Fix: more resilient conversation start

[tofarr]

Summary

This PR makes the conversation start process more resilient by improving how we handle event service lifecycle and conversation reuse.

Changes Made

1. Safer Conversation Reuse Logic

• Before: Only checked if conversation_id exists in _event_services dictionary
• After: Also verifies that the existing event service is actually open/active using the new is_open() method
• Why: Prevents attempting to reuse event services that may be in an inconsistent or closed state

2. Delayed Event Service Storage

• Before: Stored event service reference in dictionary before calling start()
• After: Only stores the reference AFTER successful startup
• Why: Prevents storing references to event services that failed to start, avoiding potential memory leaks and inconsistent state

3. New EventService.is_open() Method

• Added is_open() method that returns bool(self._conversation)
• Provides a reliable way to check if an event service is in an active, usable state

Files Modified

• openhands-agent-server/openhands/agent_server/conversation_service.py
  • Updated conversation reuse logic to check is_open() status
  • Moved event service storage to after successful startup
• openhands-agent-server/openhands/agent_server/event_service.py
  • Added is_open() method for state checking

Benefits

• Improved Reliability: Reduces race conditions during conversation startup
• Better Resource Management: Prevents storing references to failed event services
• Cleaner State Management: More explicit checks for service availability
• Backward Compatible: No breaking changes to existing APIs

Testing

These changes address edge cases in conversation lifecycle management and should be particularly beneficial in high-concurrency scenarios or when dealing with network/resource constraints during startup.

---

Agent Server images for this PR

• GHCR package: https://github.com/OpenHands/agent-sdk/pkgs/container/agent-server

Variants & Base Images

Variant	Architectures	Base Image	Docs / Tags
java	amd64, arm64	eclipse-temurin:17-jdk	Link
python	amd64, arm64	nikolaik/python-nodejs:python3.12-nodejs22	Link
golang	amd64, arm64	golang:1.21-bookworm	Link

Pull (multi-arch manifest)

# Each variant is a multi-arch manifest supporting both amd64 and arm64
docker pull ghcr.io/openhands/agent-server:24a7d50-python

Run

docker run -it --rm \
  -p 8000:8000 \
  --name agent-server-24a7d50-python \
  ghcr.io/openhands/agent-server:24a7d50-python

All tags pushed for this build

ghcr.io/openhands/agent-server:24a7d50-golang-amd64
ghcr.io/openhands/agent-server:24a7d50-golang_tag_1.21-bookworm-amd64
ghcr.io/openhands/agent-server:24a7d50-golang-arm64
ghcr.io/openhands/agent-server:24a7d50-golang_tag_1.21-bookworm-arm64
ghcr.io/openhands/agent-server:24a7d50-java-amd64
ghcr.io/openhands/agent-server:24a7d50-eclipse-temurin_tag_17-jdk-amd64
ghcr.io/openhands/agent-server:24a7d50-java-arm64
ghcr.io/openhands/agent-server:24a7d50-eclipse-temurin_tag_17-jdk-arm64
ghcr.io/openhands/agent-server:24a7d50-python-amd64
ghcr.io/openhands/agent-server:24a7d50-nikolaik_s_python-nodejs_tag_python3.12-nodejs22-amd64
ghcr.io/openhands/agent-server:24a7d50-python-arm64
ghcr.io/openhands/agent-server:24a7d50-nikolaik_s_python-nodejs_tag_python3.12-nodejs22-arm64
ghcr.io/openhands/agent-server:24a7d50-golang
ghcr.io/openhands/agent-server:24a7d50-java
ghcr.io/openhands/agent-server:24a7d50-python

About Multi-Architecture Support

• Each variant tag (e.g., 24a7d50-python) is a multi-arch manifest supporting both amd64 and arm64
• Docker automatically pulls the correct architecture for your platform
• Individual architecture tags (e.g., 24a7d50-python-amd64) are also available if needed

[github-actions]

Coverage Report •

File	Stmts	Miss	Cover	Missing
openhands-agent-server/openhands/agent_server
conversation_service.py	292	177	39%	63, 66, 77–78, 81–84, 86, 90, 92, 95–102, 105–106, 109–113, 116–118, 120–123, 125, 132–133, 135–137, 140, 144, 146, 148, 155, 161, 169, 184, 189–190, 193, 199, 202, 213–217, 219–222, 225–230, 233–236, 238–240, 243–245, 250–253, 261, 266–268, 282–286, 289, 291, 294–296, 298, 302, 306, 313–317, 320–321, 327–332, 338–340, 358, 382, 407, 409–410, 436, 438, 440–443, 448, 450–451, 455–456, 458–459, 462–464, 467, 473, 478–481, 488–489, 493–497, 499, 504, 508–510, 514–515, 517–519, 521, 523, 536–538, 541, 544, 547–550, 557–558, 562–564, 567–568, 570
event_service.py	203	84	58%	52–53, 72–74, 77–82, 94, 110, 117, 119, 126–127, 135–138, 145–147, 159–160, 163–164, 166–168, 170, 175, 178, 182–184, 186, 188, 192, 195, 199, 203–204, 206, 223–224, 271, 278–279, 281, 284–286, 288, 292–295, 299–302, 310–313, 332–333, 335–342, 344–345, 351, 357, 370–371, 378
TOTAL	12163	5573	54%

[hieptl]

Thank you! 🙏