Client graph verbs: add_input and add_data_join (#30) - #31
Merged
Conversation
The runtime had both; the service client had neither, which is what blocks the servicified demo. Two new graph verbs, passthrough on the session side: `add_input` carries data only -- dtype, channel, codec -- because the producer lives outside the framework, and `add_data_join` carries the joined dtype. Unit tests cover the binding, the runtime's channel refusal surfacing as 409, and join registration. `add_barrier` / `add_data_split_task` from #30 stay open: no consumer exists yet, and the barrier verb carries a wire-design decision (instances hold asyncio primitives and must not cross; a declarative spec should). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016Npyz3Hbnwos12ESsdJ2YU
`ChannelPublisher.open()` defaults to `timeout=None` -- the streaming contract's "wait forever" -- but the orbit backend forwarded that None into ORBIT's `start(timeout: float)`, which dies on `monotonic() + None`. Every external producer on the orbit data plane with the default timeout hit this. None is now approximated by a day. The new integration test runs the servicified demo shape end to end: two external channels bound with `add_input`, joined with `add_data_join`, consumed by a `JoinSink` -- producers are plain `ChannelPublisher`s using exactly that default. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016Npyz3Hbnwos12ESsdJ2YU
Three review findings applied: - `runtime.add_input` returns its subscription task and the service verb awaits it, so the client's success means the binding is live -- a producer publishing right after the call no longer races the broker registration, and a subscribe failure reaches the caller instead of only a server-side log. - a re-bind that changes the codec is refused (ValueError -> 409): the stream client dedupes on (channel, dtype), so the change would be recorded in `describe()` yet never applied. - `ChannelPublisher.open` defaults to CLIENT_CONNECT_TIMEOUT like every other orbit connect path -- an external producer pointed at an unreachable broker fails in 30s instead of holding an uncancellable thread for the day-long None approximation, which stays as the backstop for an explicit None (and its timeout message now names the effective deadline, not "None"). Plus: the client docstring states the stream's real delivery contract, the join test asserts against `join_components` directly, and the None substitution is pinned at unit level. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016Npyz3Hbnwos12ESsdJ2YU
- the add_input unit test drops its settling sleep -- the verb answers only once the subscription is live, and the test now pins exactly that (with the sleep, a regression back to fire-and-forget would still pass) - local imports hoisted to the module blocks they belong in - the integration helper loses a parameter nothing passed - `add_input` carries its new return contract in the signature - a dead fake attribute removed Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016Npyz3Hbnwos12ESsdJ2YU
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The two verbs the servicified demo actually calls (
run_me_service.pyin radical-collaboration/amsc#5):add_inputbinds an external channel to an input dtype (data-only wire: dtype, channel, codec — the producer lives outside the framework),add_data_joinregisters a join. Session side is passthrough to the runtime, which already validates and subscribes at bind time.Second commit fixes a crash this surfaced:
ChannelPublisher.open()defaults totimeout=None("wait forever"), which the orbit stream backend forwarded into ORBIT'sstart(timeout: float)—monotonic() + NoneTypeError. Every external producer on the orbit data plane with the default timeout hit it; None is now approximated by a day. The demo's sensors use exactly that default.Tests: 3 unit (binding lands, channel refusal surfaces as 409, join registers) + 1 integration running the demo shape end to end (two
ChannelPublishers →add_input×2 →add_data_join→JoinSink, complete pairs in order). Local: 175 unit + 30 integration passed.Out of scope, staying open on #30:
add_barrier/add_data_split_task— no consumer yet, and the barrier verb needs a wire decision (aBarrierinstance holds asyncio primitives and must not cross; a declarative spec should).🤖 Generated with Claude Code
https://claude.ai/code/session_016Npyz3Hbnwos12ESsdJ2YU