Synchronous response parsing pins Java 21 virtual threads during body reads
I maintainer di solito rispondono entro 1 giorno
Nessuno ha ancora preso questa issue.
Valutazione
- Difficoltà
- 3/5
- Tempo stimato
- Mezza giornata
- Idoneità per principianti
- 42/100
- Tipo di issue
- Bug
- Chiarezza
- Specificata chiaramente
- Stato di attività
- Attiva
- Ambito
- api, performance
Direzione di ricerca
Start with HttpResponseFor.kt, where parseable wraps the parsed response in Kotlin's synchronized lazy, and trace how the initializer reads the body through ObjectMapper.readValue. The pinning repro in the issue is a standalone Java program with a raw control mode, so run it before and after a change. Done means parsed mode no longer reports SynchronizedLazyImpl in jdk.VirtualThreadPinned events while one-time parsing semantics stay intact.
Scritto dal modello di indicizzazione a partire dal testo della issue.
Descrizione
Problem
Synchronous non-streaming calls can pin Java 21 virtual threads while waiting for response body data, including with the SDK's default OkHttp transport.
HttpResponseFor.kt caches the parsed response using Kotlin's default synchronized lazy. Its initializer performs JSON parsing and reads the response body while holding the lazy monitor. If that read blocks, the virtual thread cannot unmount from its carrier on Java 21.
With slow responses and concurrent calls, this can occupy the available carriers and delay unrelated virtual-thread work.
Reproduction conditions
- Java 21.0.4
openai-java-client-okhttp:4.71.0(default OkHttp 4.12.0)- JVM option:
-Djdk.virtualThreadScheduler.parallelism=2 - A local HTTP/1.1 server running on platform threads returns a valid non-streaming chat completion: send the response headers and the first byte of JSON, flush, wait 500ms, then send the remaining body.
- Run
client.chat().completions().create(params)on virtual threads, submitting seven calls together per round, for three rounds. - Record JFR
jdk.VirtualThreadPinnedevents with a 10ms threshold and stack traces.
Observed pinning
JFR recorded nine pinning events containing SynchronizedLazyImpl, with a maximum duration of approximately 505ms. A representative stack, shown from the call entry to the blocking read, is:
ChatCompletionServiceImpl.create
HttpResponseForKt$parseable$1.parse
kotlin.SynchronizedLazyImpl.getValue
HttpResponseForKt$parseable$1$parsed$1.invoke
ObjectMapper.readValue
okhttp3.internal.connection.Exchange$ResponseBodySource.read
okhttp3.internal.http1.Http1ExchangeCodec$FixedLengthSource.read
okio.InputStreamSource.read
sun.nio.ch.NioSocketImpl.timedRead
sun.nio.ch.Poller.pollIndirect
java.lang.VirtualThread.parkOnCarrierThread
As a control, reading response.body() directly after withRawResponse().create(params), under the same slow-body conditions, produced no pinning events above the threshold.
Expected behavior
Response parsing should preserve its one-time initialization semantics without holding a JVM monitor across blocking response-body reads, so synchronous calls can release their carriers while waiting for I/O on Java 21.
Runnable reproduction
Use JDK 21 and Maven. The program uses only a local server and a dummy API key.
Save this as pom.xml:
<project xmlns="http://maven.apache.org/POM/4.0.0">
<modelVersion>4.0.0</modelVersion>
<groupId>repro</groupId>
<artifactId>virtual-thread-pinning</artifactId>
<version>1.0</version>
<dependencies>
<dependency>
<groupId>com.openai</groupId>
<artifactId>openai-java-client-okhttp</artifactId>
<version>4.71.0</version>
</dependency>
</dependencies>
</project>
Save this as PinningRepro.java in the same directory:
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import com.sun.net.httpserver.HttpServer;
import jdk.jfr.Recording;
import jdk.jfr.consumer.RecordingFile;
import java.net.InetSocketAddress;
import java.nio.charset.StandardCharsets;
import java.nio.file.Path;
import java.time.Duration;
import java.util.ArrayList;
import java.util.concurrent.CountDownLatch;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import java.util.concurrent.TimeUnit;
public class PinningRepro {
public static void main(String[] args) throws Exception {
boolean raw = args.length > 0 && args[0].equals("raw");
byte[] json = ("{\"id\":\"test\",\"object\":\"chat.completion\",\"created\":1,"
+ "\"model\":\"test\",\"choices\":[{\"index\":0,\"message\":{\"role\":\"assistant\","
+ "\"content\":\"ok\"},\"finish_reason\":\"stop\"}]}")
.getBytes(StandardCharsets.UTF_8);
var serverThreads = Executors.newFixedThreadPool(16, Thread.ofPlatform().factory());
var server = HttpServer.create(new InetSocketAddress("127.0.0.1", 0), 0);
server.setExecutor(serverThreads);
server.createContext("/v1/chat/completions", exchange -> {
try (exchange) {
exchange.getRequestBody().readAllBytes();
exchange.getResponseHeaders().set("Content-Type", "application/json");
exchange.sendResponseHeaders(200, json.length);
var out = exchange.getResponseBody();
out.write(json, 0, 1);
out.flush();
try {
Thread.sleep(500);
} catch (InterruptedException error) {
Thread.currentThread().interrupt();
return;
}
out.write(json, 1, json.length - 1);
}
});
server.start();
var client = OpenAIOkHttpClient.builder()
.baseUrl("http://127.0.0.1:" + server.getAddress().getPort() + "/v1")
.apiKey("dummy").maxRetries(0).timeout(Duration.ofSeconds(5)).build();
var params = ChatCompletionCreateParams.builder().model("test").addUserMessage("test").build();
Path file = Path.of(raw ? "raw.jfr" : "pinning.jfr");
try (var recording = new Recording();
var callers = Executors.newThreadPerTaskExecutor(Thread.ofVirtual().name("repro-", 1).factory())) {
// Exclude class initialization and the platform-thread warmup from JFR.
client.chat().completions().create(params);
recording.enable("jdk.VirtualThreadPinned").withThreshold(Duration.ofMillis(10)).withStackTrace();
recording.start();
for (int round = 0; round < 3; round++) {
var start = new CountDownLatch(1);
var futures = new ArrayList<Future<?>>();
for (int i = 0; i < 7; i++) {
futures.add(callers.submit(() -> {
start.await();
if (raw) {
try (var response = client.chat().completions().withRawResponse().create(params)) {
response.body().readAllBytes();
}
} else {
client.chat().completions().create(params);
}
return null;
}));
}
start.countDown();
for (var future : futures) {
future.get(10, TimeUnit.SECONDS);
}
}
recording.stop();
recording.dump(file);
} finally {
client.close();
server.stop(0);
serverThreads.shutdownNow();
}
var pins = RecordingFile.readAllEvents(file).stream()
.filter(e -> e.getEventType().getName().equals("jdk.VirtualThreadPinned")).toList();
System.out.println("mode=" + (raw ? "raw" : "parsed") + ", completed=21, pinned events=" + pins.size());
pins.stream().findFirst().ifPresent(System.out::println);
}
}
Run:
mvn -q dependency:build-classpath -Dmdep.outputFile=classpath.txt
javac --release 21 -cp "$(cat classpath.txt)" PinningRepro.java
java -Djdk.virtualThreadScheduler.parallelism=2 -cp ".:$(cat classpath.txt)" PinningRepro
jfr print --events jdk.VirtualThreadPinned pinning.jfr
# Control: read the raw response body without SDK lazy parsing.
java -Djdk.virtualThreadScheduler.parallelism=2 -cp ".:$(cat classpath.txt)" PinningRepro raw
The parsed mode produces pinning events whose full JFR stacks include SynchronizedLazyImpl and the socket read. The raw mode is the control. Event counts vary with scheduling; the standalone program produced eight events in my Java 21.0.4 run.
- Lingua principale
- Kotlin
- Stelle
- 1.5k
- Fork
- 266
- Merge medio
- 10h 11m
- PR unite (30g)
- 141
Preparare l'ambiente
Avvia il container di sviluppo del progetto nel browser, con il tuo account GitHub.
- Nessun Dockerfile né file Docker Compose
- Nessun modello di pull request
- Leggi la guida per i contributori
Come iniziare
- Leggi tutta la issue e poi la guida ai contributi del progetto.
- Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
- Fai un fork del repository e lavora su un branch.
- Apri una pull request che faccia riferimento al numero della issue.
Altre issue di openai/openai-java
-
HttpRequest.url() logs spaces in path segments as literal plus signsForse già presa @sylvesterkaczmarek l’ha presa 54 giorni fa. Aperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 78/100
openai/openai-java#886 · 2 commenti ·
I maintainer di solito rispondono entro 1 giorno
-
Make diarized transcription duration and task optionalForse già presa @wskr00 l’ha presa 84 giorni fa. Aperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 68/100
openai/openai-java#802 · 2 commenti · 3 reazioni ·
I maintainer di solito rispondono entro 1 giorno
-
AsyncStreamResponse can hang when the subscriber executor rejects workForse già presa @sylvesterkaczmarek l’ha presa 37 giorni fa. Aperta
Difficoltà 3/5 1-2 giorni Idoneità per principianti 74/100
openai/openai-java#973 ·
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 5/5 Più di una settimana Idoneità per principianti 35/100
openai/openai-java#956 ·
I maintainer di solito rispondono entro 1 giorno
-
Async internal plumbing bypasses both `dispatcherExecutorService` and `streamHandlerExecutor`, forcing a hop through `ForkJoinPool.commonPool()`Forse già presa @sylvesterkaczmarek l’ha presa 31 giorni fa. Aperta
Difficoltà 5/5 Più di una settimana Idoneità per principianti 35/100
openai/openai-java#952 ·
I maintainer di solito rispondono entro 1 giorno
Tutte le issue di openai/openai-java
Issue simili
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 79/100
supabase-community/supabase-kt#1413 ·
I maintainer di solito rispondono entro 5 giorni
-
[Submission] 抖音火山版Apertasubmit-adaption submit-adaption-pre
Difficoltà 2/5 1-3 ore Idoneità per principianti 62/100
BetterAndroid/android-notification-icon-project#744 · 1 commento ·
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 64/100
utopia-rise/godot-jvm#1004 ·
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 74/100
pedroSG94/RootEncoder#2213 ·
I maintainer di solito rispondono entro 2 giorni
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 65/100
elm-tooling/intellij-elm#85 · 1 commento ·