Repository navigation
feat(mux): report TS health and the export's release clock - #5849
Conversation
01560c5 to
3f2cb92
Compare
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info
WalkthroughThe change adds cumulative rendition counters and MPEG-TS and release metrics to the mux. The CLI adds Priority: ➖ Normal Merge Risk: ⚪ Minimal · up to
|
| Check name | Status | Explanation |
|---|---|---|
| Title check |  |
The title clearly summarizes the main changes: MPEG-TS health reporting and export release-clock reporting. |
| Docstring Coverage | |
Docstring coverage is 84.66% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 176 functions across 22 files. (1 skipped: … |
| Linked Issues check | |
Check skipped because no linked issues were found for this pull request. |
| Out of Scope Changes check | |
Check skipped because no linked issues were found for this pull request. |
| Description check | |
The pull request description directly explains the MPEG-TS health statistics, export release statistics, CLI behavior, serialization changes, documentation updates, tests, and migration impact covered… |
Full details: Docstring Coverage
-
Explanation
Docstring coverage is 84.66% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 176 functions across 22 files. (1 skipped: 1 unsupported.)
Full details: Description check
-
Explanation
The pull request description directly explains the MPEG-TS health statistics, export release statistics, CLI behavior, serialization changes, documentation updates, tests, and migration impact covered by the changeset.
✨ Finishing Touches: 💡 1
-
⚔️ Resolve merge conflicts: 💡
-
- Resolve merge conflict in branch
quest/m2/ts-health-stats
- Resolve merge conflict in branch
✨ Simplify code
-
- Create a new PR
-
- Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.
Comment @coderabbitai help to get the list of available commands.
`moq import --stats` names a stats track in the catalog the import builds and fills it with what each rendition has written: frames, payload bytes, and video keyframes, with the connection's transport. `moq export --echo <name>` publishes the feedback a catalog solicits, counting what the export reads and its newest arrival. `moq export stats` and `moq export echo` print either as JSON lines. moq-mux counts at the container producer and the source, so every container import and export shares the counters. Capture encoders, gateways, `archive`, and `ts --program all` refuse `--stats`; MSF catalogs refuse `--echo`. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
- Count a buffered frame once the container writes it, not on the way in: an abort discards the buffer. - Refuse `--echo` for an MSF catalog detected from a `.msf` broadcast, not only one named. - Refuse an echo name starting with `.`, which discovery hides. - `export echo` replaces a reader on a restart, lets one drain after its broadcast ends, and bounds the lines it queues for stdout. - A report track that refuses a write is closed with the cause and reported once, instead of logged at debug every second; a failed import aborts its stats track with the import's error. - The stats integration test joins `tests/integration`; the arrival test fakes the wall clock. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
A TS import's stats and a TS export's feedback now carry an `mpegts` section: each elementary stream's access units and quiet time, and the TR 101 290 counters. An export adds `release`, how its output clock keeps up with the source: frames dropped at the deadline, the least margin frames arrived with over the last second, the clock's offset and drift from the source, and measurements past the 30 ppm a system clock may run off. `ts::stats` gains serde and `Report`, the extension both snapshots flatten. `stats::Export` moves its clock fields under `release`, and `ts::Release` reports the same for a passthrough export. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
3f2cb92 to
9974c23
Compare
|
@codex review (Written by Claude Opus 5.5) |
|
Codex Review: Something went wrong. Try again later by commenting “@codex review”. ℹ️ About Codex in GitHubCodex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback". |
1 similar comment
|
Codex Review: Something went wrong. Try again later by commenting “@codex review”. ℹ️ About Codex in GitHubCodex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback". |
`--epoch` makes instances one publisher of identical tracks, and a stats track carries this instance's own counts and transport, so the pair is refused. Each `.echo` feedback broadcast now announces a fresh epoch like every other CLI publish, so a replaced viewer reads as a restart. The echo target is a struct instead of a tuple, and the CLI page drops the edge cases and unshipped fields the API docs and quest already carry. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # rs/moq-cli/src/main.rs # rs/moq-cli/src/publish.rs # rs/moq-cli/tests/integration/main.rs
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # rs/moq-cli/src/main.rs
spawn_import gained a shared-epoch argument on main, which pushed it past clippy's argument limit; the stats report now attaches to the returned Publish instead, still before any track is declared. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
--epoch is gone (moq-dev#5967), so its refusal of --stats goes with it; the stats report asks Demand::wanted (moq-dev#5930) whether anyone reads it. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Takes moq-dev#5848's merge of main. The decoder's stats handle is `report` there, so the MPEG-TS snapshot closure becomes `report_mpegts`; main's `unplaced` and `refused` counters are omitted at zero like the rest, and the stats doc names them; a declared break is not a gap (moq-dev#5815). Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8a0af79cdc
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| track = match &stats { | ||
| Some(name) => { | ||
| let subscriber = broadcast.track(name)?.subscribe(None).await?; | ||
| Some(moq_json::snapshot::Consumer::new(subscriber, Default::default())) |
There was a problem hiding this comment.
When either report sink reads a track from an untrusted publisher, this constructs moq_json::snapshot::Consumer with the default config, whose max_size is unset and explicitly leaves reconstructed state size unlimited; read_echo does the same at line 505. A publisher can send merge patches that continually add keys and make moq export stats or moq export echo retain unbounded JSON, so both consumers need a finite state budget.
AGENTS.md reference: AGENTS.md:L18-L18
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 7bc5705 (#5848, merged here): export stats and export echo both read a report through one reader capped at 1 MiB of reconstructed state (MAX_REPORT). A report is a few KiB; patches that grow one past the cap end the read with TooLarge. a_report_past_the_cap_ends_the_read fails without the cap.
(Written by Claude Opus 5.5)
| } else { | ||
| // Detached, not aborted: the final snapshot can still be in flight, and the reader | ||
| // ends with its track. | ||
| readers.remove(&path); |
There was a problem hiding this comment.
When a peer retracts a feedback route without closing its already-subscribed track, removing and dropping this JoinHandle detaches the read_echo task rather than cancelling it; the origin contract permits tracks already in flight to continue after route retraction. Repeated announce/retract cycles can therefore leave an unbounded number of hidden tasks and subscriptions, and a later start at the same path can interleave old and new snapshots, so ended readers must remain tracked under a bound or be cancelled.
AGENTS.md reference: AGENTS.md:L18-L18
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 7bc5705 (#5848, merged here): an ended reader now stays in the map, so a start at the same path aborts it before the new reader prints, and it is aborted 5 s after the End (DRAIN) even if its track stays open; finished readers are pruned. The drain keeps the final snapshot that the earlier review asked to let through. an_ended_feedback_reader_stops_after_draining retracts a route whose track stays open and fails with the reader detached.
(Written by Claude Opus 5.5)
| /// The counters for `name`, continuing those of an earlier track under that name. | ||
| pub(crate) fn track(&self, name: &str) -> Arc<T> { | ||
| let mut entries = self.0.lock().unwrap_or_else(PoisonError::into_inner); | ||
| entries.entry(name.to_string()).or_default().clone() |
There was a problem hiding this comment.
When a remote catalog repeatedly rotates through fresh rendition or stream names, every ExportSource::for_* call inserts another entry here, while the registry intentionally never evicts entries and is shared for the source's lifetime even when --echo is not enabled. An fMP4 or FLV exporter waiting for its initial header can process arbitrarily many valid catalog updates, letting a hostile publisher grow both memory and later stats() sampling cost without bound; cap lifetime distinct IDs or reject further churn.
AGENTS.md reference: AGENTS.md:L18-L18
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 7bc5705 (#5848, merged here): a registry keeps the first 1,024 rendition IDs (MAX_ENTRIES). Past that a new name counts into counters no report reads, so neither the map nor stats() sampling grows; a known name still continues its counts. renditions_past_the_cap_are_not_kept fails without the cap.
(Written by Claude Opus 5.5)
| continue; | ||
| } | ||
| if active { | ||
| let reader = tokio::spawn(read_echo(origin.clone(), path.clone(), lines.clone())); |
There was a problem hiding this comment.
When export echo starts on a large origin, announcement replay visits all routes and this path creates one task and subscription for every matching feedback broadcast, but the change adds no benchmark sweeping total announced routes against matching echo publishers. Add that two-axis benchmark so growth with the whole route table rather than the touched feedback set is visible.
AGENTS.md reference: AGENTS.md:L40-L40
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Not adding one here. The replay over the whole route table is the origin's announced(), shared by every announcement consumer, so a sweep of it belongs with moq-net's benchmarks rather than this sink's. What export echo adds is a prefix comparison per announcement and one reader per matching .echo broadcast, which is the touched set by design, and readers are now bounded in time after their broadcast ends (7bc5705).
(Written by Claude Opus 5.5)
| #[default] | ||
| #[serde(other)] | ||
| Data, |
There was a problem hiding this comment.
For downstream code that casts or persists this public fieldless enum, moving Data after Audio and Video silently changes the implicit values from Data=0, Audio=1, Video=2 to Audio=0, Video=1, Data=2. Keep the old discriminants explicitly while satisfying serde(other) so existing Class as ... values remain stable.
AGENTS.md reference: rs/AGENTS.md:L48-L48
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 0f0d474. #[serde(other)] has to sit on the last variant, so Data stays last, and explicit discriminants (Audio = 1, Video = 2, Data = 0) keep the values main gave each class. class_keeps_its_discriminants checks the casts and that an unknown name reads as data; it fails with the implicit values.
(Written by Claude Opus 5.5)
| /// One row per elementary stream PID the exporter writes, ordered by PID. | ||
| pub streams: BTreeMap<u16, Stream>, | ||
| /// How the release clock is keeping up with the source. | ||
| pub release: Release, |
There was a problem hiding this comment.
This replaces the public ts::stats::Export fields dropped, trimmed, gaps, drift, and out_of_tolerance with the nested release field, so existing callers must rewrite field access, yet rs/moq-mux/MIGRATE.md's ## Unreleased section is untouched. Add an entry explaining the move and the new access path.
AGENTS.md reference: AGENTS.md:L63-L63
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in 0f0d474: rs/moq-mux/MIGRATE.md now has an Unreleased entry saying that the five counters are fields of Export::release, a ts::stats::Release, which ts::Release::stats also returns, and that callers read export.release.dropped where they read export.dropped.
(Written by Claude Opus 5.5)
- `export stats` and `export echo` read a report with a 1 MiB cap on its reconstructed state, so patches that keep adding keys end the read instead of growing the sink. - `export echo` keeps a reader whose broadcast ended, so a start at the path replaces it, and aborts it after 5 s: the final snapshot still drains, and a track that outlives its route no longer leaves a reader behind. - A stats registry keeps the first 1,024 rendition IDs; a catalog that keeps renaming its renditions no longer grows it, or each report, without bound. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Takes moq-dev#5848's bounds on the report sinks: a 1 MiB cap on a report's reconstructed state, a 5 s drain for an ended echo reader, and 1,024 rendition IDs per stats registry. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
`Data` stays last for `#[serde(other)]`, and explicit discriminants keep the `as` values `main` gave each class. MIGRATE.md tells callers where `Export`'s release counters went. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…readers At the registry cap, a new rendition ID now evicts a single ended one, so the others keep their final counts. `export echo` forgets a reader once its drain passes instead of on the next `End`. The sinks read `latest`, unwrapping main's `Timed` snapshots. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Drop recovery's stale #5849 gate (gaps is on main), move the encrypted moq-json round trip to the json adopter, scope the e2ee consumer to Rust and JS, aim binding docs at the generated C package, keep Unix listeners on one acceptor, and tighten several goals. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # rs/moq-mux/MIGRATE.md
|
Summary before merge: merged #5848's current head ( Every review finding on this PR is fixed or answered on its thread: Local This PR contains all of #5848, so it should land after #5848. (Written by Claude Opus 5.5) |
|
Merged The conflicts all came from #5848 landing as a squash: both sides added its code, and this branch adds the TS sections on top. Each hunk keeps this branch's side; main's later changes in the same files carry over untouched ( Against the pre-merge head, the merge adds only main's own changes. The resolution is mechanical, so the earlier review still covers the PR's content. Checks: (Written by Claude Opus 5.5) |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1bc982cff0
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
|
||
| impl Release { | ||
| pub(crate) fn of<K: Ord + Clone, T>(jitter: &crate::jitter::Buffer<K, T>) -> Self { | ||
| let millis = |seconds: f64| (seconds * 1e3).round() as i64; |
There was a problem hiding this comment.
Release::of rounds both signed second values into bare i64 milliseconds; besides making the public API unit-unsafe, a deadline missed by less than 0.5 ms becomes margin == 0, even though the field promises that every miss is negative, so sign-based consumers cannot detect that deficit. Keep a typed signed duration through the Rust API and perform a sign-preserving millisecond conversion only at the serde edge.
AGENTS.md reference: rs/AGENTS.md:L37-L37
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in bda2cf4. Release::margin and offset are now Option<SignedDuration>, a signed nanosecond span (std::time::Duration cannot go below zero), so the Rust API keeps the unit and the precision. Whole milliseconds appear only at the serde edge, rounded away from zero, so a deadline missed by 0.3 ms reports -1, where the old rounding gave 0. a_release_span_under_a_millisecond_keeps_its_sign pins that, and the doc/concept/stats.md rows say so.
(Written by Claude Opus 5.5)
| let sections = report.as_ref().map(crate::report::Stats::sections); | ||
| let report_mpegts = |snapshot: &ts::stats::Snapshot| { | ||
| if let Some(sections) = §ions { | ||
| sections.set(|report| report.mpegts = Some(snapshot.clone())); |
There was a problem hiding this comment.
When a TS feed rotates its PMT through enough elementary PIDs, Import::stats retains one row per PID across those routes, and populated health fields can make the mpegts object installed here exceed report.rs's 1 MiB MAX_REPORT. The project's own export stats consumer then returns TooLarge and permanently stops monitoring that otherwise valid producer report; either bound/evict the reported PID rows below that limit or choose a bounded reader limit that covers every report this producer can generate.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Fixed in bda2cf4 by sizing the cap to what this producer can publish rather than evicting rows. An import keeps one row per PID, and PIDs are 13 bits, so a report holds at most 8,192 rows. the_largest_ts_report_reads_whole builds that worst case (every PID, every counter and field at its limit, 1,024 renditions) and reads it back through moq_json::snapshot: 3.48 MB, which the old 1 MiB cap refused as TooLarge. MAX_REPORT is now 8 MiB, over twice that, and doc/bin/cli.md states the limit.
(Written by Claude Opus 5.5)
# Conflicts: # rs/moq-mux/MIGRATE.md
|
Merged The only conflict was Checks: (Written by Claude Opus 5.5) |
…the report cap for the largest TS report Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: bda2cf44bc
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| sampled = tokio::time::Instant::now(); | ||
| let stats = ts.export().stats(); | ||
| let counts = (stats.dropped, stats.trimmed, stats.gaps, stats.out_of_tolerance); | ||
| report(&stats); |
There was a problem hiding this comment.
When --delay is still buffering or every incoming frame is late, ts.next().await remains pending while it consumes, queues, or drops input, so this output-only block is never reached even though Echo::run keeps emitting snapshots once per second. The new mpegts and release fields therefore stay absent or stale and can hide an indefinitely rising dropped count or negative margin until a frame is emitted or the source ends; drive section sampling from an independent interval, including in the passthrough loop that repeats this pattern.
Useful? React with 👍 / 👎.
| `mpegts` holds `streams`, one row per elementary stream keyed by its PID in | ||
| decimal, and the stream-wide counters. A row has `track` (the track suffix), | ||
| `class` (`audio`, `video`, or `data`; an unknown class reads as `data`), |
There was a problem hiding this comment.
This new section enumerates the mpegts and release wire shape, including PID-key encoding and individual fields, even though scoped doc/AGENTS.md requires feature docs to leave field lists and wire encodings to API docs or drafts. Keeping a second schema copy here will drift from the serde types, so retain the user-facing monitoring behavior and link to a colocated reference instead.
AGENTS.md reference: doc/AGENTS.md:L4-L6
Useful? React with 👍 / 👎.
| | `margin` | Gauge, milliseconds rounded away from zero: the least time any frame arrived before its deadline over the last second. Negative when one missed, however narrowly. | | ||
| | `offset` | Gauge, milliseconds rounded away from zero: how far the output clock sits from the source's plus the delay. Positive holds frames longer than the delay. The clock pulls it shut over ten minutes. | | ||
| | `drift` | Gauge, parts per million: the source clock's rate against ours, positive running fast. | |
There was a problem hiding this comment.
With export ts --delay 0, including passthrough, Buffer::queue deliberately skips margin collection and steering never runs, so margin, offset, and drift remain absent forever rather than merely pending a measurement. The new table describes these gauges without that condition, which can make consumers treat a healthy zero-delay export as missing data; state the nonzero-delay requirement alongside the claim.
AGENTS.md reference: doc/AGENTS.md:L9-L9
Useful? React with 👍 / 👎.
|
|
||
| impl<'de> serde_with::DeserializeAs<'de, SignedDuration> for SignedMillis { | ||
| fn deserialize_as<D: serde::Deserializer<'de>>(deserializer: D) -> Result<SignedDuration, D::Error> { | ||
| i64::deserialize(deserializer).map(SignedDuration::from_millis) |
There was a problem hiding this comment.
When a peer report contains margin or offset outside the roughly 292-year nanosecond range but still inside i64 milliseconds, this deserializer calls from_millis, whose saturating multiplication silently changes the value. The typed report then reserializes to a different duration instead of refusing malformed input; use checked conversion here and return a serde error.
AGENTS.md reference: AGENTS.md:L17-L18
Useful? React with 👍 / 👎.
main now carries moq-dev#5849, whose stats::Release holds the release clock's counters. --sync's overruns and late join it: set beside gaps after Release::of, logged on the release-clock line and watched for change there, serialized only when nonzero like the rest, and listed in the stats concept page. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Replaces #5784, which GitHub closed when #5643 merged and its branch was deleted. Stacked on the Rust reporters PR, #5848; until that merges, this diff shows both commits. Merged #5848's head, which carries main at
f13283b30, bounds the report sinks (a cap on a report's reconstructed state, a 5 s drain for an ended echo reader, 1,024 rendition IDs per stats registry, ended ones making way past it), and letsexport statsread the connection root andexport echoa hidden prefix. From main: the decoder's stats handle isreportthere, main'sunplacedandrefusedcounters are omitted at zero like the rest and the stats doc names them, and a declared break is not a gap (#5815). Rebased onto main, which now carries #5643: #5713'strimmedandgapssit underreleasebesidedropped, and the passthrough counts its gaps too. That is the re-timing section the decision recorded on #5643 describes.Problem
ts::statscounts the TR 101 290 checks and each elementary stream's delivery, but only logs them.The TS health stats quest
asks for them in the media reports. On #5643, the maintainer asked for the export's re-timing stats
(frames dropped at the release deadline, the release margin, and the output clock's offset from the
source) to sit in their own section beside
mpegts.Approach
serde on
ts::stats.Snapshot,Stream, andClassare camelCase, with every fielddefaulted. Counters are omitted at zero, as hang's are.
quietis whole milliseconds. An unknownclassreads asdata. PID keys go throughDisplayFromStr: flattened into a snapshot, a sectionloses serde_json's integer-key parsing, and
hang::stats::Snapshot<Report>would not read back("invalid type: string "256", expected u16").
ts::stats::Report { mpegts, release }is theEof bothhang::stats::Snapshotandhang::echo::Snapshot. An import fillsmpegtsfromImport::stats(). A TS export fills it withits rows and adds
release. A passthrough export reportsreleasealone.ts::stats::Release:dropped,margin,offset,drift(ppm), andoutOfTolerance.stats::Exportbecomes{ streams, release }, andts::Release::stats()gives the passthrough's.The jitter buffer measures the two new gauges:
margin: the least time any frame arrived before its deadline over each whole second of ourclock, negative when one missed.
offset: the gap between the slack of the track sent latest and the delay, which the clock pullsshut, as last measured. Like
drift, it is absent until the rate is measured.In Rust both are a
SignedDuration, signed nanoseconds. A report carries whole milliseconds,rounded away from zero, so a miss under a millisecond stays negative.
CLI: the import's decode loop sets
mpegtsat each one-second sample and after the finaldrain. The TS export and the passthrough set theirs as they sample. The release-clock log line
gains
margin_msandoffset_ms. Both sinks stop reading a report past 8 MiB rather than 1 MiB:twice the largest a TS import can publish, a row for each of the 8,192 PIDs.
Docs: a
doc/concept/stats.md§ MPEG-TS sections covers:downstream wire);
PID_errorcovered byunitsandquiet;releasefields.Also a paragraph in
doc/bin/cli.md. The quest file is removed, with its links.Decisions a reviewer may want to revisit:
release, andtrimmedandgapsin it: they are theexport's own counts, not a row's, though
gapsis about the source rather than the clock.marginover one second (a report's interval) rather than since the last sample, so two readers donot interfere.
marginandoffset, a typed span in Rust.Impact
ts::stats::Export:dropped,trimmed,gaps,drift, andout_of_tolerancemove underrelease.rs/moq-mux/MIGRATE.mdgives the new path.Class::Datamoves last, which#[serde(other)]requires; explicit discriminants keep eachclass's
asvalue..zdeltas stay small, since counters only climb.Results
just cipasses: 3,557 and 480 tests, bar two ofmain's own checks that fail on macOS only(
moq-carla-bridge's rustdoc andbench/relay.test.py), the steps after each run by hand. On macOS,play's link step tripsbuild.warningson the__eh_framelinker note; with warnings allowed itpasses 224 (one skipped), that note being its only warning.
a_report_flattens_beside_the_media_countersquietin ms, zeros omitted) and a full read-back throughhang::stats::Snapshot<Report>; an unknown class reads as dataDisplayFromStrclass_keeps_its_discriminantsClass'sasvalues stayData = 0,Audio = 1,Video = 2withDatalast; an unknown name reads as datathe_margin_is_the_least_spare_over_a_seconda_release_span_under_a_millisecond_keeps_its_sign-1, 0.3 ms to spare1, −12 ms-120the_largest_ts_report_reads_whole(CLI)moq_json::snapshot: 3.48 MB, under half the capTooLargeat 1 MiBthe_clock_follows_a_drifting_source_for_a_dayoffsetwithin the run's existing bound at ±25 and ±30 ppm, andmarginequals the delay plusoffsetto 1 msa_slow_source_clock_keeps_the_delay(export)import_stats_count_what_each_rendition_wrote(CLI)mpegtspresent, every TR 101 290 counter zero on the clean clip, every PID'sunitsadvancing and ending equal to an in-process import's, and noreleaseat importimport_stats_report_a_silent_video_pid(CLI)quietpeaks (the import logged 1.24 s) while audio stays under 500 ms; video counts again after; no priority 1 check firesexport_echo_reports_what_the_export_reads(CLI)releasecarries a margin (2,737 ms on a 4 s delay, nothing dropped)(Written by Claude Opus 5.5)