fix(middleware): avoid event-loop blocking in WorkspaceContextMiddleware onSystemPrompt by rudy2steiner · Pull Request #2632 · agentscope-ai/agentscope-java

rudy2steiner · 2026-08-09T08:39:29Z

Background

WorkspaceContextMiddleware.onSystemPrompt() assembled workspace context synchronously on the caller thread #2578. In WebFlux deployments, that caller can be a Netty event-loop thread.
This PR fixes the issue by moving that workspace context assembly work off the caller thread.
Other middleware paths are not changed here; they need a separate audit and shared blocking-boundary policy.

What's Changed

Deferred workspace context assembly with Mono.fromCallable(...).
Scheduled the blocking assembly work on Schedulers.boundedElastic().
Added a regression test that verifies workspace context reads do not run on the caller thread.

…are onSystemPrompt

codecov · 2026-08-09T08:57:03Z

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

Copilot

Pull request overview

This PR addresses #2578 by ensuring WorkspaceContextMiddleware.onSystemPrompt() does not assemble workspace context on the caller thread (which can be a Netty event-loop thread under WebFlux), moving that potentially blocking work onto Reactor’s bounded-elastic scheduler.

Changes:

Wrap workspace context assembly in Mono.fromCallable(...) and schedule it via subscribeOn(Schedulers.boundedElastic()).
Add a regression test to verify workspace context reads do not execute on the caller thread.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
agentscope-harness/src/main/java/io/agentscope/harness/agent/middleware/WorkspaceContextMiddleware.java	Moves workspace context assembly off the caller thread using bounded-elastic scheduling.
agentscope-harness/src/test/java/io/agentscope/harness/agent/middleware/WorkspaceContextMiddlewareMemoryPromptTest.java	Adds a regression test asserting workspace reads occur on a different thread than the caller.

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

zouyx

Codecov 的覆盖度问题能修复吗？

rudy2steiner · 2026-08-10T02:02:58Z

Codecov 的覆盖度问题能修复吗？
我覆盖下

AgentScopeJavaBot

🤖 AI Review

This PR moves the blocking workspace-context assembly (filesystem reads for AGENTS.md, MEMORY.md, knowledge files) off the reactive subscriber thread by wrapping buildWorkspaceSection() in Mono.fromCallable().subscribeOn(Schedulers.boundedElastic()). The intent is sound — preventing event-loop blocking when workspace files are read synchronously. A new test validates that the callable indeed executes on a different thread than the caller. However, the removal of the empty-section early-return guard and the lack of onErrorResume for graceful degradation are worth addressing before merge.

AgentScopeJavaBot · 2026-08-12T01:42:33Z

+                            return base + separator + section;
+                        })
+                .subscribeOn(Schedulers.boundedElastic());
    }


[major] Missing onErrorResume for graceful degradation. buildWorkspaceSection() performs multiple filesystem reads. If any throws, the Mono.fromCallable will propagate the error through the middleware chain, failing the entire agent call. For a supplementary-context middleware, this is overly aggressive. Consider adding .onErrorResume(e -> { log.warn(...); return Mono.just(currentPrompt); }) after subscribeOn to degrade gracefully.

AgentScopeJavaBot · 2026-08-12T01:42:33Z

-        String separator = base.isEmpty() || base.endsWith("\n") ? "" : "\n";
-        return Mono.just(base + separator + section);
+        return Mono.fromCallable(
+                        () -> {


[minor] Removed empty-section early-return guard. The original code had if (section.isEmpty()) return Mono.just(currentPrompt) which short-circuited when no workspace files existed. The new code always concatenates the section (even if empty), which may append an unnecessary trailing newline. Consider restoring the isEmpty guard inside the callable.

AgentScopeJavaBot · 2026-08-12T01:42:33Z

    @TempDir Path workspace;

+    @Test
+    void onSystemPromptBuildsWorkspaceContextOnBoundedElastic() {


[nitpick] The test assertion assertFalse(callerThread[0]) is somewhat fragile — it only checks that the caller thread differs from the current thread at assertion time. If the test happens to run on the bounded-elastic scheduler itself, the assertion would still pass but the intent (offloading to a different scheduler) would be better captured by checking the thread name contains 'boundedElastic'.

AgentScopeJavaBot

🤖 AI Review

This PR moves the blocking workspace-context assembly (filesystem reads for AGENTS.md, MEMORY.md, knowledge files) off the reactive subscriber thread by wrapping buildWorkspaceSection() in Mono.fromCallable().subscribeOn(Schedulers.boundedElastic()). The intent is sound — preventing event-loop blocking when workspace files are read synchronously. A new test validates that the callable indeed executes on a different thread than the caller. However, the removal of the empty-section early-return guard and the lack of onErrorResume for graceful degradation are worth addressing before merge.

AgentScopeJavaBot · 2026-08-12T01:43:50Z

+                            return base + separator + section;
+                        })
+                .subscribeOn(Schedulers.boundedElastic());
    }


[major] Missing onErrorResume for graceful degradation. buildWorkspaceSection() performs multiple filesystem reads. If any throws, the Mono.fromCallable will propagate the error through the middleware chain, failing the entire agent call. For a supplementary-context middleware, this is overly aggressive. Consider adding .onErrorResume(e -> { log.warn(...); return Mono.just(currentPrompt); }) after subscribeOn to degrade gracefully.

AgentScopeJavaBot · 2026-08-12T01:43:50Z

-        String separator = base.isEmpty() || base.endsWith("\n") ? "" : "\n";
-        return Mono.just(base + separator + section);
+        return Mono.fromCallable(
+                        () -> {


[minor] Removed empty-section early-return guard. The original code had if (section.isEmpty()) return Mono.just(currentPrompt) which short-circuited when no workspace files existed. The new code always concatenates the section (even if empty), which may append an unnecessary trailing newline. Consider restoring the isEmpty guard inside the callable.

AgentScopeJavaBot · 2026-08-12T01:43:50Z

    @TempDir Path workspace;

+    @Test
+    void onSystemPromptBuildsWorkspaceContextOnBoundedElastic() {


[nitpick] The test assertion assertFalse(callerThread[0]) is somewhat fragile — it only checks that the caller thread differs from the current thread at assertion time. If the test happens to run on the bounded-elastic scheduler itself, the assertion would still pass but the intent (offloading to a different scheduler) would be better captured by checking the thread name contains 'boundedElastic'.

rudy2steiner · 2026-08-12T13:59:07Z

@zouyx 有空再看下: 删除不可达/无意义的空 section 分支，另一处是补了 currentPrompt 为 null 和不以换行结尾时的测试

rudy2steiner added 2 commits August 9, 2026 01:13

fix(middleware): avoid event-loop blocking in WorkspaceContextMiddlew…

9c31ea6

…are onSystemPrompt

Merge branch 'main' into none_blocking

b320802

rudy2steiner changed the title ~~fix(middleware): avoid event-loop blocking in WorkspaceContextMiddlew…~~ fix(middleware): avoid event-loop blocking in WorkspaceContextMiddleware onSystemPrompt Aug 9, 2026

zouyx approved these changes Aug 9, 2026

View reviewed changes

zouyx requested a lite review from Copilot August 9, 2026 10:55

Copilot started reviewing on behalf of zouyx August 9, 2026 10:56 View session

Copilot AI reviewed Aug 9, 2026

View reviewed changes

Comment thread ...harness/src/main/java/io/agentscope/harness/agent/middleware/WorkspaceContextMiddleware.java

zouyx self-assigned this Aug 9, 2026

Potential fix for pull request finding

f1c544d

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

zouyx reviewed Aug 9, 2026

View reviewed changes

AgentScopeJavaBot added bug Something isn't working area/harness agentscope-harness (test/runtime support) labels Aug 11, 2026

rudy2steiner added 2 commits August 11, 2026 22:56

Merge branch 'main' into none_blocking

1482a1e

fix(test): coverage issue

f9e4ca4

AgentScopeJavaBot reviewed Aug 12, 2026

View reviewed changes

Merge branch 'main' into none_blocking

3dacad1

zouyx approved these changes Aug 12, 2026

View reviewed changes

zouyx merged commit 40e77c0 into agentscope-ai:main Aug 12, 2026
6 checks passed

Uh oh!

Conversation

rudy2steiner commented Aug 9, 2026

Background

What's Changed

Uh oh!

codecov Bot commented Aug 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

zouyx left a comment

Choose a reason for hiding this comment

Uh oh!

rudy2steiner commented Aug 10, 2026

Uh oh!

AgentScopeJavaBot left a comment

Choose a reason for hiding this comment

Uh oh!

AgentScopeJavaBot Aug 12, 2026

Choose a reason for hiding this comment

Uh oh!

AgentScopeJavaBot Aug 12, 2026

Choose a reason for hiding this comment

Uh oh!

AgentScopeJavaBot Aug 12, 2026

Choose a reason for hiding this comment

Uh oh!

AgentScopeJavaBot left a comment

Choose a reason for hiding this comment

Uh oh!

AgentScopeJavaBot Aug 12, 2026

Choose a reason for hiding this comment

Uh oh!

AgentScopeJavaBot Aug 12, 2026

Choose a reason for hiding this comment

Uh oh!

AgentScopeJavaBot Aug 12, 2026

Choose a reason for hiding this comment

Uh oh!

rudy2steiner commented Aug 12, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

codecov Bot commented Aug 9, 2026 •

edited

Loading