ON CONFLICT now only replaces score/factors when the score actually changed,
avoiding dead tuple generation for the ~80% of grid points that don't change
between consecutive HRRR runs.
Migration sets aggressive autovacuum on propagation_scores: zero cost delay,
2000 cost limit, 1% scale factor. The table was at 72GB with 108M dead rows
because default autovacuum couldn't keep pace with 14M upserts per hour.
- Only deduplicate against active jobs (available/scheduled/executing/
retryable), not completed ones. Completed jobs no longer block new
inserts for the same args.
- Period reduced from 3600s to 300s
- Fixes contacts stuck in :queued with no actual Oban jobs
- Database triggers on contacts and oban_jobs fire pg_notify on
status/state changes
- RepoListener GenServer subscribes via Postgrex.Notifications and
broadcasts to PubSub
- Backfill dashboard subscribes to PubSub instead of 2-second polling
- Eliminates periodic SELECT queries, updates arrive instantly
- Unified queue_info helper shows "(N jobs in queue)" on all spinners
- Solar index auto-enqueues when missing on contact page
- Freshness monitor disabled in dev
- Fix test stubs for solar client
- SolarIndexWorker accepts date arg for specific date fetch
- Contact show page enqueues solar fetch when data missing
- PubSub broadcast on solar data arrival, live-updates page
- Spinner shown while fetching
- Migration renames table, column, all indexes, and FK constraints
- Contact schema now references "contacts" table
- TerrainProfile foreign key renamed from qso_id to contact_id
- All code and tests updated to use contact_id
Previously, if a job was cancelled or lost while status was :queued,
viewing the contact detail page would not re-enqueue. Now HRRR,
weather/soundings, and terrain all re-enqueue when :queued with no
data, relying on Oban uniqueness for deduplication.
The 6 sequential count(*) queries on large tables were timing out
at 15s when concurrent HRRR batch inserts held locks. Reuse the
n_live_tup estimates already fetched from pg_stat_user_tables.
Shows provenance of all data used in the analysis:
- HRRR: path profile count, pressure levels, valid time, grid point
- Elevation: sample count, path distance, source (SRTM), duct layers
- Terrain: sample count, verdict, max elevation, diffraction loss
- Surface observations: station count and time range
- Soundings: profile count, stations, levels, ducting detection
- Solar indices: SFI, SSN, Ap
- Enrichment status breakdown per type (HRRR/Weather/Terrain/IEMRE)
with color-coded status counts (complete/pending/queued/failed)
- Database overview: total DB size, HRRR profiles, observations,
soundings with per-contact averages
- Table sizes: rows and disk size per table, sorted by size
- All stats auto-refresh every 2 seconds
Collapsible control panel with hamburger toggle, detail panel as
bottom sheet, compact legend moved to top-right, and scaled timeline
buttons for narrow viewports.
