Full cutover: the propagation_scores Postgres table is gone, and the binary files under /data/scores are the sole source of truth for the map render path. Three stacked changes: 1. New migration drops the propagation_scores table and its indexes (the earlier tuning migrations for it were already applied and are now no-ops against a missing table, which is fine — Ecto just runs them on fresh environments). 2. Propagation context is gutted of every GridScore reference. replace_scores/2 writes files only. upsert_scores/2 is deleted. load_scores_from_db, available_valid_times_from_db, point_detail_from_db, point_forecast_from_db, fetch_factors, coalesce_factors, the Postgres side of prune_old_scores, and the Postgres fallbacks in latest/earliest_valid_time are all removed. point_detail always returns an empty factors map now since factor breakdowns were retired with the table. 3. Deleted modules: - Propagation.GridScore (the schema) - Propagation.ScorerDiff (read factors from the table) - Propagation.AsosNudge (helper for AsosAdjustmentWorker) - Workers.AsosAdjustmentWorker (its cron was already disabled) - Mix.Tasks.ScorerDiff (wrapper around the deleted module) And their tests. AdminTaskWorker's scorer_diff task is a logging no-op so any queued Oban rows drain cleanly. Release.scorer_diff stays as a stub that tells the operator. propagation_prune_worker_test rewritten to exercise ScoresFile pruning. propagation_test.exs rewritten to use replace_scores + ScoresFile throughout (no more GridScore / upsert_scores paths).
227 lines
8 KiB
Elixir
227 lines
8 KiB
Elixir
defmodule Microwaveprop.Workers.AdminTaskWorker do
|
|
@moduledoc """
|
|
Oban worker for long-running admin tasks (backtests, climatology, etc).
|
|
|
|
Enqueued via `Microwaveprop.Release` so that `eval` returns immediately
|
|
and the work runs in the normal Oban pipeline without blocking.
|
|
|
|
Each task type is dispatched by the `"task"` arg:
|
|
|
|
Microwaveprop.Workers.AdminTaskWorker.new(%{task: "backtest_all"})
|
|
Microwaveprop.Workers.AdminTaskWorker.new(%{task: "backtest", feature: "naive_gradient"})
|
|
Microwaveprop.Workers.AdminTaskWorker.new(%{task: "climatology", min_samples: 3})
|
|
Microwaveprop.Workers.AdminTaskWorker.new(%{task: "native_derive", limit: 10000})
|
|
"""
|
|
use Oban.Worker, queue: :admin, max_attempts: 1, unique: [period: 60]
|
|
|
|
import Ecto.Query
|
|
|
|
alias Microwaveprop.Backtest
|
|
alias Microwaveprop.Backtest.Features
|
|
alias Microwaveprop.Propagation.Duct
|
|
alias Microwaveprop.Propagation.Inversion
|
|
alias Microwaveprop.Propagation.Recalibrator
|
|
alias Microwaveprop.Repo
|
|
alias Microwaveprop.Weather.HrrrNativeProfile
|
|
alias Microwaveprop.Weather.ThetaE
|
|
|
|
require Logger
|
|
|
|
@impl Oban.Worker
|
|
def perform(%Oban.Job{args: %{"task" => "backtest_all"} = args}) do
|
|
sample_size = Map.get(args, "sample_size", 5000)
|
|
features = Features.all_features()
|
|
|
|
Logger.info("AdminTask: running consolidated backtest for #{map_size(features)} features")
|
|
|
|
results = Backtest.consolidated_report(features, sample_size: sample_size)
|
|
markdown = Backtest.to_consolidated_markdown(results)
|
|
|
|
path = "priv/backtest_reports/consolidated.md"
|
|
File.mkdir_p!(Path.dirname(path))
|
|
File.write!(path, markdown)
|
|
|
|
Logger.info("AdminTask: consolidated backtest complete, wrote #{path}")
|
|
:ok
|
|
end
|
|
|
|
def perform(%Oban.Job{args: %{"task" => "backtest", "feature" => feature_name} = args}) do
|
|
sample_size = Map.get(args, "sample_size", 5000)
|
|
fun_atom = String.to_atom(feature_name)
|
|
|
|
if !function_exported?(Features, fun_atom, 3) do
|
|
Logger.error("AdminTask: unknown feature #{feature_name}")
|
|
{:error, "unknown feature: #{feature_name}"}
|
|
end
|
|
|
|
feature_fun = &apply(Features, fun_atom, [&1, &2, &3])
|
|
name = "Microwaveprop.Backtest.Features.#{feature_name}"
|
|
|
|
Logger.info("AdminTask: running backtest for #{feature_name}")
|
|
|
|
report = Backtest.evaluate(feature_fun, sample_size: sample_size, feature_name: name)
|
|
distance_bins = Backtest.lift_by_distance(feature_fun, sample_size: sample_size)
|
|
band_stats = Backtest.lift_by_band(feature_fun, sample_size: sample_size)
|
|
markdown = Backtest.to_markdown(report, distance_bins: distance_bins, band_stats: band_stats)
|
|
|
|
path = "priv/backtest_reports/#{feature_name}.md"
|
|
File.mkdir_p!(Path.dirname(path))
|
|
File.write!(path, markdown)
|
|
|
|
Logger.info("AdminTask: backtest for #{feature_name} complete, wrote #{path}")
|
|
:ok
|
|
end
|
|
|
|
def perform(%Oban.Job{args: %{"task" => "climatology"} = args}) do
|
|
min_samples = Map.get(args, "min_samples", 3)
|
|
|
|
%{rows: combos} =
|
|
Repo.query!(
|
|
"""
|
|
SELECT EXTRACT(MONTH FROM valid_time)::int AS month,
|
|
EXTRACT(HOUR FROM valid_time)::int AS hour
|
|
FROM hrrr_profiles
|
|
WHERE surface_temp_c IS NOT NULL AND is_grid_point = true
|
|
GROUP BY 1, 2
|
|
ORDER BY 1, 2
|
|
""",
|
|
[],
|
|
timeout: 120_000
|
|
)
|
|
|
|
Logger.info("AdminTask: building climatology (min_samples=#{min_samples}, #{length(combos)} batches)")
|
|
|
|
total =
|
|
combos
|
|
|> Enum.with_index(1)
|
|
|> Enum.reduce(0, fn {[month, hour], idx}, acc ->
|
|
%{num_rows: count} =
|
|
Repo.query!(
|
|
"""
|
|
INSERT INTO hrrr_climatology (id, lat, lon, month, hour,
|
|
mean_surface_temp_c, stddev_surface_temp_c, sample_count)
|
|
SELECT gen_random_uuid(), lat, lon,
|
|
$2 AS month,
|
|
$3 AS hour,
|
|
AVG(surface_temp_c),
|
|
STDDEV_SAMP(surface_temp_c),
|
|
COUNT(*)
|
|
FROM hrrr_profiles
|
|
WHERE surface_temp_c IS NOT NULL
|
|
AND is_grid_point = true
|
|
AND EXTRACT(MONTH FROM valid_time)::int = $2
|
|
AND EXTRACT(HOUR FROM valid_time)::int = $3
|
|
GROUP BY lat, lon
|
|
HAVING COUNT(*) >= $1
|
|
ON CONFLICT (lat, lon, month, hour)
|
|
DO UPDATE SET
|
|
mean_surface_temp_c = EXCLUDED.mean_surface_temp_c,
|
|
stddev_surface_temp_c = EXCLUDED.stddev_surface_temp_c,
|
|
sample_count = EXCLUDED.sample_count
|
|
""",
|
|
[min_samples, month, hour],
|
|
timeout: 300_000
|
|
)
|
|
|
|
if rem(idx, 10) == 0, do: Logger.info("AdminTask: climatology [#{idx}/#{length(combos)}]")
|
|
acc + count
|
|
end)
|
|
|
|
Logger.info("AdminTask: climatology complete, upserted #{total} records")
|
|
:ok
|
|
end
|
|
|
|
def perform(%Oban.Job{args: %{"task" => "native_derive"} = args}) do
|
|
limit = Map.get(args, "limit", 10_000)
|
|
|
|
profiles =
|
|
HrrrNativeProfile
|
|
|> where([p], is_nil(p.bulk_richardson) and p.level_count > 2)
|
|
|> limit(^limit)
|
|
|> Repo.all()
|
|
|
|
Logger.info("AdminTask: deriving fields for #{length(profiles)} native profiles")
|
|
|
|
count =
|
|
Enum.count(profiles, fn profile ->
|
|
p = %{
|
|
heights_m: profile.heights_m,
|
|
temp_k: profile.temp_k,
|
|
spfh: profile.spfh,
|
|
pressure_pa: profile.pressure_pa,
|
|
u_wind_ms: profile.u_wind_ms,
|
|
v_wind_ms: profile.v_wind_ms,
|
|
tke_m2s2: profile.tke_m2s2,
|
|
level_count: profile.level_count
|
|
}
|
|
|
|
inversion_fields =
|
|
case Inversion.find_inversion_top(p) do
|
|
{:ok, %{height_m: top_h, level_idx: top_idx, base_idx: base_idx}} ->
|
|
[
|
|
inversion_top_m: top_h,
|
|
bulk_richardson: Inversion.bulk_richardson(p, base_idx, top_idx),
|
|
shear_at_top_ms: Inversion.shear_magnitude(p, base_idx, top_idx),
|
|
theta_e_jump_k: ThetaE.theta_e_jump(p, base_idx, top_idx)
|
|
]
|
|
|
|
:none ->
|
|
[inversion_top_m: nil, bulk_richardson: nil, shear_at_top_ms: nil, theta_e_jump_k: nil]
|
|
end
|
|
|
|
duct_result = Duct.analyze(p)
|
|
fields = inversion_fields ++ [ducts: duct_result.ducts, best_duct_band_ghz: duct_result.best_duct_band_ghz]
|
|
|
|
{1, _} =
|
|
HrrrNativeProfile
|
|
|> where([pr], pr.id == ^profile.id)
|
|
|> Repo.update_all(set: fields)
|
|
|
|
true
|
|
end)
|
|
|
|
Logger.info("AdminTask: derived fields for #{count} profiles")
|
|
:ok
|
|
end
|
|
|
|
def perform(%Oban.Job{args: %{"task" => "recalibrate"} = args}) do
|
|
sample_size = Map.get(args, "sample_size", 5000)
|
|
epochs = Map.get(args, "epochs", 2000)
|
|
learning_rate = Map.get(args, "learning_rate", 0.01)
|
|
|
|
Logger.info("AdminTask: running weight recalibration (sample=#{sample_size}, epochs=#{epochs}, lr=#{learning_rate})")
|
|
|
|
result =
|
|
Recalibrator.fit(
|
|
sample_size: sample_size,
|
|
epochs: epochs,
|
|
learning_rate: learning_rate
|
|
)
|
|
|
|
Logger.info("AdminTask: recalibration complete")
|
|
|
|
Logger.info(" train_loss=#{result.train_loss}, val_loss=#{result.val_loss}, initial_loss=#{result.initial_loss}")
|
|
|
|
for {factor, weight} <- Enum.sort(result.weights) do
|
|
Logger.info(" #{factor}: #{weight}")
|
|
end
|
|
|
|
:ok
|
|
end
|
|
|
|
def perform(%Oban.Job{args: %{"task" => "scorer_diff"}}) do
|
|
# scorer_diff was a calibration tool that read the per-cell
|
|
# factors JSONB from the propagation_scores table. That table is
|
|
# gone — factors are no longer persisted across the CONUS grid —
|
|
# so the diff has nothing to compute against. Left as a no-op so
|
|
# any in-flight Oban rows from the pre-cutover era drain cleanly.
|
|
Logger.warning("AdminTask: scorer_diff is disabled — propagation_scores table + factors storage have been dropped")
|
|
|
|
:ok
|
|
end
|
|
|
|
def perform(%Oban.Job{args: %{"task" => task}}) do
|
|
Logger.error("AdminTask: unknown task #{task}")
|
|
{:error, "unknown task: #{task}"}
|
|
end
|
|
end
|