[Bug]: 0.3.3.Final - Agent terminates early when ADK tool returning RxJava Single hops threads

Open
#872 7 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

@malladinagarjuna2 is already working on this.

Since Aug 9, 2026.

  • #1027 by @malladinagarjuna2 — open

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
45/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Quiet
Tech stack
kotlin

Research direction

The bug is in the cleanupProducer method of DefaultRequestHandler where the ChildQueue closes early when an RxJava Single hops threads. Look at the agent endpoint handling and how it waits for asynchronous tool results. Start by reproducing with a Kotlin tool that uses rxSingle on a different dispatcher. Check the 1.0.0 fix with AgentEmitter to understand the resolution. The fix likely involves proper synchronization or waiting for the Single's completion on the new thread.

Written by the indexing model from the issue text.

Description

What happened?

When an ADK tool returns an RxJava Single executes on a different thread to the executor, the agent endpoint returns right after the function call is made. I noticed this bug using Kotlin, where using rxSingle executes on the global dispatcher, rather than the executor, e.g:

@Schema(description = "Get the weather summary in a city")
fun getWeather(@Schema(name = "city", description = "City to query") city: String): Single<Map<String, Any>> {
    return rxSingle {
        mapOf<String, Any>("result" to "It's 50 degrees and partly cloudy in $city")
    }
}

When running an agent with this tool, e.g. with message/send, the result cuts out after the tool call.

{
  "jsonrpc": "2.0",
  "id": "cli-check-2",
  "result": {
    "id": "565c05e1-8bfc-414c-950e-80d74e14350c",
    "contextId": "cli-demo-context",
    "status": {
      "state": "working",
      "timestamp": "2026-05-12T15:09:36.256177Z"
    },
    "artifacts": [
      {
        "artifactId": "80c03215-3225-4108-9f7f-040d5ad2ec82",
        "parts": [
          {
            "data": {
              "name": "getWeather",
              "id": "adk-183771c0-58a0-4f1b-87e9-d406ffaef646",
              "args": {
                "city": "Edinburgh"
              }
            },
            "metadata": {
              "adk_type": "function_call"
            },
            "kind": "data"
          }
        ],
        "metadata": {}
      }
    ],
    "history": [
      {
        "role": "user",
        "parts": [
          {
            "text": "What is the weather in Edinburgh?",
            "kind": "text"
          }
        ],
        "messageId": "cli-check-2",
        "contextId": "cli-demo-context",
        "taskId": "565c05e1-8bfc-414c-950e-80d74e14350c",
        "kind": "message"
      }
    ],
    "kind": "task"
  }
}

Using rxSingle(Dispatchers.Unconfined) works only if the function never actually suspends, otherwise the issue persists. A Java equivalent would be to e.g. call .subscribeOn(Schedulers.io()) on the Single, or any other change of Scheduler.

The issue appears to be in the cleanupProducer method of the DefaultRequestHandler, where the ChildQueue is closed prematurely. It does appear to be fixed in 1.0.0 with AgentEmitter - but I'm unsure if we'll be able to adopt it yet as the ADK doesn't support it (I've raised an issue for it here). Is there any chance a fix could be backported to 0.3? I'm happy to provide help if needed.

Code of Conduct
  • I agree to follow this project's Code of Conduct
Dominant language
Java
Stars
514
Forks
200
Avg merge
1d 2h
Merged PRs (30d)
53

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from a2aproject/a2a-java

All issues in a2aproject/a2a-java

Similar issues

More Java issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.