Truncate noisy build-failure output, fix the recurring terrain death

The web UI was showing a raw ~500k-character stdout/stderr dump on a
failed build — go-pmtiles' extract logs one progress line per chunk,
which is fine on a terminal (it overwrites itself) but not captured
non-interactively. command_output now keeps just the tail, and the
status line renders it with preserved line breaks in a scrollable box
instead of one unreadable blob.

Root cause of the actual failure (terrain build dying ~11 minutes in
with an HTTP/2 stream reset): go-pmtiles' default --overfetch=0.05
merged this multi-country planet.pmtiles extract into a single ~60GB
request with no way to resume after a dropped stream. --overfetch=0
splits it into many smaller requests instead.

docs/development.adoc

@@ -71,13 +71,23 @@ and similar, varying by region):
 
 [source,shell]
 ----
-pmtiles extract https://download.mapterhorn.com/planet.pmtiles planet-extract.pmtiles --bbox=west,south,east,north --maxzoom=15
+pmtiles extract https://download.mapterhorn.com/planet.pmtiles planet-extract.pmtiles --bbox=west,south,east,north --maxzoom=15 --overfetch=0
 # only if a regional chunk covers the bbox too (see download_urls.json) —
 # most bboxes hit exactly one, a bbox straddling a chunk boundary hits two:
-pmtiles extract https://download.mapterhorn.com/<chunk>.pmtiles chunk-extract.pmtiles --bbox=west,south,east,north --maxzoom=15
+pmtiles extract https://download.mapterhorn.com/<chunk>.pmtiles chunk-extract.pmtiles --bbox=west,south,east,north --maxzoom=15 --overfetch=0
 pmtiles merge planet-extract.pmtiles chunk-extract.pmtiles data/terrain.pmtiles
 ----
 
+`--overfetch=0`: the default (0.05) merges nearby byte ranges into fewer,
+bigger requests to save round-trips — for `planet.pmtiles` (global z0-12,
+tens of GB) over a bbox spanning several countries this can collapse the
+whole extract into a *single* multi-GB request with no way to resume after
+a mid-transfer HTTP/2 stream reset. `--overfetch=0` trades that one
+fragile mega-request for many small ones instead (spread over
+`--download-threads`, default 4) — slower end to end, but a dropped
+connection only loses one small request's worth of progress, not the
+whole extract.
+
 Tiles are 512px WebP, not the 256px PNG this app's *online* Terrarium
 source uses — `public/js/dem.js`'s own point-elevation lookups
 (`shadow.js`'s terrain-shadow overlay) read the actual decoded tile

lib/maptool/pmtiles_builder.rb

@@ -513,7 +513,17 @@ module Maptool
       extracts = sources.each_with_index.map do |item, i|
         report_step.call(progress_step("extracting #{item["name"]}", labels, i))
         extract_path = File.join(scratch_dir, "extract-#{i}.pmtiles")
-        run_pmtiles!("extract", item["url"], extract_path, "--region=#{region_path}", "--maxzoom=#{TERRAIN_MAX_ZOOM}")
+        # --overfetch=0: go-pmtiles' default (0.05) merges nearby byte
+        # ranges into fewer, bigger HTTP requests to save round-trips —
+        # for planet.pmtiles (global z0-12, tens of GB) covering a
+        # multi-country selection that collapsed the whole extract into
+        # a *single* ~60GB request, which then had no way to resume after
+        # an HTTP/2 stream reset 11 minutes in (see the "died again"
+        # 2026-09-12 build). Disabling merging trades that one fragile,
+        # unresumable mega-request for many small ones spread over
+        # --download-threads (default 4) — slower on a clean run, but no
+        # longer one dropped stream away from losing 20 minutes of transfer.
+        run_pmtiles!("extract", item["url"], extract_path, "--region=#{region_path}", "--maxzoom=#{TERRAIN_MAX_ZOOM}", "--overfetch=0")
         extract_path
       end
 
@@ -621,11 +631,23 @@ module Maptool
       ENV.fetch("PLANETILER_JAR", "/usr/local/lib/planetiler.jar")
     end
 
+    # go-pmtiles' extract logs one progress line per chunk fetched
+    # ("fetching N dirs, M chunks, K requests") — fine on a terminal where
+    # it overwrites itself, but captured non-interactively that's one line
+    # per chunk, easily hundreds of thousands of characters for a big
+    # region. The actual failure is always at the end, so keep only the
+    # tail — same story for osmium/planetiler's own chatty stdout.
+    OUTPUT_TAIL_LINES = 40
+
     # Some CLI tools (go-pmtiles included) write their actual error to
     # stdout rather than stderr — combine both rather than risk raising a
     # BuildError with nothing useful in it.
     def self.command_output(stdout, stderr)
-      [stdout, stderr].map(&:strip).reject(&:empty?).join("\n")
+      lines = [stdout, stderr].map(&:strip).reject(&:empty?).join("\n").lines
+      return lines.join if lines.size <= OUTPUT_TAIL_LINES
+
+      omitted = lines.size - OUTPUT_TAIL_LINES
+      "(#{omitted} earlier lines omitted)\n#{lines.last(OUTPUT_TAIL_LINES).join}"
     end
     private_class_method :command_output
 

public/css/style.css

@@ -542,6 +542,14 @@ html, body {
   font-size: 12px;
 }
 
+.offline-data-panel__status--error {
+  white-space: pre-wrap;
+  word-break: break-word;
+  max-height: 200px;
+  overflow-y: auto;
+  color: #b3261e;
+}
+
 .offline-data-panel__update-badge {
   color: #b8860b;
 }

public/js/offline_data.js

@@ -206,23 +206,31 @@ class DataFileRow {
     this._refreshStatus();
   }
 
+  // Single place that writes to _statusEl, so the error styling (pre-wrap,
+  // scrollable, red — see style.css) never stays stuck on once a build's
+  // failure message has scrolled past into a normal status line.
+  _setStatus(text, isError = false) {
+    this._statusEl.textContent = text;
+    this._statusEl.classList.toggle("offline-data-panel__status--error", isError);
+  }
+
   async _refreshStatus() {
     let data;
     try {
       data = await (await fetch(this._statusUrl)).json();
     } catch {
-      this._statusEl.textContent = "Could not check status.";
+      this._setStatus("Could not check status.");
       return;
     }
 
     if (!data.available) {
-      this._statusEl.textContent = "Not generated yet.";
+      this._setStatus("Not generated yet.");
       this._deleteButton.hidden = true;
       return;
     }
 
     const updated = new Date(data.mtime).toLocaleString([], { dateStyle: "short", timeStyle: "short" });
-    this._statusEl.textContent = `${formatBytes(data.size)} · updated ${updated}${this._formatMeta(data.meta)}`;
+    this._setStatus(`${formatBytes(data.size)} · updated ${updated}${this._formatMeta(data.meta)}`);
     this._deleteButton.hidden = !isLoggedIn();
     this._checkUpdate();
   }
@@ -274,7 +282,7 @@ class DataFileRow {
   // Called by OfflineDataControl once RegionPicker's Apply has saved a
   // new selection — not tied to a button of this row's own.
   async triggerBuild() {
-    this._statusEl.textContent = "Checking…";
+    this._setStatus("Checking…");
     try {
       // false (not null/true — see offline_update_available? in
       // pmtiles_builder.rb) means the just-saved selection matches what's
@@ -291,17 +299,17 @@ class DataFileRow {
       // couldn't tell — fall through and build rather than silently skip
     }
 
-    this._statusEl.textContent = "Starting…";
+    this._setStatus("Starting…");
     try {
       const response = await fetch(this._buildUrl, { method: "POST" });
       const data = await response.json().catch(() => ({}));
       if (!response.ok) throw new Error(data.error || `build failed: ${response.status}`);
     } catch (err) {
-      this._statusEl.textContent = err.message;
+      this._setStatus(err.message, true);
       return;
     }
 
-    this._statusEl.textContent = "Building…";
+    this._setStatus("Building…");
     this._poll();
   }
 
@@ -318,13 +326,13 @@ class DataFileRow {
       }
 
       if (status.state === "running") {
-        this._statusEl.textContent = formatRunning(status);
+        this._setStatus(formatRunning(status));
         this._poll();
         return;
       }
 
       if (status.state === "error") {
-        this._statusEl.textContent = `Build failed: ${status.message}`;
+        this._setStatus(`Build failed: ${status.message}`, true);
       } else {
         this._refreshStatus();
       }

test/test_pmtiles_builder.rb

@@ -47,6 +47,21 @@ class PmtilesBuilderTest < Minitest::Test
     File.write(@pmtiles_fail, "#!/bin/sh\necho 'boom' >&2\nexit 1\n")
     File.chmod(0o755, @pmtiles_fail)
 
+    # Mimics go-pmtiles' real extract failure mode: hundreds of progress
+    # lines before the actual error — see command_output's tail-truncation.
+    @pmtiles_fail_noisy = File.join(@bin_dir, "pmtiles-fail-noisy")
+    File.write(@pmtiles_fail_noisy, <<~SH)
+      #!/bin/sh
+      i=0
+      while [ "$i" -lt 200 ]; do
+        echo "fetching $i dirs, $i chunks, 1 requests" >&2
+        i=$((i + 1))
+      done
+      echo "stream error: stream ID 9; INTERNAL_ERROR" >&2
+      exit 1
+    SH
+    File.chmod(0o755, @pmtiles_fail_noisy)
+
     # osmium merge PBF1 PBF2 ... -o OUTPUT — output is always "-o"'s next
     # arg, unlike pmtiles' varying-position output above.
     @osmium_ok = File.join(@bin_dir, "osmium-ok")
@@ -545,6 +560,22 @@ class PmtilesBuilderTest < Minitest::Test
     assert_match(/exited with status/, status[:message])
   end
 
+  # The real trigger for this: a Mapterhorn extract that failed after ~7000
+  # progress lines, producing a 500k+ character status[:message] that was
+  # unreadable in the web UI.
+  def test_build_terrain_truncates_a_noisy_command_failure
+    select_regions("ES")
+    ENV["PMTILES_BIN"] = @pmtiles_fail_noisy
+    thread = Maptool::PmtilesBuilder.build_terrain(data_dir: @data_dir, fetch_manifest: -> { fake_manifest })
+    thread.join
+
+    status = Maptool::PmtilesBuilder.terrain_status
+    assert_equal "error", status[:state]
+    assert_match(/\(\d+ earlier lines omitted\)/, status[:message])
+    assert_match(/stream error: stream ID 9/, status[:message])
+    assert_operator status[:message].length, :<, 5000
+  end
+
   def test_build_terrain_raises_when_one_is_already_running
     select_regions("ES")
     ENV["PMTILES_BIN"] = @pmtiles_ok