feat(playback): freeze visibility + cache-resume tuning (v10.11.0)#46
Merged
Conversation
…nstrumentation blind spot Owner keeps seeing freezes that every counter calls clean. They're right: paused-for-cache episodes (network starvation — greg TCP resets, slow opens) are invisible to the frame-drop counters AND shorter than the 20s stall watchdog. The hour bench logged 5 connection resets that 'recovered' — each one on a young clip is a visible freeze nothing counted. - paused-for-cache is now observed per cell (bare-emit observer per the CLAUDE.md rules): every episode shows a pulsing BUFFERING card on the cell (a freeze now explains itself instead of looking broken), logs a WARNING with measured duration + cache-buffering-state, and rides the stats dump as freezes / freeze_seconds per cell — benches now quantify the exact thing the owner experiences. - cache-pause-wait=3: mpv resumes only with 3s of buffer in hand, so a starvation is one slightly-longer pause instead of a resume/starve flicker loop (probed: accepted + read back). - Track changes bank an open freeze episode (no cross-track leakage). - Seek cap 0.90 → 0.98: the property-driven EOF advance made the clip tail safe; 10% of every video was unreachable for no remaining reason. - Buffering card replaces stale title/loading cards (test-caught: it deferred to them initially). tests/test_freeze_visibility.py pins: episode counting + card lifecycle, startup-fill exclusion, stale-generation guard, track-change banking, and the 0.98 cap. CLAUDE.md gains the freeze-triage rule. Version 10.11.0. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
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 blind spot
The owner keeps seeing freezes that every counter calls clean — and the counters are the problem.
paused-for-cacheepisodes (network starvation: greg's TCP resets, slow stream opens) are invisible to the frame-drop counters and shorter than the 20s stall watchdog. The hour bench logged five connection resets that "recovered" — each one, on a young clip with a shallow buffer, is a visible freeze nothing measured.What ships
paused-for-cacheis observed per cell → a pulsing BUFFERING card on the frozen cell (replacing stale title/loading cards), a WARNING log with the measured duration and cache state, andfreezes/freeze_secondsper cell in the stats dump. The next bench quantifies the exact thing the owner experiences.cache-pause-wait=3— mpv resumes only with 3s of buffer in hand, converting resume/starve flicker-loops into one slightly-longer pause (probed: accepted + read back).Verification
5 new tests (episode counting + card lifecycle, startup exclusion, stale-gen guard, track-change banking, 0.98 cap) — one of which caught a real design flaw before ship (the card deferred to stale overlays). Suite green; CLAUDE.md gains the freeze-triage rule; version 10.11.0; exe rebuilt.
The bigger revamp
This is deliberately the first tranche: it converts an invisible complaint into measured data. One normal viewing session on this build tells us whether the remaining freezes are network (expect FREEZE lines naming greg) or something new — and that evidence decides whether the next tranche is server-side (Emby/greg tuning), buffer strategy, or moot.
🤖 Generated with Claude Code