[Tracker] 2026-06-10 issue triage — 26 open

[jakebromberg]

Cross-repo rollup: items from this tracker that block other repos or carry external deadlines are mirrored in Cross-repo critical path (org Project #33).

End-of-day triage of all 26 open issues in WXYC/discogs-etl. Intended as the entry point for any contributor or agent picking up after 2026-06-10. Supersedes #238 (2026-05-29). Updated 2026-06-10 — original 2026-06-03 snapshot has been refreshed in place; what changed since is summarised in the section below.

What's changed since 2026-06-03

• The 2026-06-07 manual jumpstart rebuild (Manual jumpstart rebuild to land #497 fields ahead of 2026-07-04 cron #267, closed) ran end-to-end and populated artist.profile, artist_alias, artist_name_variation, artist_member, artist_url for the first time. This is the most substantive shift in the repo's coverage shape since Production cron pair-wise filter never executed against Railway prod #188 (the 2026-05-14 rebuild). Consequences land throughout this tracker.
• Rebuild-hygiene cluster: heavy churn. Rebuild fetches artists.xml.gz; direct-pg path loads artist details #264 (artists.xml.gz fetch + directory mode), rebuild-cache.sh:153 ERR trap clobbers $? via rm -rf — silently exits 0 on pipeline failure #269 (ERR-trap exit-code clobber), rebuild-cache.sh: redirect TMPDIR to $WORK_DIR — converter CSVs land in tmpfs (~2 GB RAM cap) not EBS #271 (TMPDIR → $WORK_DIR), cache_metadata missing UNIQUE(release_id) — populate_cache_metadata ON CONFLICT fails on load #273 (cache_metadata UNIQUE) — all closed 2026-06-07 as part of getting the jumpstart over the line. ephemeral-rebuild: 100 GB EBS exhausted at 94% of release scan — bump volume or split /tmp #268 (EBS exhaustion at 94%) and Add PG integration test for populate_cache_metadata's ON CONFLICT race-tolerance #207 (cache_metadata race-tolerance integration test) also closed.
• LML side of the artwork-NULL epic finished. Library-metadata-lookup#221 (one-shot live-API drain) closed 2026-06-07. With Track % NULL artwork_url in release table over time; alarm on stagnation #241 + Monthly cache rebuild wipes LML-back-patched release.artwork_url; preserve via COALESCE upsert #242 already closed, every Scope-row in Epic I ([Epic I] Artwork-NULL ambiguity #247) is now ✅; this tracker recommends close (status comment posted 2026-06-10).
• LML#356 (ANV-from-dump) closed 2026-06-07 as a dup of LML#497 — the underlying gap is fixed by the same converter+loader work the jumpstart rebuild exercised. The cross-repo runtime risk called out in the 2026-06-03 cut is materially resolved.
• [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229 epic's scope shrank by ~50%. The rebuild now fetches+processes artists.xml.gz, so the epic's residual scope is labels + masters only, not all three. Per-issue status comments posted on rebuild-cache.sh: fetch + process all four Discogs dump files (artists, labels, masters in addition to releases) #228 and [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229 noting the shrink. Stub bodies still uninterpolated 21 days post-filing.
• New issue: rebuild-cache-bootstrap.sh: skip the discogs-xml-converter clone+build when prebuilt download will succeed #275 (bootstrap clone+build redundant given prebuilt-download happy path). Low-priority cold-start latency cleanup; labeled enhancement.
• New cross-repo dependency: LML#510 (Tombstone 404s for get_artist_details + get_release via not_found column discriminator) — requires an alembic migration in this repo to add the column. Scheduled after BS#1361's BS-side mitigation per LML's tracker (#515).
• PR test: bump charset-torture corpus pin to v0.12.0 #150 (charset-torture corpus pin v0.12.0) — CI all green; mergeable-status is now DIRTY (rebase needed). No activity since 2026-06-03.

P0 sequencing — recommended order

1. Close-out [Epic I] Artwork-NULL ambiguity #247 (Epic I — Artwork-NULL ambiguity). Every Scope row is ✅ as of 2026-06-07. Either back-publish the post-drain artwork_null_share measurement and close, or close now and file a thin follow-up specifically for the published-baseline record. Cheapest piece of outstanding work in the queue and removes the highest-noise stale tracker.
2. Repair or close + re-file the four stub-body issues (check_cache_drift.py: assert artist_name_variation populated + release_track_artist.extra column present #226 / rebuild-cache.sh hygiene: non-zero exit on wait_for_postgres timeout + mask DSN in import_csv.py log output #227 / rebuild-cache.sh: fetch + process all four Discogs dump files (artists, labels, masters in addition to releases) #228 / [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229). Now 21 days post-filing without repair. [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229 in particular is materially smaller post-jumpstart (labels+masters only); a clean re-file with the post-Manual jumpstart rebuild to land #497 fields ahead of 2026-07-04 cron #267 framing is probably less work than reconstructing the original intent. Per-issue status comments dated 2026-06-10 spell out the changes.
3. rebuild-cache-bootstrap.sh: skip the discogs-xml-converter clone+build when prebuilt download will succeed #275 — move the discogs-xml-converter clone+build out of rebuild-cache-bootstrap.sh to lean on the prebuilt-download path. Low-priority but mechanically straightforward. ~2-5 min off every cold start, no correctness impact.
4. Resolve release_artist.role schema drift: drop the column or wire the converter #210 — execute Option A (drop release_artist.role). One alembic migration plus mirror in schema/create_database.sql:77-83. Decision input was complete in [Tracker] 2026-06-10 issue triage — 26 open #263's 2026-06-03 cut; remaining work is mechanical.
5. Pick DuckDB + extension versions; create plans/duckdb-version-pin.md in WXYC/wiki #189 — pick DuckDB + extension versions and create WXYC/wiki/plans/duckdb-version-pin.md. First-mover prerequisite for the entire Epic: rewrite verify_cache.py against DuckDB ATTACH (Project 2 candidate 1) #186 epic chain (Rewrite scripts/verify_cache.py against DuckDB ATTACH #190 → Parity test: rewritten verify_cache.py vs original on fixture corpus #191 → Update README.md and CLAUDE.md for the new DuckDB-based pattern #192 → Production smoke-test against full prod-shaped corpus #193). Cheap, unblocks four downstream issues.
6. Add PG integration test for populate_cache_metadata's ON CONFLICT race-tolerance #207 + Add integration test for release_artist import with the converter's actual CSV shape #208 + Add PG integration test for the NOT VALID FK race-tolerance pattern #212 — three follow-up PG integration tests for the recent race-tolerance fixes. (Note Add PG integration test for populate_cache_metadata's ON CONFLICT race-tolerance #207 closed 2026-06-07; Add integration test for release_artist import with the converter's actual CSV shape #208 + Add PG integration test for the NOT VALID FK race-tolerance pattern #212 remain.)
7. PR test: bump charset-torture corpus pin to v0.12.0 #150 — rebase against current main and merge; CI is all green and the original gating dep (wxyc-shared v0.12.0) is long-resolved.

Critical / high severity

• CRITICAL: none active in the queue.
• HIGH: check_cache_drift.py: assert artist_name_variation populated + release_track_artist.extra column present #226, rebuild-cache.sh hygiene: non-zero exit on wait_for_postgres timeout + mask DSN in import_csv.py log output #227, rebuild-cache.sh: fetch + process all four Discogs dump files (artists, labels, masters in addition to releases) #228, [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229 (stub bodies — load-bearing intent partially overtaken by 2026-06-07 work but never repaired; the four issues collectively cover the full-dump expansion epic and the rebuild hygiene cluster).
• MEDIUM: Add CI parity check: converter CSV output vs import_csv.py expected columns #209, Resolve release_artist.role schema drift: drop the column or wire the converter #210, Decide: run VALIDATE CONSTRAINT after rebuild, or accept NOT VALID as steady state #213 (operational/test gaps that gate the next drift class), plus rebuild-cache-bootstrap.sh: skip the discogs-xml-converter clone+build when prebuilt download will succeed #275 (cold-start latency).

Cross-cluster conflicts (decide once before starting)

DuckDB rewrite vs. verify_cache hardening (carried)

#186's epic rewrites the entire scripts/verify_cache.py (75 KB / 1854 LOC currently). The recent column-loss fixes (#232/#234, closed) were targeted patches on that same script. If the rewrite lands, those patches don't migrate automatically. Decision still needed before #190 starts: does the parity test in #191 assert the same column-preservation invariants those patches were defending? If yes, fold into parity. If no, file a follow-up to re-derive them in the DuckDB shape.

Full-dump expansion (#229) vs. ANV ask — RESOLVED

The LML#356 ANV ask that was previously called out as natural-fit-inside-#229 closed 2026-06-07 as dup of LML#497. The underlying gap was already shipped by the jumpstart rebuild's converter+loader work. #229's residual scope is labels+masters only; the ANV thread is closed end-to-end.

release_artist.role decision (#210) — RESOLVED at decision-input level (carried)

Cross-repo grep + converter inspection done 2026-06-03; no consumer references the column. Mechanical work remaining: ship Option A. Not really a conflict anymore.

Obsolete / needs revision (recommend close)

• feat: Dual-source catalog abstraction for tubafrenzy and Backend-Service #50 (Dual-source catalog abstraction). scripts/run_pipeline.py:187-335 implements --catalog-source / --catalog-db-url / --wxyc-db-url. Phases 1-6 all in tree. Already labeled obsolete.
• Add name collision resolution tooling #59 (Add name collision resolution tooling). scripts/resolve_collisions.py (20 KB) is on main. Already labeled obsolete.
• Integrate ON_STREAMING sync into daily library pipeline #96 (Integrate ON_STREAMING sync into daily library pipeline). scripts/sync-library.sh:152-162 runs export_streaming_links.py on every build; already labeled obsolete.
• deploy-ephemeral-rebuild workflow lacks AWS credentials; CI infra updates blocked #202 (deploy-ephemeral-rebuild workflow lacks AWS credentials). Workflow migrated to OIDC; already labeled obsolete.
• [Epic I] Artwork-NULL ambiguity #247 (Epic I — Artwork-NULL ambiguity). Substance fully shipped 2026-05-30 + 2026-06-07 LML drain. Status comment posted 2026-06-10. Recommend close once acceptance measurement back-published (or split that into its own thin issue).

Stub-body issues — bodies failed to interpolate at filing (carried, 21 days uninterpolated)

• check_cache_drift.py: assert artist_name_variation populated + release_track_artist.extra column present #226 / rebuild-cache.sh hygiene: non-zero exit on wait_for_postgres timeout + mask DSN in import_csv.py log output #227 / rebuild-cache.sh: fetch + process all four Discogs dump files (artists, labels, masters in addition to releases) #228 / [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229 — filed 2026-05-20 with body content @/tmp/<file>.md (literal). Per-issue status comments dated 2026-06-10 spell out what the 2026-06-07 work shifted in each one. Until repaired, do not treat any of the four as actionable in their current shape.

Cross-repo dependencies (encoded both natively and in body prose)

Issue	Blocked by	Method	State
#190	#189	native blockedBy	OPEN
#191	#190	native blockedBy	OPEN
#192	#190	native blockedBy	OPEN
#193	#191	native blockedBy	OPEN
#228	#226 (#222 closed; stale link)	native blockedBy	OPEN
#186 (epic)	sub-issues #189-#193	sub-issue relationships	OPEN
#229 (epic)	sub-issues #226/#227/#228	sub-issue relationships	OPEN

Cross-repo runtime risks

• feat(discogs): tombstone 404s for get_artist_details + get_release via not_found discriminator library-metadata-lookup#510 (Tombstone 404s for get_artist_details + get_release via not_found column discriminator) — REQUIRES a discogs-etl alembic migration for the not_found column. Per LML's 2026-06-07 tracker (#515), schedules after BS#1361's BS-side mitigation lands. Net-new cross-repo dep since [Tracker] 2026-06-10 issue triage — 26 open #263's last cut.
• feat(cache): POST /api/v1/cache/refresh-for-identities — source-agnostic cache backfill by identity_id library-metadata-lookup#525 (POST /api/v1/cache/refresh-for-identities — source-agnostic cache backfill by identity_id) — likely interacts with the cache schema and rebuild semantics; watch for downstream alembic asks.
• feat(identity): release-level identity — schema, mint-on-resolve, reconciliation log shape library-metadata-lookup#526 (release-level identity — schema, mint-on-resolve, reconciliation log shape) — cache shape implications; may surface a downstream discogs-etl backfill or schema-mirror issue once landed.
• Cache schema-skew hardening: UndefinedColumnError → CacheUnavailableError arms process-wide cool-down library-metadata-lookup#476 (cache schema-skew hardening with process-wide cool-down). Open. Defends against discogs-etl schema changes that LML hasn't picked up yet; relevant whenever Resolve release_artist.role schema drift: drop the column or wire the converter #210 (role drop) or [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229 (labels+masters expansion) or the #510 migration ships.
• Model sub_tracks parent-child relationship via release_sub_track table discogs-xml-converter#61 (model sub_tracks parent-child via release_sub_track table) — open. A non-trivial fraction of release_track rows in the current cache may be wrong shape because of nested <track> inside <sub_tracks>. Affects Audit: rebuild release count below expected pair-wise band (38,662 vs 50–80K) #217's release-count audit (candidate root cause), and would surface in the DuckDB rewrite's parity test (Parity test: rewritten verify_cache.py vs original on fixture corpus #191) — parity may need to be measured against a corpus where this is fixed, otherwise both implementations reproduce the same wrong rows.
• [Epic] MD-set "Not on Discogs" flag for catalog releases Backend-Service#1280 epic + #1281/#1282/#1283/#1294 — "MD-set 'Not on Discogs' flag for catalog releases". Daily cron and lml-client extensions interact with discogs-cache reads; align with the cache-side 404 tombstone story (LML#510) so the gate semantics don't fork.
• feat(schema): add rotation.lml_identity_id; mint LML identities on dj-site paste Backend-Service#1380 + #1381 — rotation-artist-backfill migrating from Discogs-specific carving to lml_identity_id fan-out. Reduces direct discogs-cache write pressure from the rotation backfill path.

Cross-repo runtime risks — RESOLVED since 2026-06-03

• Populate artist_name_variation during cache rebuild (not only runtime api_fetch) library-metadata-lookup#356 (populate artist_name_variation during cache rebuild) — closed 2026-06-07 as dup of LML#497; the converter+loader work the jumpstart rebuild exercised covers it.
• validate_track_on_release should filter release_track_artist to extra = 0 library-metadata-lookup#333 (extra=0 filter on validate_track_on_release) — closed; the column has been in tree and is now consumed.

Open PR in flight

• PR test: bump charset-torture corpus pin to v0.12.0 #150 — bump charset-torture corpus pin to v0.12.0. Author: jakebromberg; created 2026-05-05. CI all green (lint, drift-check, test, pg, marker-sync). mergeStateStatus: DIRTY — needs rebase against current main. Gating dep (wxyc-shared v0.12.0) is long-resolved. Rebase + merge.

Cluster map (26 open)

• DuckDB verify_cache rewrite (epic): Epic: rewrite verify_cache.py against DuckDB ATTACH (Project 2 candidate 1) #186, Pick DuckDB + extension versions; create plans/duckdb-version-pin.md in WXYC/wiki #189, Rewrite scripts/verify_cache.py against DuckDB ATTACH #190, Parity test: rewritten verify_cache.py vs original on fixture corpus #191, Update README.md and CLAUDE.md for the new DuckDB-based pattern #192, Production smoke-test against full prod-shaped corpus #193.
• Full-dump expansion (epic; body broken; now labels+masters-only post-jumpstart): [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229, rebuild-cache.sh: fetch + process all four Discogs dump files (artists, labels, masters in addition to releases) #228, check_cache_drift.py: assert artist_name_variation populated + release_track_artist.extra column present #226.
• Rebuild hygiene (stub-bodied; partially overtaken by 2026-06-07 work): rebuild-cache.sh hygiene: non-zero exit on wait_for_postgres timeout + mask DSN in import_csv.py log output #227, rebuild-cache-bootstrap.sh: skip the discogs-xml-converter clone+build when prebuilt download will succeed #275, Audit: rebuild release count below expected pair-wise band (38,662 vs 50–80K) #217.
• Test gap follow-ups from 2026-05-13 incident: Add integration test for release_artist import with the converter's actual CSV shape #208, Add CI parity check: converter CSV output vs import_csv.py expected columns #209, Add PG integration test for the NOT VALID FK race-tolerance pattern #212.
• Schema / operational decisions: Resolve release_artist.role schema drift: drop the column or wire the converter #210, Decide: run VALIDATE CONSTRAINT after rebuild, or accept NOT VALID as steady state #213.
• Infra / CI hygiene: Reshape test_base_trigram_indexes_use_unaccent from blocklist to allowlist #184.
• Epic close-out (recommend close): [Epic I] Artwork-NULL ambiguity #247.
• Likely-obsolete (recommend close): feat: Dual-source catalog abstraction for tubafrenzy and Backend-Service #50, Add name collision resolution tooling #59, Integrate ON_STREAMING sync into daily library pipeline #96, deploy-ephemeral-rebuild workflow lacks AWS credentials; CI infra updates blocked #202.

Active cross-cutting projects this repo participates in

• Music Data Pipeline Hardening (org Project Pipeline performance optimizations #19) — phases A–D done; E (monorepo) + F (split semantic-index) outstanding org-wide; discogs-etl issues touched by this stay on this tracker.
• Cross-repo critical path (org Project fix: dedup release IDs during CSV import #33) — mirror of the items in this tracker that block other repos (currently the latent Resolve release_artist.role schema drift: drop the column or wire the converter #210 / migration cluster; expected to gain the #510-driven alembic migration once filed).
• Linkage stabilization & metadata backfill (org Project Split combined multi-artist library entries for matching #24) — LML-anchored; discogs-cache substrate is the read surface for entity.identity reconciliation; Audit: rebuild release count below expected pair-wise band (38,662 vs 50–80K) #217's release-count audit feeds the calibration positive class.
• cross-cache-identity (org Project Split combined multi-artist library entries for matching #25) — bulk-resolve-libraries + LML#271 + #431 + #490; discogs-cache supplies the artist/release entity rows.
• Catalog Track Search (org Project perf: Set tables UNLOGGED before converter COPY in direct-PG mode #30) — flags ON 2026-05-21; LML-anchored; discogs-cache substrate is the read surface; Audit: rebuild release count below expected pair-wise band (38,662 vs 50–80K) #217's audit is a coverage check.

How to find work

• All blocked / sequenced issues carry a "## Blocked by" body section and a native blockedBy relationship (rendered under the "Relationships" pill in the GitHub UI).
• Per-issue triage commentary is on each issue's page (look for "Triage finding" / "Status update" / "Sequencing concern" / "Recommend closing" / "Stub-body finding" comments dated 2026-06-03 and 2026-06-10).
• Open PRs in flight: test: bump charset-torture corpus pin to v0.12.0 #150 (charset-torture corpus pin; CI green, needs rebase).
• Supersedes tracker [Tracker] 2026-05-29 issue triage — 24 open #238 (2026-05-29).

[jakebromberg]

Tracker delta since 2026-06-03 (T+4 days):

New work landed (#264, #265, #266 chain — LML#497 unblock):

• Rebuild fetches artists.xml.gz; direct-pg path loads artist details #264 (closed; PR feat(rebuild): fetch artists.xml.gz; load artist details in direct-pg path #265 merged 2026-06-07T01:27Z): rebuild fetches artists.xml.gz; direct-pg path loads artist details
• feat(import-csv): load artist_url.csv into artist_url table #266 (closed; PR merged 2026-06-07T04:56Z): load artist_url.csv into artist_url table

Manual jumpstart incident chain (2026-06-07T14:09 → T17:00):

• Manual jumpstart rebuild to land #497 fields ahead of 2026-07-04 cron #267 (OPEN): manual jumpstart to land #497 fields ahead of 2026-07-04 cron — first attempt failed at 94% of release scan; re-trigger pending after fix(rebuild-cache): redirect TMPDIR to EBS so converter CSVs do not fill tmpfs #272
• ephemeral-rebuild: 100 GB EBS exhausted at 94% of release scan — bump volume or split /tmp #268 (OPEN): EBS volume hypothesis (200 GB bump landed; was the wrong fix — benign; leave open until next jumpstart succeeds)
• rebuild-cache.sh:153 ERR trap clobbers $? via rm -rf — silently exits 0 on pipeline failure #269 (CLOSED; PR fix(rebuild-cache): propagate real exit code through ERR-trap cleanup #270 merged 15:42Z): ERR-trap clobbers $? — silent-failure mode confirmed before fix landed
• rebuild-cache.sh: redirect TMPDIR to $WORK_DIR — converter CSVs land in tmpfs (~2 GB RAM cap) not EBS #271 (CLOSED; PR fix(rebuild-cache): redirect TMPDIR to EBS so converter CSVs do not fill tmpfs #272 merged 16:15Z): real root cause — tempfile.TemporaryDirectory defaulted to /tmp (tmpfs ~2 GB cap on c6i.large) instead of EBS $WORK_DIR

New P0 (supersedes prior P0 ordering):

1. Re-trigger the Manual jumpstart rebuild to land #497 fields ahead of 2026-07-04 cron #267 manual jumpstart now that fix(rebuild-cache): redirect TMPDIR to EBS so converter CSVs do not fill tmpfs #272 is in main. Verify acceptance: artist_url populated, artist.profile coverage jumps, Yetsuby smoke test, fixture preservation across rebuild. Close Manual jumpstart rebuild to land #497 fields ahead of 2026-07-04 cron #267 and ephemeral-rebuild: 100 GB EBS exhausted at 94% of release scan — bump volume or split /tmp #268 on success.
2. Then update Populate api_fetch-only artist columns during cache rebuild (profile, urls, members, name_variations) library-metadata-lookup#497 with rebuild timestamp + row counts; close #497.
3. Then Rotation-scoped backfill: refresh artist cache rows for current-rotation releases library-metadata-lookup#501 (rotation-scoped backfill) can run against the meaningfully-populated baseline.

Carrying over from #263 unchanged:

• Stub-bodies check_cache_drift.py: assert artist_name_variation populated + release_track_artist.extra column present #226/rebuild-cache.sh hygiene: non-zero exit on wait_for_postgres timeout + mask DSN in import_csv.py log output #227/rebuild-cache.sh: fetch + process all four Discogs dump files (artists, labels, masters in addition to releases) #228/[Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229 — still uninterpolated (10 days since prior triage)
• Recommend-close items: feat: Dual-source catalog abstraction for tubafrenzy and Backend-Service #50, Add name collision resolution tooling #59, Integrate ON_STREAMING sync into daily library pipeline #96, deploy-ephemeral-rebuild workflow lacks AWS credentials; CI infra updates blocked #202, [Epic I] Artwork-NULL ambiguity #247 (still recommend-close after baseline question)
• DuckDB rewrite epic Epic: rewrite verify_cache.py against DuckDB ATTACH (Project 2 candidate 1) #186 chain — Pick DuckDB + extension versions; create plans/duckdb-version-pin.md in WXYC/wiki #189 still the first-mover gate
• PR test: bump charset-torture corpus pin to v0.12.0 #150 — charset-torture corpus pin v0.12.0; CI green; needs rebase or admin merge
• Resolve release_artist.role schema drift: drop the column or wire the converter #210 Option A (drop release_artist.role) — ready to ship

Cross-repo signal: WXYC/library-metadata-lookup#499 (?refresh=true) merged 04:01Z today; LML#497 still OPEN; LML#501 still OPEN. discogs-etl is the active substrate for the end-to-end unblock — the LML side is waiting on a successful rebuild here.

[jakebromberg]

[2026-06-08] Tracker delta since 2026-06-07 (T+1 day):

Manual jumpstart succeeded; full upstream chain closed.

• Manual jumpstart rebuild to land #497 fields ahead of 2026-07-04 cron #267 (manual jumpstart) — CLOSED 2026-06-07. Re-trigger landed after rebuild-cache.sh: redirect TMPDIR to $WORK_DIR — converter CSVs land in tmpfs (~2 GB RAM cap) not EBS #271/fix(rebuild-cache): redirect TMPDIR to EBS so converter CSVs do not fill tmpfs #272 in main.
• ephemeral-rebuild: 100 GB EBS exhausted at 94% of release scan — bump volume or split /tmp #268 (100 GB EBS) — CLOSED 2026-06-07.
• rebuild-cache.sh:153 ERR trap clobbers $? via rm -rf — silently exits 0 on pipeline failure #269 (ERR-trap clobbers $?) — CLOSED 2026-06-07 (PR fix(rebuild-cache): propagate real exit code through ERR-trap cleanup #270).
• rebuild-cache.sh: redirect TMPDIR to $WORK_DIR — converter CSVs land in tmpfs (~2 GB RAM cap) not EBS #271 (TMPDIR → EBS) — CLOSED 2026-06-07 (PR fix(rebuild-cache): redirect TMPDIR to EBS so converter CSVs do not fill tmpfs #272).
• cache_metadata missing UNIQUE(release_id) — populate_cache_metadata ON CONFLICT fails on load #273 (cache_metadata UNIQUE) — CLOSED 2026-06-07.

LML downstream consumers all closed (post-jumpstart):

• Populate api_fetch-only artist columns during cache rebuild (profile, urls, members, name_variations) library-metadata-lookup#497 — CLOSED. artist.profile, artist_alias, artist_name_variation, artist_member, artist_url populated; ?refresh=true per-entity escape hatch live.
• Rotation-scoped backfill: refresh artist cache rows for current-rotation releases library-metadata-lookup#501 — CLOSED (rotation-scoped backfill ran).
• Populate artist_name_variation during cache rebuild (not only runtime api_fetch) library-metadata-lookup#356 — CLOSED (ANV populated from dump; [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229's natural overlap retired).
• validate_track_on_release should filter release_track_artist to extra = 0 library-metadata-lookup#333 — CLOSED (release_track_artist extra=0 filter shipped).

That removes one cross-cluster conflict from the #263 body (#229 ↔ LML#356) and one cross-repo runtime risk (LML#333). The remaining runtime risk is WXYC/library-metadata-lookup#476 (schema-skew cool-down) and WXYC/discogs-xml-converter#61 (sub_tracks modelling) — both still OPEN; both still relevant to #229 / #217.

New issue filed today:

• rebuild-cache-bootstrap.sh: skip the discogs-xml-converter clone+build when prebuilt download will succeed #275 — cold-start bootstrap waste: rebuild-cache-bootstrap.sh:179-183 clones+builds the converter that rebuild-cache.sh:110-113 then ignores in the happy path. LOW severity (wall-clock only, no correctness issue), XS-S complexity. Surfaced during the Manual jumpstart rebuild to land #497 fields ahead of 2026-07-04 cron #267 jumpstart logs. Not blocking anything; fits into the post-jumpstart cleanup tier alongside Audit: rebuild release count below expected pair-wise band (38,662 vs 50–80K) #217's count audit.

Unchanged P0 carries forward:

The substance of the #263 body still holds with these corrections:

1. Repair or close+refile check_cache_drift.py: assert artist_name_variation populated + release_track_artist.extra column present #226 / rebuild-cache.sh hygiene: non-zero exit on wait_for_postgres timeout + mask DSN in import_csv.py log output #227 / rebuild-cache.sh: fetch + process all four Discogs dump files (artists, labels, masters in addition to releases) #228 / [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229. Now 19 days uninterpolated since 2026-05-20 filing. Confirmed today: bodies still literal @/tmp/parent.md / @/tmp/child-b.md / @/tmp/child-c.md / @/tmp/child-d.md. With LML#356 now closed, [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229's external pressure is reduced — but the four issues continue to mis-represent ~15% of the open queue.
2. Resolve release_artist.role schema drift: drop the column or wire the converter #210 Option A still ready to ship. schema/create_database.sql:82 still holds the dead role column; no consumer references it.
3. Pick DuckDB + extension versions; create plans/duckdb-version-pin.md in WXYC/wiki #189 still first-mover gate for the Epic: rewrite verify_cache.py against DuckDB ATTACH (Project 2 candidate 1) #186 DuckDB rewrite chain.
4. [Epic I] Artwork-NULL ambiguity #247 (Epic I): all 8 Scope rows closed per the 2026-06-03 in-thread comments; only acceptance criterion 3 (post-drain artwork_null_share baseline) outstanding. Recommend close once baseline is back-published or split off into a thin follow-up.
5. PR test: bump charset-torture corpus pin to v0.12.0 #150 — still OPEN, CI green (5/5 SUCCESS), last touched 2026-06-03. Needs rebase + merge.

Current open queue (26): #50, #55, #59, #96, #184, #186, #189, #190, #191, #192, #193, #202, #208, #209, #210, #212, #213, #217, #226, #227, #228, #229, #238 (prior tracker; can close), #247, #263 (this tracker), #275.

Recommend-close in addition to prior list: #238 (superseded by #263 — explicit in #263's body).

[jakebromberg]

Tracker body refreshed in place 2026-06-10. Title bumped to reflect the current 26-open count and the 2026-06-10 cut.

What shifted since the 2026-06-03 cut:

• 2026-06-07 jumpstart rebuild (Manual jumpstart rebuild to land #497 fields ahead of 2026-07-04 cron #267) closed; landed artist.profile, artist_alias, artist_name_variation, artist_member, artist_url populations from the dump for the first time.
• Seven issues closed (Add PG integration test for populate_cache_metadata's ON CONFLICT race-tolerance #207, Rebuild fetches artists.xml.gz; direct-pg path loads artist details #264, Manual jumpstart rebuild to land #497 fields ahead of 2026-07-04 cron #267, ephemeral-rebuild: 100 GB EBS exhausted at 94% of release scan — bump volume or split /tmp #268, rebuild-cache.sh:153 ERR trap clobbers $? via rm -rf — silently exits 0 on pipeline failure #269, rebuild-cache.sh: redirect TMPDIR to $WORK_DIR — converter CSVs land in tmpfs (~2 GB RAM cap) not EBS #271, cache_metadata missing UNIQUE(release_id) — populate_cache_metadata ON CONFLICT fails on load #273) — mostly rebuild hygiene + integration test landing alongside the jumpstart.
• LML#221 + LML#356 + LML#497 all closed; Epic I ([Epic I] Artwork-NULL ambiguity #247) is end-to-end ✅; cross-repo ANV runtime risk resolved.
• New issue rebuild-cache-bootstrap.sh: skip the discogs-xml-converter clone+build when prebuilt download will succeed #275 (bootstrap clone+build redundancy).
• New cross-repo dep: LML#510 will require an alembic migration here for a not_found discriminator column on get_artist_details / get_release rows.
• [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229 epic's scope shrank ~50% — artists dump now in tree; residual is labels+masters only.
• Per-issue status comments dated 2026-06-10 posted on check_cache_drift.py: assert artist_name_variation populated + release_track_artist.extra column present #226, rebuild-cache.sh hygiene: non-zero exit on wait_for_postgres timeout + mask DSN in import_csv.py log output #227, rebuild-cache.sh: fetch + process all four Discogs dump files (artists, labels, masters in addition to releases) #228, [Epic] Expand monthly cache rebuild to ingest the full Discogs dump (releases + artists + labels + masters) #229, [Epic I] Artwork-NULL ambiguity #247.

[jakebromberg]

Refresh 2026-06-12

Minor delta from the 2026-06-10 snapshot — no triage-reshaping events; primarily closures and one new cross-repo signal worth flagging.

Closed since 2026-06-10 (3)

• docs(migrations): extend per-revision history to cover 0009-0012 #283 (2026-06-11) — docs(migrations): extend per-revision history to cover 0009-0012. Per-revision bullets for 0009 (cache_metadata_unique), 0010 (release.not_found), 0011 (artist.not_found), 0012 (entity.release_identity adoption) now in docs/migrations.md.
• docs(migrations-runbook): out-of-band single-migration deploy procedure #281 (2026-06-11) — docs(migrations-runbook): out-of-band single-migration deploy procedure. Documents the single-migration out-of-band deploy path used to ship 0012 ahead of the next monthly rebuild (gated on LML#530).
• Adopt entity.identity / entity.reconciliation_log into alembic chain #279 (2026-06-12) — Adopt entity.identity / entity.reconciliation_log into alembic chain. Revision 0013 (adopt_entity_identity) lands in alembic chain; the two tables that previously existed only in prod are now declared in tree.

Net: alembic chain advanced from 0008 (the 2026-06-03 cut) to 0013. Five revisions in nine days; all behind the rebuild-hygiene + LML#510/#525/#526 cluster.

Cross-repo dependencies — status update

• LML#510 (Tombstone 404s) — RESOLVED on the discogs-etl side: 0010 (release.not_found) + 0011 (artist.not_found) are in tree and were deployed out-of-band per the docs(migrations-runbook): out-of-band single-migration deploy procedure #281 runbook. The cross-repo dep flagged in [Tracker] 2026-06-10 issue triage — 26 open #263's body is satisfied; LML#510 close-out is now an LML-side decision.
• LML#525 (POST /api/v1/cache/refresh-for-identities) — alembic 0012 (entity.release_identity) + 0013 (adopt_entity_identity) deployed. Discogs-cache substrate ready; remaining work is LML-side.
• LML#526 (release-level identity) — substrate landed via 0012 + 0013. No further discogs-etl asks visible.

New cross-repo signal — LML#537 (2026-06-12)

investigate: Discogs cache hit ratio plateaued at ~50% (search_releases_by_track at 34%) after write_release fix — H1 (ETL coverage gap) is confirmed and explicitly names the structural cause. search_releases_by_track has pg_write=None by design (discogs/fallthrough.py:23); the only path that warms release_track is get_release → write_release, which only runs when the orchestrator calls validate_track_on_release on a candidate. The 34% hit ratio is downstream-of-validation, not a search-cache property. Not blocking for this repo today, but if LML#537 lands a "warm the track index from search results" proposal, it will likely surface a discogs-etl ask (broader periodic warm-up of release_track from the dump, or a synthetic backfill driven by recent search misses). Watch.

#247 (Epic I — Artwork-NULL ambiguity) — still open, still recommended for close

No movement since the 2026-06-10 status comment. Every Scope row remains ✅; either back-publish the post-drain artwork_null_share measurement and close, or close now and file a thin follow-up for the published-baseline record. Highest-leverage cheap close in the queue.

PR #150 — no change

mergeStateStatus: UNKNOWN, no activity since 2026-06-03. Still needs rebase + merge.

Severity / complexity snapshot (unchanged from 2026-06-10)

No new HIGH or CRITICAL issues. No new issues created in the last 7 days (last was #275 on 2026-06-07). Severity ladder unchanged: stub-body cluster (#226/#227/#228/#229) still the highest pending judgement, #247 close-out the highest-leverage cheap fix.