Repository navigation
row 155 F25 row 728: the completion tap carries no response_tokens_out, as in 1.5.5 (stacked on #912) - #919
Merged
Conversation
…t, as 1.5.5's never did The coordinator granted this lane the fix: "response_tap must carry the response's token signals ... as 1.5.5 does; cite the 1.5.5 line". Reading 1.5.5 shows it never computed the signal. Its stage taps projected no catalog signal (v1.5.5 crates/busbar/src/proxy/hooks.rs:1348-1352, `signals: Default::default()`, "the wire is byte-identical"), and ResponseTokensOut was reserved with no compute fn (crates/api/src/signal.rs:90-95). The driver's empty SignalBag in response_tap (busbar-kernel plane_driver/hooks/mod.rs) is therefore 1.5.5's wire, and the legacy crate's 1 output token on the tap was post-1.5.5 drift. #912's port of the legacy figure would go red against the 1.5.5 behaviour; this pins 1.5.5 instead. No kernel change.
MattJackson
marked this pull request as draft
October 14, 2026 03:36
MattJackson
added a commit
that referenced
this pull request
Oct 15, 2026
…redev, by cause - cap_perf_streams_2000 (#1039 made the llm door the default path): every piece of an admitted unit took an instance max_inflight unit on worker 0, so 2000 stream endings overflowed the door's cap and the refused pieces cut answered streams. plugin-loader: Dispatcher::submit_unit (an admitted unit's on_piece/serve hold no max_inflight unit, as driver-ticket ops do; spec A.4.1, one ticket per unit reused for its ops); PlaneInstance spreads unit tickets over the dispatcher's workers (one slab per instance and worker); a session's two legs mint on one worker (PlaneCalls::mint_beside). - #958 (bedrock over-budget = 429 ThrottlingException, accepted difference C-22): exchange_refuse x2, on_driver refusals, plane_node bedrock fixture, pools_door spent-budget x2 and governance. - llm door projection reads bedrock inferenceConfig.maxTokens where the top-level cap is absent (#912's on_driver pin, now that the door serves by default, #824). - on_driver seam: a hook veto ends the route leg Failed(Route, HookVeto) as the kernel pins; the completion tap carries no response_tokens_out, as 1.5.5 (#919's test change only). - kernel config_validate: the unknown-protocol list names each dialect once (cli_validate x2). - root model_egress: the far end's RecordNothing is the caller's fault, counted in observe (the one call it records through). - kernel secret_refs: SpawnEnv inventoried (#865); plugin-loader: the forged-ticket latch test reaches its 16th forgery, the dispatch fixture's stale wakes use the tickets it was handed (#1018). - xtask: posture case reads construction figures (fixfwd-2 emptied structure-lint's); reachability selftest's name-only case uses busbar-plane-llm; segregation SRC_FLOOR follows xtask/src's growth (a floor, stricter); conformance_rigs counts 18 suites (#885). - busbar-plane-llm dialect.rs: every dialect states its `protocol` param (the auth-header plugin's no-header line reads it), with the fixture's families/own/passthrough.
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.
Row 155, F25 row 728. Stacked on #912, which is in a train batch, so this follow-up does not push into it.
Finding. 1.5.5 never computed
response_tokens_out:crates/busbar/src/proxy/hooks.rs:1348-1352: stage taps project no catalog signal (signals: Default::default(), "the wire is byte-identical").crates/api/src/signal.rs:90-95: ResponseTokensOut is reserved, with no compute fn.So the driver's empty SignalBag in
response_tapis already 1.5.5's wire. The legacy crate's figure (1 output token on the tap) was post-1.5.5 drift, and #912's port of it would go red against the 1.5.5 behaviour.Change. The test now pins 1.5.5: no
response_tokens_outkey on the completion tap, whether the signal is declared or not. The kernel is unchanged.RED:
on_driver/seam.rs::completion_tap_carries_no_response_tokens_out_as_in_1_5_5. Not run locally.