Parallelize GRIB2 downloads using a dedicated rayon thread pool

Downloads clat/clon concurrently via rayon::join and all forecast steps
in parallel via par_iter, then renders frames sequentially. A separate
thread pool (up to 16 threads) is used for downloads to avoid blocking
the global pool during CPU-bound rendering.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/main.rs

@@ -6,6 +6,7 @@ mod render;
 use anyhow::{bail, Context, Result};
 use chrono::{DateTime, Datelike, Timelike, Utc};
 use clap::Parser;
+use rayon::prelude::*;
 use std::path::PathBuf;
 
 const BASE_URL: &str = "https://opendata.dwd.de/weather/nwp/icon-d2/grib/";
@@ -77,26 +78,22 @@ fn main() -> Result<()> {
         dt.hour()
     );
 
-    // Download & decode coordinate grids
+    // Build a dedicated thread pool for I/O-bound downloads, separate from
+    // the global rayon pool used for CPU-bound rendering.
+    let num_download_threads = (args.prediction_hours as usize + 2).min(16);
+    let download_pool = rayon::ThreadPoolBuilder::new()
+        .num_threads(num_download_threads)
+        .build()
+        .context("Failed to build download thread pool")?;
+
+    // Download coordinate grids (clat / clon) concurrently.
     eprintln!("Downloading coordinate grids (clat / clon)...");
-    let lats = download_and_decode(
+    let (lats, lons) = download_pool.join(
-        &run_url,
+        || download_and_decode(&run_url, &date_str, "clat", "time-invariant", None, &cache_dir).context("clat"),
-        &date_str,
+        || download_and_decode(&run_url, &date_str, "clon", "time-invariant", None, &cache_dir).context("clon"),
-        "clat",
+    );
-        "time-invariant",
+    let lats = lats?;
-        None,
+    let lons = lons?;
-        &cache_dir,
-    )
-    .context("clat")?;
-    let lons = download_and_decode(
-        &run_url,
-        &date_str,
-        "clon",
-        "time-invariant",
-        None,
-        &cache_dir,
-    )
-    .context("clon")?;
 
     if lats.len() != lons.len() {
         bail!(
@@ -107,11 +104,12 @@ fn main() -> Result<()> {
     }
     eprintln!("  Grid has {} points.", lats.len());
 
-    // Download, decode, and render each forecast hour
+    // Download all forecast steps in parallel, then render sequentially.
-    eprintln!("Rendering {} frame(s)...", args.prediction_hours);
+    eprintln!("Downloading {} forecast frame(s) in parallel...", args.prediction_hours);
-    let mut output_paths = Vec::new();
+    let mut cloud_data: Vec<(u32, Vec<f32>)> = download_pool.install(|| {
-
+        (0..args.prediction_hours)
-    for step in 0..args.prediction_hours {
+            .into_par_iter()
+            .map(|step| {
                 let cloud = download_and_decode(
                     &run_url,
                     &date_str,
@@ -121,7 +119,16 @@ fn main() -> Result<()> {
                     &cache_dir,
                 )
                 .with_context(|| format!("clct step {}", step))?;
+                Ok::<_, anyhow::Error>((step, cloud))
+            })
+            .collect::<Result<Vec<_>>>()
+    })?;
+    cloud_data.sort_unstable_by_key(|(step, _)| *step);
 
+    eprintln!("Rendering {} frame(s)...", args.prediction_hours);
+    let mut output_paths = Vec::new();
+
+    for (step, cloud) in cloud_data {
         if cloud.len() != lats.len() {
             bail!(
                 "clct step {} has {} values but coordinate grids have {}",