tracker

4.9 KB

1. # Report 067: tracker#18 delta sync, show record refresh, summary priority
3. **Commit**: `f83571c` on branch `t18` (worktree loreana:/media/STORAGE/projects/tracker-wt18, from t26 head 10bb3f9). Not pushed, not merged, not deployed. main, tracker-wt26 and tracker-wt19 were not touched.
5. ## Files (merge notes)
6. - New: `deltasync.hl` (the plan from the change lists, `sync-state.txt`, one-time `clearCopiedSummaries`), `tests/peek-shows.hl`, `tests/realdata-018.mjs`, `tests/realdata-018-poll.mjs`.
7. - `tmdbsync.hl`: `syncShowDelta` (the light step), `changedSeasonNumbers`, `emptyResult`. The TVmaze request is now `/shows/<id>?embed=episodes`, which also brings `tvmazeSummary` (through `stripHtml`). `syncShowPart` only reads the episodes of its own part, and takes an `only.tvmaze` flag.
8. - `details.hl`: `withDetailsFields` (shared with #26 `firstStep`, which no longer writes `summary`) and `applySynced`.
9. - `project.hl`: `syncTick` rewritten for planning, delta and full walk. The clock's `busy` now includes planning. Adds the boot-time summary clear.
10. - `tools/sync-tmdb.hl`: `--delta [--since D] [--plan]`, and uses `applySynced`.
11. - `search.hl`: `haveTvmz`/`titleIdByTvmaze`, `searchTitleRenamed` (stale entries), and import `summary = null`.
12. - `people.hl` (1 line, `summary = null`), `shows.hl` (`summaryOf`), `components/show.hl` (uses `summaryOf`).
13. - Tests: `tests/browser.mjs`, `tests/faketmdb.mjs` (change lists, updates, embed, summaries), `tests/seed-show.hl` (the movie's summary is a copy).
14. - `README.md` and `STATUS.md`: expect conflicts with the t19 and 062 merges in these two docs only. The code overlaps with people.hl and search.hl are small.
16. ## Gate
17. `TRACKER_GATE_SHOTS=/tmp/w067-shots TRACKER_GATE_PORT=8751 TRACKER_GATE_IDENT_PORT=8752 TRACKER_GATE_TMDB_PORT=8753 TRACKER_GATE_CHROME=8754-8756 node tests/browser.mjs` → **`261 passed, 0 failed`** (249 before, plus 12 #18 checks).
19. What the 12 checks prove:
20. - A changed title is refreshed: renamed, with its new text and tagline.
21. - The creator's own summary is kept.
22. - A copied summary is cleared at the first start.
23. - Titles that did not change are untouched (same `tmdbSync`).
24. - An unfollowed series gets only its newest season.
25. - Search finds a title by its new name.
26. - A failing change list falls back to the full walk and does not advance the stored day.
27. - More than 14 days since the last run gives the full walk.
28. - TVmaze's text is shown when there is no other summary.
30. ## Numbers (real data copy 2026-10-01, real TMDB and TVmaze, one day of changes)
31. - **Changes**: TMDB tv 3,088 + movie 9,559 ids (31 + 96 pages), TVmaze 312 → **336 of our titles (28 followed)**.
32. - **Delta run**: 128 change requests + 477 title requests + 29 TVmaze. It took **536 s** (511 s on a second run), 0 errors, slowest step 2.5 s. Pages during the run: median 45 ms, p99 693 ms, max 2.2 s.
33. - **Old full walk** on the same copy: 128 shows, 259 + 128 TVmaze requests, **312 s**.
34. - My first variant gave every title the full step. Today (15,716 episodes) and EastEnders (7,349) then blocked the server for 20.8 s each, and RSS reached 14 GB. That is why titles nobody follows now get the light step.
35. - **Cleared summaries: 9,647**, 24 kept. All 24 are followed shows whose tmdbSummary the #9 sync changed after the migration, so they are probably older TMDB texts, not rewrites (list in STATUS). One of them, Star-Trek-First-Contact, has another title's tmdbSummary: wrong tmdbId on that record.
36. - First day only: 24 titles renamed, 79 TMDB texts changed, 173 taglines, +79 seasons / +4,475 episodes. The new seasons and episodes belong to unfollowed series that were never synced since the migration.
37. - Screenshots (I looked at them) are on Loreana in `/tmp/w067/real/` (Andor renamed; Breaking Bad shows TMDB's current text) and `/tmp/w067-shots/delta-{show,movie}-1280.png`.
39. ## Open points
40. - It is **not "a handful of requests"**: about 634 a day against 387 for the old run, mainly because the movie change list is 96 pages. In return every title stays current. Dropping unfollowed movies would save roughly 300 requests a day; that is the creator's call.
41. - **Memory**: about 10 MB per synced title is never freed (the Hybriel leak described in #28). A delta run adds about 3.5 GB RSS, the old run about 2.8 GB. A daily restart or a runtime fix is needed either way. The one-time summary clear adds about 2 GB at that boot.
42. - **Titles**: TMDB's en-US name replaces the migrated title, including the old "Name(original)" form, e.g. Kaun Banega Crorepati(कौन बनेगा करोड़पति) → Kaun Banega Crorepati. To keep a title, a creator-title field like `summary` would be needed.
43. - The first live run after the deploy is a full walk, because there is no `sync-state.txt` yet.
44. - **Hybriel issues**: no new tickets. Two things to know:
45. - `mode` cannot be a top-level variable name; it is parsed as the file mode directive.
46. - The known fetch/decode memory leak. A small HTTPS fetch alone grows RSS by about 0.57 MB per request (60 TVmaze fetches: 6 → 51 MB).