Set initial execution status to error if it was running

[tofarr]

Summary

When a conversation is deserialized if the execution_status is running, we set the execution_status to error - because this means that the conversation stopped while executing some action - this most commonly means some sort of crash in a process started by the agent.

The Web Frontend does not pick this up yet, but the execution_status does appear as error, and the runtime_status appears as STATUS$ERROR

Testing

• Start an agent server instance with

Checklist

• If the PR is changing/adding functionality, are there tests to reflect this?
• If there is an example, have you run the example to make sure that it works?
• If there are instructions on how to run the code, have you followed the instructions and made sure that it works?
• If the feature is significant enough to require documentation, is there a PR open on the OpenHands/docs repository with the same branch name?
• Is the github CI passing?

---

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:bbc3277-python

Run

docker run -it --rm \
  -p 8000:8000 \
  --name agent-server-bbc3277-python \
  ghcr.io/openhands/agent-server:bbc3277-python

All tags pushed for this build

ghcr.io/openhands/agent-server:bbc3277-golang-amd64
ghcr.io/openhands/agent-server:bbc3277-golang_tag_1.21-bookworm-amd64
ghcr.io/openhands/agent-server:bbc3277-golang-arm64
ghcr.io/openhands/agent-server:bbc3277-golang_tag_1.21-bookworm-arm64
ghcr.io/openhands/agent-server:bbc3277-java-amd64
ghcr.io/openhands/agent-server:bbc3277-eclipse-temurin_tag_17-jdk-amd64
ghcr.io/openhands/agent-server:bbc3277-java-arm64
ghcr.io/openhands/agent-server:bbc3277-eclipse-temurin_tag_17-jdk-arm64
ghcr.io/openhands/agent-server:bbc3277-python-amd64
ghcr.io/openhands/agent-server:bbc3277-nikolaik_s_python-nodejs_tag_python3.12-nodejs22-amd64
ghcr.io/openhands/agent-server:bbc3277-python-arm64
ghcr.io/openhands/agent-server:bbc3277-nikolaik_s_python-nodejs_tag_python3.12-nodejs22-arm64
ghcr.io/openhands/agent-server:bbc3277-golang
ghcr.io/openhands/agent-server:bbc3277-java
ghcr.io/openhands/agent-server:bbc3277-python

About Multi-Architecture Support

• Each variant tag (e.g., bbc3277-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., bbc3277-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	336	209	37%	64, 67, 78–79, 82–85, 87, 91, 93, 96–103, 106–107, 110–114, 117–119, 121–124, 126, 133–134, 136–138, 141, 145, 147, 149, 156, 162, 170–171, 180–183, 192, 201, 206–207, 210, 223–224, 242, 245, 256–260, 262–265, 268–273, 276–279, 281–283, 286, 289–291, 296–299, 307, 312–314, 328–332, 335, 337, 340–342, 344, 348, 352, 359–363, 366–367, 371–375, 378–379, 383–387, 390–391, 397–402, 409–410, 414, 416–417, 422–423, 429–430, 436–438, 456, 480, 508, 510–511, 537, 539, 541–544, 549, 551–552, 556–557, 559–560, 563–565, 568, 574, 579–582, 589–590, 594–598, 600, 605, 609–611, 615–616, 618–620, 622, 624, 637–639, 642, 645, 648–651, 658–659, 663–665, 668–669, 671
event_service.py	314	158	49%	55–56, 75–77, 81–86, 89–92, 107, 123, 127, 131–132, 139, 141, 148–149, 157–160, 167–169, 186, 210–211, 214–215, 217–219, 221, 226, 229–230, 233–235, 238, 242–244, 246, 248, 259–262, 275–276, 279–280, 283, 286–288, 291–292, 295–296, 300, 303, 307, 311–312, 314, 331–332, 349, 351, 355–357, 361, 370–371, 373, 377, 383, 385, 393–398, 447, 449–452, 461, 477, 484, 488, 499–500, 510–513, 515–516, 520, 522, 526–529, 534–536, 538, 542–545, 549–552, 560–563, 582–583, 585–592, 594–595, 604–605, 607–608, 615–616, 618–619, 623, 629, 639–640, 647
openhands-sdk/openhands/sdk/llm
llm.py	420	157	62%	359, 364, 368, 372–373, 376, 380–381, 392–393, 395–396, 400, 417, 435–438, 485, 515–517, 538, 542, 557, 563–564, 588–589, 599, 624–629, 650–651, 654, 658, 670, 675–678, 687, 695–702, 706–709, 711, 724, 728–729, 731–732, 737–738, 740, 747, 750–755, 812–817, 874–875, 878–881, 923, 940, 994, 997, 1000–1008, 1012–1014, 1017, 1020–1022, 1029–1030, 1039, 1046–1048, 1052, 1054–1059, 1061–1078, 1081–1085, 1087–1088, 1094–1103, 1116, 1130, 1135
TOTAL	14512	6871	52%

[hieptl]

Thank you! 🙏

[enyst]

Are we sure "it cannot possibly be RUNNING"? I thought we have an autosave, though I could be wrong.

So I understand we may encounter a problem deserializing from RUNNING, but I'm not sure what the best solution is, could we maybe save it as something else, or avoid to save when running. Or set it as IDLE perhaps? The latter makes more sense to me, unless something prevents that.

[openhands-ai]

Looks like there are a few issues preventing this PR from being merged!

• GitHub Actions are failing:
  • Run tests

If you'd like me to help, just leave a comment, like

@OpenHands please fix the failing actions on PR #1554 at branch `fix-restarted-conversation`

Feel free to include any additional details that might help me get this PR into a better state.

_{^{You can manage your notification settings}}

[enyst]

Out of curiosity, is there a reason why this isn't suitable?

software-agent-sdk/openhands-sdk/openhands/sdk/agent/agent.py

Line 139 in 7b623da

pending_actions = ConversationState.get_unmatched_actions(state.events)

[tofarr]

Workflow is thus:

1. Agent executes some tool call which leaks memory
2. Agent server pod is evicted from K8 as a result of this.
3. Pod restarts - execution_status was running but is now error.
4. User can prompt the agent to run again - but it will run the last Action which does not have an observation - resulting in a repeat of step 1.

After change
4. The action which crashed the pod now has an AgetnErrorObservation, letting the agent know not to run the same action again (Unless prompted with something like "Please try that again!")

[enyst]

Got it, thank you. We had this kind of reset in agent controller back in V0. Indeed I think AgentErrorEvent is correct... 🤔

The other question is, is this really server-specific? If the state is RUNNING and it's auto-saved, which I think it is, is there anything preventing it from happening on some who-knows-what stuck process on LocalConversation?