[Bug]: Shared OTel plugin state mixes concurrent LMI invocations
I maintainer di solito rispondono entro 1 giorno
Valutazione
- Difficoltà
- 4/5
- Tempo stimato
- 3-5 giorni
- Idoneità per principianti
- 35/100
- Tipo di issue
- Bug
- Chiarezza
- Specificata chiaramente
- Stato di attività
- Ferma
- Stack tecnologico
- java
- Ambito
- observability
Direzione di ricerca
Start with ExecutionOtelPlugin.java and InvocationOtelPlugin.java, then read PluginRunner.java to understand the real SDK registration and dispatch path. The issue asks for overlapping-invocation tests covering both OTel views, lifecycle cases and reused operation IDs, plus deployed LMI coverage proving same-JVM overlap; verify those outcomes and any registration migration documentation. PR #721 is already open proposing per-invocation plugin instances.
Scritto dal modello di indicizzazione a partire dal testo della issue.
Descrizione
Expected Behavior
Concurrent Lambda Managed Instances invocations must keep their OTel execution state isolated. Each operation must be attributed to its owning execution, and ending/resetting one invocation must not invalidate another invocation’s telemetry.
Actual Behavior
A single published ExecutionOtelPlugin instance stores the current execution ARN and span state in mutable fields. Interleaving onInvocationStart(A), onInvocationStart(B) and onOperationEnd(A-step) exports A-step with B’s durable.execution.arn. The same sequence using separate plugin instances correctly exports A’s ARN. Java LMI shares the handler/JVM between request threads, so environment-lifetime plugin objects can be used by overlapping invocations.
This was reproduced locally with the published plugin and an in-memory exporter. The interleaving is deterministic; it is not a fresh cloud concurrency measurement.
Steps to Reproduce
Use a scratch Maven project with software.amazon.lambda.durable:aws-durable-execution-sdk-java-plugin-otel:2.2.1 and io.opentelemetry:opentelemetry-sdk-testing:1.65.0, including their transitive dependencies. Compile and run OtelLmiStateProbe below on Java 25. No Lambda credentials or service calls are involved.
The shared-object case reports A’s operation under .../B; the separate-instance control reports .../A. Assertions describe the pre-fix behavior.
Reproduction code
import java.time.Instant;
import java.util.*;
import com.fasterxml.jackson.databind.ObjectMapper;
import io.opentelemetry.api.common.AttributeKey;
import io.opentelemetry.sdk.trace.SdkTracerProvider;
import io.opentelemetry.sdk.trace.samplers.Sampler;
import io.opentelemetry.sdk.trace.export.SimpleSpanProcessor;
import io.opentelemetry.sdk.testing.exporter.InMemorySpanExporter;
import software.amazon.lambda.durable.otel.*;
import software.amazon.lambda.durable.plugin.*;
public class OtelLmiStateProbe {
static final Instant START=Instant.parse("2026-10-02T00:00:00Z");
static final String BASE="arn:aws:lambda:us-west-2:123456789012:function:probe:1/durable-execution/probe/";
static ExecutionOtelPlugin plugin(InMemorySpanExporter exporter) {
return new ExecutionOtelPlugin(
SdkTracerProvider.builder().setSampler(Sampler.alwaysOn()).addSpanProcessor(SimpleSpanProcessor.create(exporter)),
OtelPluginConfig.builder().enableMdc(false).contextExtractor(() -> null).build());
}
public static void main(String[] args) throws Exception {
var results=new ArrayList<Map<String,Object>>();
for (boolean shared:new boolean[]{true,false}) {
var exporter=InMemorySpanExporter.create();
var a=plugin(exporter);
var b=shared?a:plugin(exporter);
a.onInvocationStart(new InvocationInfo("request-A",BASE+"A",true,START));
b.onInvocationStart(new InvocationInfo("request-B",BASE+"B",true,START));
a.onOperationEnd(new OperationEndInfo("A-step","A-step","STEP","Step",null,START,START.plusSeconds(1),"SUCCEEDED",null,false,null));
var spans=exporter.getFinishedSpanItems().stream().filter(s -> s.getName().equals("A-step")).toList();
if(spans.size()!=1)throw new AssertionError("Expected one A-step span");
var assignedArn=spans.getFirst().getAttributes().get(AttributeKey.stringKey("durable.execution.arn"));
if(!(BASE+(shared?"B":"A")).equals(assignedArn))throw new AssertionError("Unexpected assignment: "+assignedArn);
results.add(Map.of("sharedPluginInstance",shared,"operation","A-step","expectedExecution",BASE+"A","actualExecution",assignedArn));
}
System.out.println(new ObjectMapper().writeValueAsString(Map.of("sdkAndPluginVersion","2.2.1","results",results,"checksPassed",true)));
}
}
SDK Version
Core SDK / OTel plugin 2.2.1. Main rechecked at 78574e3051aac9b5ca00a0c1fd2b91eab72a9e15.
Java Version
Corretto 25.0.4.1
Is this a regression?
Unknown; no last-known-good version has been established.
Additional Context
The Java LMI runtime uses request threads sharing one handler and static state. Relevant code: ExecutionOtelPlugin, InvocationOtelPlugin, and plugin dispatch.
PR #721 proposes per-invocation plugin instances. This issue tracks the OTel defect and its release/validation outcome. The core executor/deadline issues in #726 and the cloud-test request in #727 remain separate work.
Acceptance criteria
- Verify both OTel views with overlapping invocations and reused operation IDs through the real SDK registration/dispatch path.
- Ensure invocation completion, suspension, retries, errors and late hooks cannot alter another invocation’s spans or context.
- Preserve shared provider/exporter ownership and add deployed LMI coverage with proven same-JVM overlap.
- Document the factory migration and compatible SDK/plugin versions if the fix changes registration.
Related cross-SDK tracking: https://github.com/aws/aws-durable-execution-sdk-js/issues/950
- Lingua principale
- Java
- Stelle
- 28
- Fork
- 13
- Merge medio
- 2g 3h
- PR unite (30g)
- 44
Preparare l'ambiente
- Nessun Dockerfile né file Docker Compose
- Ha un 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 aws/aws-durable-execution-sdk-java
-
documentation pkg:sdk
Difficoltà 1/5 1-3 ore Idoneità per principianti 88/100
aws/aws-durable-execution-sdk-java#645 · 1 commento ·
I maintainer di solito rispondono entro 1 giorno
-
enhancement
Difficoltà 2/5 1-3 ore Idoneità per principianti 68/100
aws/aws-durable-execution-sdk-java#300 ·
I maintainer di solito rispondono entro 1 giorno
-
enhancement needs-triage
Difficoltà 5/5 Più di una settimana Idoneità per principianti 35/100
aws/aws-durable-execution-sdk-java#779 ·
I maintainer di solito rispondono entro 1 giorno
-
[Bug]: root handler instrumentation misses the canonical OTel execution contextForse già presa Una pull request collegata a questa issue è aperta o già unita. Apertaneeds-triage
Difficoltà 5/5 Più di una settimana Idoneità per principianti 40/100
aws/aws-durable-execution-sdk-java#770 ·
I maintainer di solito rispondono entro 1 giorno
-
[Feature]: Propagate per-operation trace context for chained invokesForse già presa Una pull request collegata a questa issue è aperta o già unita. Apertaenhancement needs-triage
Difficoltà 5/5 Più di una settimana Idoneità per principianti 38/100
aws/aws-durable-execution-sdk-java#764 ·
I maintainer di solito rispondono entro 1 giorno
Tutte le issue di aws/aws-durable-execution-sdk-java
Issue simili
-
Clock.MakeTime fails to validate hour, minute, and second ranges due to inert Calendar.set try-catchAperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 84/100
mit-cml/appinventor-sources#4139 ·
I maintainer di solito rispondono entro 1 giorno
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 62/100
Vakore/ZappierGames#81 ·
-
proposal
Difficoltà 2/5 1-3 ore Idoneità per principianti 85/100
beemdevelopment/Aegis#1843 · 1 reazione ·
-
[Bug] Logo style setting missing and `classic` style not applied across multiple platforms (v3.1.0)Aperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 75/100
Stirling-Tools/Stirling-PDF#8382 · 1 reazione ·
I maintainer di solito rispondono entro 2 giorni
-
>enhancement needs:triage
Difficoltà 2/5 1-3 ore Idoneità per principianti 70/100
elastic/elasticsearch#161191 ·
I maintainer di solito rispondono entro 1 giorno