Viktor Barzin
a42944a756
wrongmove: round-3 fix sweep — scrape pipeline, BUY tab, filter URL state, render hygiene, map polish
...
Coordinated fix across 31 bugs found in a parallel QA pass. Findings docs at /tmp/wrongmove-bugs/qa-round-3/qa{1,2,3,4}-*.md.
## Backend / scrape (Fix-1) — 8 bugs
- B1 [P0] Scrape totally broken on prod: pod UID 100 vs NFS dir 1000:1000 mode 775 → PermissionError on every never-seen listing. Switched Dockerfile to explicit `useradd --uid 1000 --gid 1000`; added securityContext + chown initContainer to k8s/{api,celery-beat}-deployment.yaml. Celery worker manifest lives outside this repo — Dockerfile UID change is the load-bearing fix.
- B4 [P1] Celery broker reaped every ~30s by Redis HAProxy idle timeout. Added `broker_transport_options` / `result_backend_transport_options` with `socket_keepalive=True, health_check_interval=25` in celery_app.py + same kwargs on every redis.from_url/Redis call across services/, utils/redis_lock.py, redis_repository.py.
- B5 [P1] dump_listings_task never published terminal FAILURE to the task_progress pub/sub channel — UI polled forever. Wrap body in try/except that publishes FAILURE before re-raising.
- B6 [P1] _process_worker had no per-listing exception handler — one bad listing killed the whole scrape via asyncio.gather. Wrap loop body in try/except Exception (re-raises CancelledError).
- B20 [P2] dump_listings_task gained time_limit=3600, soft_time_limit=3500, acks_late=True.
- B21 [P2] RedisRepository moved off shared db0 (was alongside paperless-ngx) to db3 via REDIS_USER_DB env var; keys prefixed `wrongmove:user:`.
- B32 [P3] redis_lock now uses uuid4() owner token + Lua compare-and-delete.
- B33 [P3] Slack notify in refresh_listings → asyncio.create_task (fire-and-forget).
## Frontend filter system (Fix-2) — 7 bugs
- B2 [P0] BUY tab click triggered "Maximum update depth exceeded" → ErrorBoundary. Replaced the three mutually-triggering useEffects in FilterBar with a single one-way controlled-value flow (URL → parent state → form), guarded by previousListingTypeRef so price-defaults fires once per real transition.
- B3 [P0] Filter values never reached the URL. Wired useFilterParams.setFilterValues into FilterBar/FilterPanel onSubmit + handleRemoveChip + new handleResetAllFilters; fed parsed filterValues into both forms' defaultValues; added URL→form sync via form.reset on browser back/forward.
- B8 [P1] Chip removal now resets form state via new FilterBar onFormReady callback — More badge no longer sticks.
- B12 [P2] Desktop swipe-review FAB added next to header (mobile FAB unchanged).
- B17 [P2] "Reset all" affordance on chip strip.
- B22 [P2] formatPrice precision: 1500 → £1.5k, 2500 → £2.5k (no longer collides with £2k/£3k defaults).
- B30 [P3] last_seen_days input gained min={0}.
## Frontend render hygiene + data integrity (Fix-3) — 8 bugs
- B7 [P1] streamingService bails on first non-NDJSON chunk (HTML response = backend down) and throws StreamParseError so the existing AlertError dialog surfaces a single user-visible error instead of 18× console.error spam.
- B9 [P1] formatDuration widened to (null|undefined|number): returns "—" for non-finite or negative, caps implausibly large values.
- B10 [P1] PropertyCard / PropertyCardCompact / SwipeCard JSX leaves render "—" for null total_price/qm/qmprice (was "£0/0 m²/£0/m²" — looked like free listings).
- B13 [P2] hexgrid worker reduceAverage uses Number.isFinite filter instead of !isNaN (which incorrectly accepted null → 0, biasing per-hex averages low).
- B14 [P2] ListingDetail Overview wraps agency in "Listed by" labelled block so it can't collapse to a bare agency name.
- B15 [P2] Compact POIDistanceBadges iterates all three travel modes with "—" for missing, matching the detail-sheet Travel table.
- B24 [P3] Drawer.Description (sr-only) added to ListingDetailSheet + MobileBottomSheet to silence Radix a11y warning.
- B25 [P3] lastSeenDays clamped to ≥0 so future timestamps don't render as "-7d ago".
## Frontend map / carousel / tasks polish (Fix-4) — 8 bugs
- B11 [P2] HexgridHeatmapClient destroy race: Map.tsx adds .catch() + ref guard so post-destroy promise rejections are silent no-ops. Verified by browser smoke (24 rapid Map↔List toggles → 0 pageErrors).
- B16 [P2] PhotoCarousel + inner CardCarousel gained keyboard nav (Arrow keys).
- B18 [P2] Default map center moved from Czech Republic to London (zoom 10).
- B19+B29 [P2/P3] Mapbox token: no longer hard-coded fallback; reads env-only and shows a clear "Map unavailable — set VITE_MAPBOX_TOKEN" banner when missing.
- B23 [P3] PhotoCarousel suppresses "1/1" counter for single-photo listings; added onError fallback for broken URLs.
- B26 [P3] PhotoCarousel only enables loop when photos.length > 1.
- B27 [P3] TaskIndicator cancel/clear-all buttons gained aria-label + data-testid.
- B28 [P3] useTaskProgress strips terminal-local task IDs from the polling union — no more forever-poll on completed tasks.
## Tests
74 new vitest tests + 18 new pytest tests. Local: tsc clean, 201 vitest tests pass, 633 pytest tests pass.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-10 22:27:29 +00:00
Viktor Barzin
67d4ab3821
Fix celery metrics not reaching Prometheus and update Grafana dashboard
...
Init OTel metrics at module level in celery_app.py so prefork child
processes inherit the MeterProvider and PrometheusMetricReader from
the parent. Previously, worker_process_init created a separate
MeterProvider in each child, disconnected from the HTTP server in
the main process — so all scrape/celery/OCR metrics were silently
lost.
Update Grafana dashboard with API Performance and Frontend Performance
sections, synced from the live cluster dashboard.
2026-02-22 17:58:20 +00:00
Viktor Barzin
f6f0aa8d14
Fix metrics not initialized in Celery prefork worker processes
...
worker_ready fires in the main process after pool workers are forked,
so init_metrics() never ran in the child processes. Add a
worker_process_init handler to initialize OTel metrics in each worker.
2026-02-15 12:58:59 +00:00
Viktor Barzin
25912eac0c
Fix metric imports: use module-level access instead of name imports
...
Type-annotated metric variables (e.g. `geojson_cache_operations: Counter`)
don't exist as importable names until init_metrics() runs. Switch all
`from api.metrics import <metric>` to `import api.metrics as m` and
access instruments as attributes at runtime to avoid ImportError.
2026-02-14 11:21:49 +00:00
Viktor Barzin
d6edb747d2
Add structured JSON logging, OTel business metrics, and Grafana dashboard
...
Structured logging via JsonFormatter replaces uvicorn's default format so
Loki can parse timestamps and fields. 14 business metrics (scrape stats,
throttle events, circuit breaker state, cache hit rate, OCR success rate,
Celery task lifecycle) are defined in a shared metrics module and
instrumented across the scraper pipeline, API, and workers. Celery
workers expose a Prometheus HTTP endpoint on configurable ports.
2026-02-14 10:59:12 +00:00
Viktor Barzin
bd788df9aa
Add POI API routes and Celery task
...
FastAPI router with CRUD endpoints for POIs, distance calculation
trigger, and distance queries. Streaming GeoJSON endpoint now accepts
include_poi_distances=true to inject travel times into features.
Celery task wraps the distance calculator with progress reporting.
2026-02-08 13:14:47 +00:00
Viktor Barzin
eafbc1ac52
Flatten repo structure: move crawler/ to root, remove vqa/ and immoweb/
...
The crawler subdirectory was the only active project. Moving it to the
repo root simplifies paths and removes the unnecessary nesting. The
vqa/ and immoweb/ directories were legacy/unused and have been removed.
Updated .drone.yml, .gitignore, .claude/ docs, and skills to reflect
the new flat structure.
2026-02-07 23:01:20 +00:00
