Repository navigation
docs(quest): record four findings from the m1 quest wave #3424
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Changes from all commits
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,47 @@ | ||
| # [M] moq-net: resume resolves track info from segment zero, not the newest | ||
|
|
||
| ## Goal | ||
|
|
||
| `resume::Consumer::poll_info` reports the track info of the generation whose | ||
| frames a reader will actually receive. A track republished on a live path | ||
| serves the successor's `priority`, `maxAge`, and `timescale`, not the | ||
| predecessor's. | ||
|
|
||
| Boundaries: this is the Rust half of the same defect `js/net` fixed by keying | ||
| its `TRACK_INFO` cache on the routing front. The wire format does not change, | ||
| so no draft moves. | ||
|
|
||
| ## Plan | ||
|
|
||
| `resume::Consumer::poll_info` (`rs/moq-net/src/model/resume.rs`) resolves info | ||
| from **segment zero**, while every data read routes to the **newest** segment. | ||
| When a broadcast is replaced on a path, those are different generations, so a | ||
| subscriber can be handed the predecessor's metadata alongside the successor's | ||
| frames. | ||
|
|
||
| `timescale` is the damaging field: the reader converts frame timestamps with | ||
| it, so a mismatched generation silently rescales every timestamp rather than | ||
| failing. `priority` and `maxAge` are wrong but survivable. | ||
|
|
||
| `origin.rs`'s `run_front` already re-resolves the successor's info on | ||
| `Step::Splice` and then discards it, which is the natural place for the fix to | ||
| hang off. | ||
|
|
||
| ### How it was found | ||
|
|
||
| Isolated while fixing the `js/net` half. Against a real relay, a publisher | ||
| wrote 1234 ticks on a successor's MICRO grid and the subscriber received | ||
| `1000us`: the JS fix alone was not enough, because the relay's own stale | ||
| metadata rescaled the frames. With the Rust side still wrong, the same | ||
| mismatch is reachable from any client. | ||
|
|
||
| ### Not covered by an existing tracker | ||
|
|
||
| [#2991](/quest/m1/2991-net-coalesce-dynamic-tracks-and-preserve-sequences-across.md) | ||
| is about sequence continuity across replacement, not info resolution. #2610's | ||
| epoch remedy was removed from the draft by #3225, so there is no wire-level | ||
| generation marker to lean on; the fix is local. | ||
|
|
||
| ## Related | ||
|
|
||
| - [#2991](/quest/m1/2991-net-coalesce-dynamic-tracks-and-preserve-sequences-across.md) - sequence continuity across the same replacement | ||
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,37 @@ | ||
| # [M] The interop matrix has no Go client | ||
|
|
||
| ## Goal | ||
|
|
||
| `test/smoke` publishes from and subscribes with Go, so a regression in | ||
| `go/wrapper` fails CI instead of shipping. | ||
|
|
||
| Boundaries: the Go wrapper only, through the same harness the other bindings | ||
| use. No new transport or feature coverage; this closes a hole in the matrix | ||
| rather than widening it. | ||
|
|
||
| ## Plan | ||
|
|
||
| `just test smoke-full` runs 21 cells today: rust, python and js publishers | ||
| against rust, python, js, js-native-node, js-native-bun, c and gst | ||
| subscribers. Go appears in neither axis, so nothing in CI exercises the Go | ||
| wrapper end to end. Its only coverage is `go/wrapper`'s own tests, which run | ||
| against the library rather than across a session. | ||
|
|
||
| That gap is load-bearing: `moq-ffi` changes are required to run `smoke-full` | ||
| (Cross-Package Sync), and a passing run currently proves nothing about Go even | ||
| though `go/wrapper` is generated from the same core. | ||
|
|
||
| Add a Go publisher and a Go subscriber to `test/smoke/smoke.sh` alongside the | ||
| existing clients. `smoke.sh` resolves dependencies from committed lock data, | ||
| so the Go client must build with the repository's pinned toolchain and module | ||
| graph rather than resolving at run time. | ||
|
|
||
| ### How it was found | ||
|
|
||
| Adding `context.Context` cancellation to every blocking Go operation. The PR | ||
| ran `smoke-full` and it passed 21/21, which said nothing at all about the Go | ||
| code the PR changed. | ||
|
|
||
| ## Related | ||
|
|
||
| - [Native Go context](/quest/m1/go-native-context.md) - the Go wrapper work whose verification this gap limits |
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,29 @@ | ||
| # [S] moq-uring does not compile with --all-features | ||
|
|
||
| ## Goal | ||
|
|
||
| `cargo clippy -p moq-uring --all-features --all-targets` compiles, so the | ||
| nightly `just rs features` gate covers `moq-uring` instead of failing on it. | ||
|
|
||
| ## Plan | ||
|
|
||
| Enabling both `noq` and `quinn` puts two `quinn_proto` versions in scope, and | ||
| `rs/moq-uring/src/quic/quinn/connection.rs`'s test module refers to | ||
| `quinn_proto` unqualified, so the reference is ambiguous and the test target | ||
| does not build. `--all-features --lib` is clean; only `--all-targets` trips it. | ||
|
|
||
| Disambiguate the reference in the test module. Check whether the same | ||
| unqualified path appears anywhere else that only compiles under one of the two | ||
| features, since the ambiguity is invisible until both are on at once. | ||
|
|
||
| This is dev-only: `moq-uring` does not exist on main. | ||
|
|
||
| ### How it was found | ||
|
|
||
| Verified while adding qlog support to the io_uring workers: reproduced on a | ||
| stashed tree at the merge base, so it predates that work. It is unrelated to | ||
| qlog and was deliberately not fixed there. | ||
|
|
||
| ## Related | ||
|
|
||
| - [Worker metrics](/quest/m1/uring-metrics.md) - other moq-uring runtime work in the same crate |
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,35 @@ | ||
| # [M] Go bindings take context.Context natively | ||
|
|
||
| ## Goal | ||
|
|
||
| `uniffi-bindgen-go` emits `context.Context` as the first parameter of every | ||
| generated blocking call, so `go/wrapper` stops carrying a hand-rolled | ||
| cancellation token and the generated layer cancels the native task directly. | ||
|
|
||
| ## Plan | ||
|
|
||
| Today every affected `moq-ffi` method takes a trailing | ||
| `Option<Arc<MoqCancel>>` purely so the Go wrapper has something to cancel: | ||
| `uniffi-bindgen-go` renders a Rust `async fn` as a blocking Go call with no | ||
| cancellation handle at all. The token works and costs the other bindings | ||
| nothing (it is additive, with a UniFFI argument default), but it exists only | ||
| to work around the generator. | ||
|
|
||
| A fork of `kixelated/uniffi-bindgen-go` adding native `context.Context` | ||
| support already exists locally, on branch `codex/context-cancel-3188`. | ||
| Landing it means publishing that branch and pinning the new rev in | ||
| `flake.nix`, after which the token becomes redundant for Go and can be removed | ||
| from the `moq-ffi` surface. | ||
|
|
||
| Decide first whether the fork is worth carrying. Keeping the token is a | ||
| supported outcome: it is already shipped and costs nothing outside Go, and a | ||
| pinned generator fork is a maintenance obligation on every UniFFI bump. Nobody | ||
| should start this until that call is made. | ||
|
|
||
| ## Required | ||
|
|
||
| - The `context.Context` support in `kixelated/uniffi-bindgen-go` is published and pinned in `flake.nix` | ||
|
Comment on lines
+29
to
+31
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
This prerequisite combines publishing the fork with changing AGENTS.md reference: quest/AGENTS.md:L95-L99 Useful? React with 👍 / 👎. |
||
|
|
||
| ## Related | ||
|
|
||
| - [Go smoke client](/quest/m0/smoke-go-client.md) - what would actually verify a change to the Go bindings | ||
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
resume::Subscriber::poll_recv_groupiterates every retainedSegmentSub, not only the newest one, because the segments partition the sequence space across a takeover. Resolvingpoll_infofrom the successor would therefore make predecessor frames use the successor's timescale, while existing subscribers would retain the predecessor's already-sentTRACK_INFOfor successor frames. Define and enforce a stable-info invariant across splices, or introduce a real generation boundary, rather than basing the fix on newest-only reads.AGENTS.md reference: AGENTS.md:L148-L149
Useful? React with 👍 / 👎.