You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
moq-hls: a named sibling rendition is pinned too late to survive a same-path republish #2870
#2795 bound renditions to the broadcast their catalog came from, so a same-path republish (whose group numbering restarts) cannot have its media served under the replaced broadcast's segment numbers, durations, and PROGRAM-DATE-TIME. It closed this for both rendition shapes:
inline (config.broadcast == None, the catalog's own broadcast) — seeded at construction via upstream.local(...).
named sibling (config.broadcast == Some(path)) — resolved at the per-rendition watcher's start, before the watcher pushed its first timeline row, so the timeline and the media it describes always agreed.
On dev (after #2852 merges main), only the first half survives. dev's HLS export is fetch-on-demand: there is one watcher per timeline section in export/mod.rs fanning entries out to every rendition, rather than main's watcher per rendition. The sibling resolution therefore has nowhere to hang, and Rendition::track resolves lazily on the first media request instead:
That leaves a window: the fanout can publish timeline rows for a sibling rendition while live.broadcast is still unset. A same-path republish landing in that window means the lazy resolve returns the replacement broadcast, whose restarted group numbers are then served under the old catalog's segment ranges, durations, and wall-clock metadata. This is exactly the failure #2795 describes, reachable only through the sibling path.
The pin itself is correct once taken (OnceLock::set, so the first resolution wins). The problem is purely when it is taken.
Note this is not a regression from the merge: dev always resolved lazily, and #2852 carries main's inline fix plus its regression test. It is main's sibling fix failing to carry across an architectural difference, so it should be closed deliberately rather than assumed handled.
Why the test doesn't catch it
a_replacement_publisher_is_not_served_under_the_replaced_catalog survived the merge and is not vacuous, but its rendition is built from a plain VideoConfig::new(VideoCodec::VP8) with no broadcast reference — an inline rendition. It exercises only the eagerly-seeded path.
Suggested direction
Resolve each rendition's sibling broadcast after construction but before the section watcher pushes any entry to it. Construction itself can't do it: renditions::Producer::sync runs synchronously under a write lock, which is why main used the watcher. Then add a regression that republishes a named sibling before the first media request and proves the replacement's bytes are not served under the old catalog.
Found by Codex adversarial review of #2852, verified against main's export/rendition.rs::watch and dev's fanout in export/mod.rs.
#594 - Rename name to path AGAIN [closed] #989 - Rename path back to name. [closed] #1371 - feat: relative broadcast paths for cross-broadcast catalog renditions (incl. WHEP egress) [merged] #2254 - fix(moq-hls): release source subscriptions when a Broadcaster is dropped [merged] #2266 - fix(moq-hls): reconcile catalog renditions [merged] #2630 - fix!: reject a catalog whose broadcast reference escapes the root [merged] #2795 - fix(moq-hls): bind renditions to the broadcast their catalog came from [merged]
📝 Issue Planner
Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.
Create Plan
🧪 Issue enrichment is currently in open beta.
You can configure auto-planning by selecting labels in the issue_enrichment configuration.
To disable automatic issue enrichment, add the following to your .coderabbit.yaml:
issue_enrichment:
auto_enrich:
enabled: false
💬 Have feedback or questions? Drop into our discord!
The window is now narrow (renditions bind at construction), and what remains is covered by the broadcast-epoch catalog references quest (quest/m0/broadcast-epoch/catalog-references.md).
Activity
coderabbitai commentedon Aug 15, 2026
🔗 Related PRs
#594 - Rename name to path AGAIN [closed]
#989 - Rename path back to name. [closed]
#1371 - feat: relative broadcast paths for cross-broadcast catalog renditions (incl. WHEP egress) [merged]
#2254 - fix(moq-hls): release source subscriptions when a Broadcaster is dropped [merged]
#2266 - fix(moq-hls): reconcile catalog renditions [merged]
#2630 - fix!: reject a catalog whose broadcast reference escapes the root [merged]
#2795 - fix(moq-hls): bind renditions to the broadcast their catalog came from [merged]
📝 Issue Planner
Check the box below or use the
@coderabbitai plancommand to generate an implementation plan and prompts that you can use with your favorite coding assistant.🧪 Issue enrichment is currently in open beta.
You can configure auto-planning by selecting labels in the issue_enrichment configuration.
To disable automatic issue enrichment, add the following to your
.coderabbit.yaml:💬 Have feedback or questions? Drop into our discord!
fix(moq-hls): bind a named sibling rendition when it is created (#3445)
kixelated commentedon Oct 19, 2026
The window is now narrow (renditions bind at construction), and what remains is covered by the broadcast-epoch catalog references quest (
quest/m0/broadcast-epoch/catalog-references.md).(Written by Claude Opus 5.5)