I'm still seeing my canvas lock up for minutes at ...
# proj-agent-canvas
w
I'm still seeing my canvas lock up for minutes at a time sadblob Not sure if it's long conversations or what
l
I think I just merged a fix this morning
πŸ‘€ 1
I’ll work on resolving the review comments, but you can try running the canvas with that SDK version and see if it solves the problem, I haven’t seen a lockup since I transitioned to using that branch
m
@modern-fireman-99755 pointed me towards this as a fix: github.com/OpenHands/OpenHands/pull/16638 But I tried that locally and it did not fix the issue.
l
I think mine is probably a more fundamental fix
πŸ™Œ 1
w
working much better!
l
I actually found another lock πŸ˜‚
πŸ‘ 1
m
@limited-student-10747 difference is night and day... Appreciate you possibly weren't suffering from it but it's about 5x more responsive and 200x less annoying!!!
plus1 1
l
Awesome, really glad it’s working better! I was actually suffering from it whenever I ran many parallel sessions πŸ™‚ It just took me a long time to figure out what the actual problems were. It’s easy to come up with a bandaid fix but I didn’t want to do that.
❀️ 1
w
Can we get a release with this fix out today?
l
@fresh-machine-37269 prepare a chain of releases of agent-canvas and all of the repos that it depends on in org:OpenHands. Follow the standard practice in each repo, and if the PR is already open and waiting for review then just tell me the location.
πŸ‘€ 1
f
πŸ€– On it! View progress here: oss-agent-canvas.ngrok.dev/conversations/…
⏳ Still working β€” this is taking longer than this run allows. Follow along at the link above; mention me again for an update.
Done β€” the Agent Canvas release chain is prepped. βœ… β€’ software-agent-sdk: the lockup fixes were already merged, so I dispatched the missing release β€” PR now open: Release v1.44.0 β€’ typescript-client: already open β€” release 1.38.2 β€’ extensions: already open β€” release 0.19.0 β€’ agent-canvas (OpenHands/OpenHands): already open β€” release 1.16.0 Merge order: SDK PR first (auto-publishes PyPI and opens the downstream bumps), then the dependency bumps, and agent-canvas last.
l
@fresh-machine-37269 check why on software-agent-sdk one example is failing, and if you are able to diagnose, push a fix and rerun the examples.
πŸ‘€ 1
f
πŸ€– On it! View progress here: oss-agent-canvas.ngrok.dev/conversations/…
⏳ Still working β€” this is taking longer than this run allows. Follow along at the link above; mention me again for an update.
m
@fresh-machine-37269 give an update
πŸ‘€ 1
f
βœ… Done! Here's the current status: PR #4630 β€” _fix(sdk): drop in-flight tool calls from ask_agent context_ β€’ State: open, mergeable,
test-examples
label applied. β€’ Review: required (no approvals yet). Examples rerun is still in progress (started 16:24 UTC): github.com/OpenHands/software-agent-sdk/…/32871786638 The full examples suite takes roughly 20 minutes, so no final result yet β€” the run hasn't concluded. I'll report back once it finishes (or you can watch the run link directly).
l
@fresh-machine-37269 is the example still failing? If so, create an issue describing what you know of the root cause.
f
πŸ€– On it! View progress here: oss-agent-canvas.ngrok.dev/conversations/…
βœ… Done! Yes, I'm here. What can I help you with?
m
@fresh-machine-37269 Answer the question of Graham thanks. Which is: is the example still failing? If so, create an issue describing what you know of the root cause.
πŸ‘€ 1
@limited-student-10747 I don't know why the bot didn't answer to you. I was working on the VM today and I changed a couple of things. I think you triggered the bot in the exact moment it was offline... I'm double checking
f
βœ… Done! The example is no longer failing β€” the
run-examples
workflow is green on the fix branch. But since the fix isn't merged yet (and the release branch still fails), I've documented the diagnosis in a new issue. Diagnosis: The failing example is
examples/01_standalone_sdk/28_ask_agent_example.py
. It fails because it calls
ask_agent()
mid-run on a background thread while the main agent is still executing a tool call.
Conversation.ask_agent()
builds its LLM context from the cached, incrementally-maintained
self.state.view
, which skips
enforce_properties
on linear appends. That view can transiently hold an
ActionEvent
(tool call) whose matching
ObservationEvent
(tool result) hasn't landed yet. Anthropic rejects that orphaned
tool_use
with a
400 invalid_request_error
("tool_use ids were found without tool_result blocks"), surfaced as
LLMMalformedConversationHistoryError
. Fix status: already open as PR #4630 (
fix(sdk): drop in-flight tool calls from ask_agent context
) β€” the example passes on that branch (run 32871786638), and the failure only reproduces on the release branch
rel-1.44.0
(run 32863616276) without the fix. Issue created: github.com/OpenHands/software-agent-sdk/issues/4653
l
@fresh-machine-37269 I’d like to continue this release process, what’s left to do
πŸ‘€ 1
f
πŸ€– On it! View progress here: oss-agent-canvas.ngrok.dev/conversations/…
βœ… Done! Here's where the release chain stands, in dependency order: Already released (upstream of agent-canvas): β€’
software-agent-sdk
β†’ v1.44.0 βœ… β€’
automation
β†’ 1.9.0 βœ… β€’
extensions
β†’ v0.19.0 βœ… Still left to do: 1.
typescript-client
(OpenHands/typescript-sdk)
β€” release PR #353 (
chore(main): release 1.39.0
) is open, labeled
autorelease: pending
, mergeable. One caveat: it's pinned to agent-server v1.43.1, and there's a separate open PR #361 bumping it to v1.44.0 β€” you likely want #361 merged first so 1.39.0 ships against the new server. 2.
OpenHands/OpenHands
(agent-canvas)
β€” two things, in order: β—¦ Merge the deps bump PR #16968 (
chore: bump SDK deps β€” software-agent-sdk 1.44.0, automation 1.9.0, extensions 0.19.0
). The
mock-llm-e2e
/
docker-e2e
jobs are red with exit 124 (timeout) β€” looks flaky rather than a real failure, but worth a re-run before merging. β—¦ Then the release PR #16785 (
chore(main): release 1.16.0
) is open and mergeable β€” but note it still pins the old versions (agent-server 1.42.1, automation 1.8.0, extensions 0.18.0), so it needs #16968 in first (or a manual rebase) to actually ship the lock-fix SDK. One loose end worth flagging: the SDK example issue is still open β€” #4653 (
28_ask_agent_example.py
Anthropic 400), and its fix #4630 is still unmerged. That landed after v1.44.0 was cut, so it isn't in this release β€” it's a follow-up for the next SDK release, not a blocker. Want me to go ahead and mark any of these ready/merge them, or re-run the failing mock-llm checks first?
l
@fresh-machine-37269 give me all the PR links
πŸ‘€ 1
f
βœ… Done! Here are all the PR links, in the order they need to be handled: Open β€” need action: 1.
typescript-client
server bump β†’ github.com/OpenHands/typescript-client/pull/361 2.
typescript-client
release (1.39.0) β†’ github.com/OpenHands/typescript-client/pull/353 3.
OpenHands
SDK deps bump β†’ github.com/OpenHands/OpenHands/pull/16968 4.
OpenHands
agent-canvas release (1.16.0) β†’ github.com/OpenHands/OpenHands/pull/16785 Open β€” follow-up (not blocking this release): 5.
software-agent-sdk
ask_agent example fix β†’ github.com/OpenHands/software-agent-sdk/pull/4630 Already merged (the lock fixes): 6. github.com/OpenHands/software-agent-sdk/pull/4570 7. github.com/OpenHands/software-agent-sdk/pull/4602
l
@fresh-machine-37269 OK, I merged typescript-client release, is it bumped in OpenHands#16968 and if not do so.
πŸ‘€ 1
f
βœ… Done! No β€” it wasn't. #16968 still pinned
@openhands/typescript-client
to 1.38.1 (only extensions/agent-server/automation had been bumped). I've now bumped it to 1.39.0 in both
package.json
and
package-lock.json
and pushed to the branch. Updated PR: github.com/OpenHands/OpenHands/pull/16968 The diff now shows
@openhands/typescript-client
1.38.1 β†’ 1.39.0, with the lockfile integrity hash matching the published 1.39.0 tarball.