feat: remove skippy HRRR proxy + disk cache, fetch directly from NOAA S3

- Drop skippy.w5isp.com:8080 as HRRR_BASE_URL in dev/prod config and
  all K8s manifests. Default to direct NOAA S3 reads.
- Strip hrrr_cache_dir disk-caching infrastructure from HrrrClient
  (read_cache, write_cache, download_and_cache_grib_ranges). Files are
  now fetched fresh each time — no local filesystem cache.
- Update stale skippy references in network-policy, test comments, docs.

config/dev.exs

@@ -117,10 +117,7 @@ config :microwaveprop, :propagation_scores_dir, Path.expand("priv/dev_scores", F
 # Enable dev routes for dashboard and mailbox
 config :microwaveprop, dev_routes: true
 
-# HRRR caching handled by nginx proxy on skippy — no local filesystem cache
-
-# Proxy HRRR downloads through local caching nginx (set to nil to use S3 directly)
-config :microwaveprop, hrrr_base_url: "http://skippy.w5isp.com:8080"
+# HRRR fetched directly from NOAA S3 — no local filesystem cache
 
 # Load ML model at startup (Nx/Axon/EXLA only available in dev/test)
 config :microwaveprop, load_ml_model: true

config/runtime.exs

@@ -492,7 +492,7 @@ if config_env() == :prod do
   # replica. Queue depth is better observed from the oban_web
   # dashboard anyway.
   config :microwaveprop, :prom_ex_include_oban_plugin, false
-  config :microwaveprop, hrrr_base_url: System.get_env("HRRR_BASE_URL", "http://skippy.w5isp.com:8080")
+  config :microwaveprop, hrrr_base_url: System.get_env("HRRR_BASE_URL", "https://noaa-hrrr-bdp-pds.s3.amazonaws.com")
   config :microwaveprop, srtm_tiles_dir: "/data/srtm"
   config :microwaveprop, start_freshness_monitor: false
 end

docs/plans/2026-04-13-rdps-vertical-profiles.md

@@ -302,6 +302,6 @@ Do NOT push to `dokku`. Let Flux roll out the new image from the CI build.
 ## Notes / risks
 
 - **Rotated lat/lon projection**: wgrib2 handles it correctly for point extraction but caution needed if the team ever decides to store full-grid binary rasters.
-- **File size**: RDPS pressure-level files are ~50 MB each. At 4 runs/day × ~10 forecast hours × ~5 variables × ~10 pressure levels, total download ≈ 2 GB/day. Not a problem on a home lab with skippy.w5isp.com cache, but worth caching aggressively.
+- **File size**: RDPS pressure-level files are ~50 MB each. At 4 runs/day × ~10 forecast hours × ~5 variables × ~10 pressure levels, total download ≈ 2 GB/day. Worth caching aggressively in a home lab.
 - **Data latency**: RDPS publishes ~90 minutes after synoptic; the cron is already aligned.
 - **Fallback when RDPS is unavailable**: `scores_at/3` should gracefully degrade to HRRR-only if `rdps_profiles` is empty for the requested `valid_time`.

k8s/deployment-backfill.yaml

@@ -67,8 +67,6 @@ spec:
               value: "backfill"
             - name: PHX_HOST
               value: "prop.w5isp.com"
-            - name: HRRR_BASE_URL
-              value: "http://skippy.w5isp.com:8080"
           envFrom:
             - secretRef:
                 name: prop-secrets

k8s/deployment-grid-rs.yaml

@@ -74,12 +74,6 @@ spec:
           env:
             # Elixir secrets bundle carries DATABASE_URL already. Rust
             # reads it directly (sqlx connects with the same URL).
-            - name: HRRR_BASE_URL
-              value: "http://skippy.w5isp.com:8080"
-            # When the LAN caching proxy at skippy is unreachable, fetches
-            # retry once against the NOAA S3 public bucket so a proxy
-            # outage degrades to direct S3 reads instead of stalling the
-            # hourly pipeline.
             - name: HRRR_FALLBACK_BASE_URL
               value: "https://noaa-hrrr-bdp-pds.s3.amazonaws.com"
             - name: PROP_SCORES_DIR

k8s/deployment-hrrr-point-rs.yaml

@@ -67,8 +67,6 @@ spec:
           imagePullPolicy: IfNotPresent
           command: ["/usr/bin/tini", "--", "/usr/local/bin/hrrr-point-worker"]
           env:
-            - name: HRRR_BASE_URL
-              value: "http://skippy.w5isp.com:8080"
             - name: HRRR_FALLBACK_BASE_URL
               value: "https://noaa-hrrr-bdp-pds.s3.amazonaws.com"
             - name: RUST_LOG

k8s/deployment.yaml

@@ -89,8 +89,6 @@ spec:
               value: "prop.w5isp.com"
             - name: PORT
               value: "5000"
-            - name: HRRR_BASE_URL
-              value: "http://skippy.w5isp.com:8080"
           envFrom:
             - secretRef:
                 name: prop-secrets

k8s/network-policy.yaml

@@ -5,7 +5,7 @@
 #   - Erlang distribution (EPMD + ephemeral ports) between prop + backfill pods
 #
 # Egress is unrestricted — pods need to reach Postgres, NFS (10.0.19.103),
-# HRRR (skippy + NOAA S3), and IEM.
+# HRRR (NOAA S3), and IEM.
 apiVersion: networking.k8s.io/v1
 kind: NetworkPolicy
 metadata:

lib/microwaveprop/weather/hrrr_client.ex

@@ -490,8 +490,7 @@ defmodule Microwaveprop.Weather.HrrrClient do
   @doc """
   Downloads the given byte ranges from a GRIB2 URL and concatenates
   them in offset order. Merges adjacent/overlapping ranges and runs
-  requests in parallel. Backed by a disk cache keyed on
-  `(url, ranges)` in dev.
+  requests in parallel.
 
   Public so the native-level client can reuse it rather than
   duplicating the HTTP/parallelism/cache logic.
@@ -509,27 +508,7 @@ defmodule Microwaveprop.Weather.HrrrClient do
   end
 
   defp do_download_grib_ranges(url, ranges) do
-    cache_key = url_to_cache_key(url, ranges)
-
-    case read_cache(cache_key) do
-      {:ok, binary} ->
-        Logger.info("HRRR cache hit: #{cache_key}")
-        {:ok, binary}
-
-      :miss ->
-        download_and_cache_grib_ranges(url, ranges, cache_key)
-    end
-  end
-
-  defp download_and_cache_grib_ranges(url, ranges, cache_key) do
-    case download_grib_ranges_remote(url, ranges) do
-      {:ok, binary} = ok ->
-        write_cache(cache_key, binary)
-        ok
-
-      error ->
-        error
-    end
+    download_grib_ranges_remote(url, ranges)
   end
 
   @doc """
@@ -613,44 +592,6 @@ defmodule Microwaveprop.Weather.HrrrClient do
     end
   end
 
-  defp hrrr_cache_dir, do: Application.get_env(:microwaveprop, :hrrr_cache_dir)
-
-  defp url_to_cache_key(url, ranges) do
-    # e.g. hrrr.20240922_conus_hrrr.t18z.wrfprsf00_r42.grib2
-    uri = URI.parse(url)
-    path_part = uri.path |> String.trim_leading("/") |> String.replace("/", "_")
-    range_hash = ranges |> :erlang.phash2() |> Integer.to_string(16)
-    "#{path_part}_r#{range_hash}.grib2"
-  end
-
-  defp read_cache(key) do
-    case hrrr_cache_dir() do
-      nil ->
-        :miss
-
-      dir ->
-        path = Path.join(dir, key)
-
-        case File.read(path) do
-          {:ok, binary} -> {:ok, binary}
-          {:error, _} -> :miss
-        end
-    end
-  end
-
-  defp write_cache(key, binary) do
-    case hrrr_cache_dir() do
-      nil ->
-        :ok
-
-      dir ->
-        File.mkdir_p!(dir)
-        path = Path.join(dir, key)
-        File.write!(path, binary)
-        Logger.info("HRRR cached: #{path} (#{byte_size(binary)} bytes)")
-    end
-  end
-
   defp merge_ranges(ranges) do
     ranges
     |> Enum.sort_by(&elem(&1, 0))

test/mix/tasks/simple_tasks_test.exs

@@ -309,7 +309,7 @@ defmodule Mix.Tasks.SimpleTasksTest do
   describe "Mix.Tasks.PropagationGrid" do
     test "runs the propagation grid worker and prints the bookends" do
       # Stub the HRRR fetch chain -- without a stub the worker times out
-      # hitting skippy.w5isp.com. Elevation client too for safety.
+      # hitting NOAA S3. Elevation client too for safety.
       Req.Test.stub(HrrrClient, fn conn ->
         Plug.Conn.send_resp(conn, 404, "not found")
       end)