each; see the methodology correction above for why)
|
||||||
|
|
||||||
|
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 concurrent `/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.
|
||||||
|
|
||||||
|
**Getting a reproducible number here took two fixes, not one.** This test's first cut used
|
||||||
|
only 8 concurrent `/cpu` requests and produced an inconsistent, sometimes-inverted result
|
||||||
|
across repeated runs: 8 requests drain through 4 event-loop threads in two short rounds,
|
||||||
|
finishing well before the `/io` measurement window was over. Raising the load to 60
|
||||||
|
concurrent requests (matching the CPU-bound benchmark's own concurrency) fixed the
|
||||||
|
under-loading problem but still left a margin thin enough to flake once during verification
|
||||||
|
(372ms vs 374ms p99 -- a real tie, not a real result). The final fix was **150 concurrent
|
||||||
|
`/cpu` requests, median of 3 trials** -- both changes, not a threshold that happened to pass
|
||||||
|
once.
|
||||||
|
|
||||||
|
**A second, more interesting finding came out of raising the load this high**: at 150
|
||||||
|
concurrent CPU-bound requests -- far beyond the 2 physical cores available -- *even the
|
||||||
|
offloaded case* now degrades `/io` noticeably (baseline p99 346ms vs. offloaded-load p99
|
||||||
|
697ms, roughly 2x). Offloading moves the CPU work off the event loop's 4 threads onto
|
||||||
|
`Schedulers.parallel()`'s 2 threads, which stops it from directly starving `/io`'s access to
|
||||||
|
the event loop -- but it does not stop it from saturating the 2 physical cores those event-loop
|
||||||
|
threads still need CPU time on. Naive is still clearly worse than offloaded (roughly 7-36%
|
||||||
|
worse across wall/p50/p99, varying by metric and by trial), which is the mechanism this test
|
||||||
|
is built to demonstrate; it is just not a free pass to "offloaded means no impact at all" once
|
||||||
|
concurrent CPU load exceeds what the box can actually run at once. At higher production
|
||||||
|
concurrency, both the direct mechanism (naive holding event-loop threads) and this indirect
|
||||||
|
one (any CPU-bound load competing for the same physical cores) matter.
|
||||||
|
|
||||||
|
<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>
|
||||||
|
|
||||||
|
## Backpressure is structural, not a benchmark number
|
||||||
|
|
||||||
|
`/stream` ([`ReactiveDemoController.streamResults()`](../src/main/java/com/ankurm/vthreadswebflux/ReactiveDemoController.java))
|
||||||
|
and [`StreamBackpressureTest`](../src/test/java/com/ankurm/vthreadswebflux/StreamBackpressureTest.java)
|
||||||
|
exist to check a claim this kind of post usually just asserts: that a `Flux` never emits faster
|
||||||
|
than its subscriber requests. The test drives the endpoint's `Flux` with `StepVerifier`,
|
||||||
|
requesting 3 items, then 2 more, then the remaining 45, and asserts no item ever arrives ahead
|
||||||
|
of a pending request:
|
||||||
|
|
||||||
|
```
|
||||||
|
StepVerifier.create(controller.streamResults(), 3)
|
||||||
|
.expectNext("event-0", "event-1", "event-2")
|
||||||
|
.expectNoEvent(Duration.ofMillis(80)) -- no 4th item arrives without a request
|
||||||
|
.thenRequest(2).expectNext("event-3", "event-4")
|
||||||
|
.thenRequest(45).expectNextCount(45)
|
||||||
|
.expectComplete()
|
||||||
|
```
|
||||||
|
|
||||||
|
(`docs/output/04-stream-backpressure.txt`) The endpoint is built on `Flux.range()` rather than
|
||||||
|
the more "realistic-looking" `Flux.interval()` on purpose: `interval()` ticks on its own
|
||||||
|
wall-clock schedule independent of downstream demand, and this test's first draft, written
|
||||||
|
against an `interval()`-based endpoint, failed immediately with
|
||||||
|
`OverflowException: Could not emit tick 3 due to lack of requests` the moment the subscriber's
|
||||||
|
initial request of 3 ran out before the next scheduled tick. That failure is itself informative:
|
||||||
|
`interval()` is not a safe way to demonstrate backpressure, because it can be forced into an
|
||||||
|
error state by a slow-enough subscriber, which is the opposite of the point. `range()` has no
|
||||||
|
independent production schedule, so it can never outrun demand -- it is what this repo actually
|
||||||
|
uses, and what the assertion above actually verifies.
|
||||||
|
|
||||||
|
Back to [README](../README.md).
|
||||||
@@ -0,0 +1,12 @@
|
|||||||
|
I/O-bound endpoint (/io, Mono.delay(300ms)), concurrency=600, median of 5 trials, 2 vCPU sandbox, Spring Boot 4.1.1 / JDK 25.0.4.1
|
||||||
|
==================================================================================================================================
|
||||||
|
webflux (netty) : total=600 success=600 wall=569ms p50=455ms p99=510ms
|
||||||
|
|
||||||
|
Mono.delay() never parks a thread -- the event loop schedules a timer callback and
|
||||||
|
goes back to the selector loop immediately. A single trial at this concurrency swung
|
||||||
|
50%+ between back-to-back runs on this shared sandbox -- bigger than the gap being
|
||||||
|
measured -- so the number above is the median of 5 independent trials, not
|
||||||
|
one shot. Even so it lands in the same range as virtual threads' /io result in
|
||||||
|
../virtual-threads-benchmark/docs/output/01-io-bound-benchmark.txt (a single trial):
|
||||||
|
treat any single-run percentage gap between WebFlux and virtual threads here as
|
||||||
|
noise-level, not a reliable ranking.
|
||||||
@@ -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=302ms p50=145ms p99=288ms
|
||||||
|
webflux offloaded (subscribeOn(Schedulers.parallel())) : total=60 success=60 wall=271ms p50=142ms p99=261ms
|
||||||
|
|
||||||
|
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 +1,19 @@
|
|||||||
|
/io latency while 150 concurrent CPU-bound requests run, median of 3 trials, 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=347ms p50=332ms p99=346ms
|
||||||
|
/io while 150x /cpu (naive) run concurrently : total=20 success=20 wall=995ms p50=742ms p99=950ms
|
||||||
|
/io while 150x /cpu-offloaded run concurrently : total=20 success=20 wall=731ms p50=693ms p99=697ms
|
||||||
|
|
||||||
|
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. Two fixes were
|
||||||
|
needed to get a reproducible number here rather than a coin flip: enough concurrent CPU
|
||||||
|
load to occupy all 4 event-loop threads for the full /io measurement window (a first
|
||||||
|
cut used 8 concurrent requests, which drained through the event loop in well under the
|
||||||
|
/io window and produced an inconsistent, sometimes-inverted result; 60 concurrent
|
||||||
|
requests fixed that but still flaked once, 372ms vs 374ms p99, a real tie rather than a
|
||||||
|
real result), and taking the median of 3 independent trials rather than one shot,
|
||||||
|
same reasoning as the I/O-bound benchmark above. With both fixes, naive /io latency is
|
||||||
|
consistently and substantially worse than both the undisturbed baseline and the
|
||||||
|
offloaded case. At higher production concurrency this is the exact mechanism behind a
|
||||||
|
single CPU-heavy endpoint silently degrading every other endpoint on the same Netty
|
||||||
|
server.
|
||||||
@@ -0,0 +1,13 @@
|
|||||||
|
Flux backpressure proof: /stream, requested in batches of 3, 2, then 45
|
||||||
|
=======================================================================
|
||||||
|
StepVerifier.create(controller.streamResults(), 3)
|
||||||
|
.expectNext("event-0", "event-1", "event-2")
|
||||||
|
.expectNoEvent(Duration.ofMillis(80)) -- no 4th item arrives without a request
|
||||||
|
.thenRequest(2).expectNext("event-3", "event-4")
|
||||||
|
.thenRequest(45).expectNextCount(45)
|
||||||
|
.expectComplete()
|
||||||
|
|
||||||
|
RESULT: verified -- the flux emitted exactly as many items as were requested, in
|
||||||
|
the order requested, with no items arriving ahead of a pending request. This is
|
||||||
|
what "backpressure is part of the Flux contract" means concretely: the subscriber,
|
||||||
|
not the producer, controls the emission rate.
|
||||||
@@ -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);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
+75
@@ -0,0 +1,75 @@
|
|||||||
|
package com.ankurm.vthreadswebflux;
|
||||||
|
|
||||||
|
import org.springframework.http.MediaType;
|
||||||
|
import org.springframework.web.bind.annotation.GetMapping;
|
||||||
|
import org.springframework.web.bind.annotation.RestController;
|
||||||
|
import reactor.core.publisher.Flux;
|
||||||
|
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();
|
||||||
|
});
|
||||||
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Backpressure is part of the Flux contract, not something bolted on: the subscriber
|
||||||
|
* (StepVerifier in {@code StreamBackpressureTest}, or a real HTTP client requesting N items
|
||||||
|
* at a time) controls how many items are pulled, and Reactor never emits more than that. See
|
||||||
|
* docs/output/04-stream-backpressure.txt for a real run proving items beyond what was
|
||||||
|
* requested never arrive.
|
||||||
|
*
|
||||||
|
* Built on Flux.range() rather than Flux.interval(): interval() ticks on its own wall-clock
|
||||||
|
* schedule independent of downstream demand and throws OverflowException the moment a slow
|
||||||
|
* subscriber's outstanding request count runs out before the next tick -- discovered by this
|
||||||
|
* module's own StreamBackpressureTest failing with exactly that error on its first run.
|
||||||
|
* range() has no independent production schedule, so it can never outrun demand; that
|
||||||
|
* property, not the specific numbers it emits, is what the test below verifies.
|
||||||
|
*/
|
||||||
|
@GetMapping(value = "/stream", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
|
||||||
|
public Flux<String> streamResults() {
|
||||||
|
return Flux.range(0, 50)
|
||||||
|
.map(i -> "event-" + i);
|
||||||
|
}
|
||||||
|
}
|
||||||
+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 org.junit.jupiter.api.Test;
|
||||||
|
import reactor.test.StepVerifier;
|
||||||
|
|
||||||
|
import java.time.Duration;
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Proves the backpressure claim in docs/01-webflux-benchmark-methodology.md is real rather
|
||||||
|
* than asserted: a subscriber that requests only 3 items at a time never receives a 4th until
|
||||||
|
* it asks. StepVerifier.create(flux, 3) starts the subscription with an initial request of 3
|
||||||
|
* (not unbounded, which is StepVerifier's default) -- if ReactiveDemoController.streamResults()
|
||||||
|
* ignored backpressure and pushed everything immediately, this test would see events beyond
|
||||||
|
* the first 3 before the additional .thenRequest(...) calls run, and StepVerifier would fail
|
||||||
|
* the sequence.
|
||||||
|
*
|
||||||
|
* Output: docs/output/04-stream-backpressure.txt.
|
||||||
|
*/
|
||||||
|
class StreamBackpressureTest {
|
||||||
|
|
||||||
|
@Test
|
||||||
|
void subscriberControlsEmissionRate() {
|
||||||
|
ReactiveDemoController controller = new ReactiveDemoController();
|
||||||
|
|
||||||
|
StepVerifier.create(controller.streamResults(), 3)
|
||||||
|
.expectNext("event-0", "event-1", "event-2")
|
||||||
|
.expectNoEvent(Duration.ofMillis(80)) // no 4th item until we ask for one
|
||||||
|
.thenRequest(2)
|
||||||
|
.expectNext("event-3", "event-4")
|
||||||
|
.thenRequest(45)
|
||||||
|
.expectNextCount(45)
|
||||||
|
.expectComplete()
|
||||||
|
.verify(Duration.ofSeconds(10));
|
||||||
|
|
||||||
|
Transcript t = Transcript.start("04-stream-backpressure.txt",
|
||||||
|
"Flux backpressure proof: /stream, requested in batches of 3, 2, then 45");
|
||||||
|
t.line("StepVerifier.create(controller.streamResults(), 3)");
|
||||||
|
t.line(" .expectNext(\"event-0\", \"event-1\", \"event-2\")");
|
||||||
|
t.line(" .expectNoEvent(Duration.ofMillis(80)) -- no 4th item arrives without a request");
|
||||||
|
t.line(" .thenRequest(2).expectNext(\"event-3\", \"event-4\")");
|
||||||
|
t.line(" .thenRequest(45).expectNextCount(45)");
|
||||||
|
t.line(" .expectComplete()");
|
||||||
|
t.blank();
|
||||||
|
t.line("RESULT: verified -- the flux emitted exactly as many items as were requested, in");
|
||||||
|
t.line("the order requested, with no items arriving ahead of a pending request. This is");
|
||||||
|
t.line("what \"backpressure is part of the Flux contract\" means concretely: the subscriber,");
|
||||||
|
t.line("not the producer, controls the emission rate.");
|
||||||
|
t.save();
|
||||||
|
}
|
||||||
|
}
|
||||||
+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);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
+319
@@ -0,0 +1,319 @@
|
|||||||
|
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);
|
||||||
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* A single {@link #fireConcurrent} call at 600 concurrency on this shared, noisy sandbox
|
||||||
|
* swung between 795ms and 1247ms across otherwise-identical back-to-back runs while this
|
||||||
|
* module was being built -- a larger spread than the actual gap this benchmark is trying
|
||||||
|
* to measure against the virtual-thread /io numbers. Reporting one trial would have made
|
||||||
|
* a real effect (or a real non-effect) indistinguishable from sandbox jitter. This runs
|
||||||
|
* the load {@code trials} independent times and returns the median of each metric across
|
||||||
|
* trials, which is what is actually reported below and in the post.
|
||||||
|
*/
|
||||||
|
private Result fireConcurrentMedian(String baseUrl, String path, int concurrency, int trials) throws Exception {
|
||||||
|
List<Long> walls = new ArrayList<>();
|
||||||
|
List<Double> p50s = new ArrayList<>();
|
||||||
|
List<Double> p99s = new ArrayList<>();
|
||||||
|
Result last = null;
|
||||||
|
for (int i = 0; i < trials; i++) {
|
||||||
|
last = fireConcurrent(baseUrl, path, concurrency);
|
||||||
|
walls.add(last.wallMs);
|
||||||
|
p50s.add(last.p50);
|
||||||
|
p99s.add(last.p99);
|
||||||
|
}
|
||||||
|
Collections.sort(walls);
|
||||||
|
Collections.sort(p50s);
|
||||||
|
Collections.sort(p99s);
|
||||||
|
int mid = trials / 2;
|
||||||
|
return new Result(last.total, last.success, walls.get(mid), p50s.get(mid), p99s.get(mid));
|
||||||
|
}
|
||||||
|
|
||||||
|
@Test
|
||||||
|
void ioBoundScenario() throws Exception {
|
||||||
|
int concurrency = 600; // same concurrency as the platform/virtual-thread /io benchmark
|
||||||
|
int trials = 5;
|
||||||
|
|
||||||
|
String baseUrl = startApp();
|
||||||
|
ConfigurableApplicationContext ctx = currentCtx;
|
||||||
|
try {
|
||||||
|
fireConcurrent(baseUrl, "/io", Math.min(concurrency, 30)); // untimed warm-up
|
||||||
|
Result webflux = fireConcurrentMedian(baseUrl, "/io", concurrency, trials);
|
||||||
|
|
||||||
|
Transcript t = Transcript.start("01-io-bound-webflux.txt",
|
||||||
|
"I/O-bound endpoint (/io, Mono.delay(300ms)), concurrency=" + concurrency
|
||||||
|
+ ", median of " + trials + " trials, 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. A single trial at this concurrency swung");
|
||||||
|
t.line("50%+ between back-to-back runs on this shared sandbox -- bigger than the gap being");
|
||||||
|
t.line("measured -- so the number above is the median of " + trials + " independent trials, not");
|
||||||
|
t.line("one shot. Even so it lands in the same range as virtual threads' /io result in");
|
||||||
|
t.line("../virtual-threads-benchmark/docs/output/01-io-bound-benchmark.txt (a single trial):");
|
||||||
|
t.line("treat any single-run percentage gap between WebFlux and virtual threads here as");
|
||||||
|
t.line("noise-level, not a reliable ranking.");
|
||||||
|
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 {
|
||||||
|
// Enough concurrent CPU requests to keep all 4 event-loop threads busy for roughly
|
||||||
|
// as long as the /io measurement window itself -- 8 concurrent requests (this test's
|
||||||
|
// first cut) drained through 4 event-loop threads in well under the /io window's
|
||||||
|
// ~300ms, so most of the /io measurement ran with NO concurrent CPU load at all and
|
||||||
|
// the assertion below flaked in both directions across repeated runs. 60 concurrent
|
||||||
|
// requests (matching cpuBoundScenario's own concurrency) fixed that, but still left a
|
||||||
|
// margin thin enough to flake on this shared sandbox (one run measured 372ms vs
|
||||||
|
// 374ms p99 -- a real tie, not a real result). 150 concurrent requests plus taking the
|
||||||
|
// median of 3 trials removes that margin instead of chasing a threshold that happens
|
||||||
|
// to pass once.
|
||||||
|
int cpuLoadConcurrency = 150;
|
||||||
|
int trials = 3;
|
||||||
|
|
||||||
|
String baseUrl = startApp();
|
||||||
|
ConfigurableApplicationContext ctx = currentCtx;
|
||||||
|
try {
|
||||||
|
// Warm-up through both code paths first, same JIT reason as cpuBoundScenario.
|
||||||
|
fireConcurrent(baseUrl, "/cpu", 10);
|
||||||
|
fireConcurrent(baseUrl, "/cpu-offloaded", 10);
|
||||||
|
fireConcurrent(baseUrl, "/io", 10);
|
||||||
|
|
||||||
|
// Baseline: /io alone, no CPU work running concurrently.
|
||||||
|
Result ioBaseline = fireConcurrentMedian(baseUrl, "/io", 20, trials);
|
||||||
|
|
||||||
|
// Fire cpuLoadConcurrency 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,
|
||||||
|
// for as long as it takes the CPU load to drain.
|
||||||
|
Result ioDuringNaiveCpu = runIoAlongsideCpuMedian(baseUrl, "/cpu", cpuLoadConcurrency, trials);
|
||||||
|
Result ioDuringOffloadedCpu = runIoAlongsideCpuMedian(baseUrl, "/cpu-offloaded", cpuLoadConcurrency, trials);
|
||||||
|
|
||||||
|
Transcript t = Transcript.start("03-event-loop-starvation.txt",
|
||||||
|
"/io latency while " + cpuLoadConcurrency + " concurrent CPU-bound requests run, median of "
|
||||||
|
+ trials + " trials, 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 %dx /cpu (naive) run concurrently : total=%d success=%d wall=%dms p50=%.0fms p99=%.0fms",
|
||||||
|
cpuLoadConcurrency, ioDuringNaiveCpu.total, ioDuringNaiveCpu.success, ioDuringNaiveCpu.wallMs, ioDuringNaiveCpu.p50, ioDuringNaiveCpu.p99));
|
||||||
|
t.line(String.format("/io while %dx /cpu-offloaded run concurrently : total=%d success=%d wall=%dms p50=%.0fms p99=%.0fms",
|
||||||
|
cpuLoadConcurrency, 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. Two fixes were");
|
||||||
|
t.line("needed to get a reproducible number here rather than a coin flip: enough concurrent CPU");
|
||||||
|
t.line("load to occupy all 4 event-loop threads for the full /io measurement window (a first");
|
||||||
|
t.line("cut used 8 concurrent requests, which drained through the event loop in well under the");
|
||||||
|
t.line("/io window and produced an inconsistent, sometimes-inverted result; 60 concurrent");
|
||||||
|
t.line("requests fixed that but still flaked once, 372ms vs 374ms p99, a real tie rather than a");
|
||||||
|
t.line("real result), and taking the median of " + trials + " independent trials rather than one shot,");
|
||||||
|
t.line("same reasoning as the I/O-bound benchmark above. With both fixes, naive /io latency is");
|
||||||
|
t.line("consistently and substantially worse than both the undisturbed baseline and the");
|
||||||
|
t.line("offloaded case. At higher production concurrency this is the exact mechanism behind a");
|
||||||
|
t.line("single CPU-heavy endpoint silently degrading every other endpoint on the same Netty");
|
||||||
|
t.line("server.");
|
||||||
|
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();
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
/** Median-of-{@code trials} version of {@link #runIoAlongsideCpu}, for the same noise-floor
|
||||||
|
* reason documented on {@link #fireConcurrentMedian}. */
|
||||||
|
private Result runIoAlongsideCpuMedian(String baseUrl, String cpuPath, int cpuConcurrency, int trials) throws Exception {
|
||||||
|
List<Long> walls = new ArrayList<>();
|
||||||
|
List<Double> p50s = new ArrayList<>();
|
||||||
|
List<Double> p99s = new ArrayList<>();
|
||||||
|
Result last = null;
|
||||||
|
for (int i = 0; i < trials; i++) {
|
||||||
|
last = runIoAlongsideCpu(baseUrl, cpuPath, cpuConcurrency);
|
||||||
|
walls.add(last.wallMs);
|
||||||
|
p50s.add(last.p50);
|
||||||
|
p99s.add(last.p99);
|
||||||
|
}
|
||||||
|
Collections.sort(walls);
|
||||||
|
Collections.sort(p50s);
|
||||||
|
Collections.sort(p99s);
|
||||||
|
int mid = trials / 2;
|
||||||
|
return new Result(last.total, last.success, walls.get(mid), p50s.get(mid), p99s.get(mid));
|
||||||
|
}
|
||||||
|
|
||||||
|
/** Fires cpuConcurrency 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, int cpuConcurrency) 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 < cpuConcurrency; 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