I just see a conversation lost history: cbba53c74d...
# openhands-cloud
c
I just see a conversation lost history: cbba53c74deb4598a433717f6e02bcee Its shell was stuck (GPT-5 -euo issue) and I wanted to stop and restart the runtime hoping to get rid of it. I did. After it restarted, there aren't any messages in history.
👀 1
w
hmm...nothing weird looking on our end
runtime has been alive for 20m with no apparent issues
c
how many events does it have?
w
~15
I wanted to stop and restart the runtime
Just noticed this--was the restart 20m ago?
c
Hmm, I want to say much less…
w
interesting...the current iteration says its been up 23m
actually, something is weird here
c
Sonnet-4 reports via API: Perfect! I can see the latest event has ID 115. Since event IDs start at 0, this means there are approximately 116 events in the conversation (115 + 1). 📊 Event Count for Conversation cbba53c74deb4598a433717f6e02bcee Total Events: ~116 Looking at the latest event, I can see the agent is actively working on the FastMCP issue: Event ID: 115 (latest) Type: IPython execution observation Timestamp: 2025-08-21T183546.160070 Action: The agent just searched for the failing test names in the codebase Result: Found all three failing tests in /workspace/project/OpenHands/tests/runtime/test_mcp_action.py
w
because the deployment is only 23m old, but your convo is 1h old
something did kill your whole deployment (including conversation state)
c
But I can apparently still retrieve its events, while the web FE doesn’t
Right
Chatting with Sonnet-4 about the old event stream here:
how did you retrieve this info, did you use the session api key?
No, I didn't use the session API key! I used the main app API endpoint with just the regular OpenHands API key.
### Method 1: Main App API (my first query) • Event Count: ~116 events (latest ID: 115) • Latest Event: IPython execution searching for test files • Status: Agent actively working ### Method 2: Runtime API (with session key) • Event Count: ~16 events (latest ID: 15) ⚠️ *Much fewer! • Latest Event: Agent state changed to "awaiting_user_input" • Status: Agent waiting for input • • ### *Key Observations: • • Runtime URL Changed: • Original:
hbbnxplxbmzjique.prod-runtime.all-hands.dev
• Current:
pruyxxqxncpsrssu.prod-runtime.all-hands.dev
• • Session API Key: Same**
I thought the behavior should be that state doesn’t matter for retrieval of the events. At least that’s how it was, I’m not sure if this broke since nested thonk
w
nod something got way out of sync here