Merge pull request 'Bound how long the blog cache serves stale entries' (#7) from fix/blog-cache-staleness into master

Reviewed-on: #7

.gitea/workflows/deploy.yml

@@ -1,4 +1,4 @@
-# .gitea/workflows/deploy.yml
+# .gitea/workflows/deploy.yml
 name: Build and Deploy
 
 on:
@@ -12,13 +12,17 @@ concurrency:
 
 jobs:
   build-deploy:
+    # Gitea's runner images are lean, so the toolchain is installed below rather
+    # than assumed. The glibc that libchess_engine.so links against comes from
+    # this image, not from the runner host — keep it matched to the prod server.
     runs-on: ubuntu-latest
 
     steps:
-      - name: Install deploy tools
+      - name: Install toolchain
         run: |
           apt-get update
-          apt-get install -y --no-install-recommends rsync openssh-client
+          apt-get install -y --no-install-recommends \
+            cmake build-essential rsync openssh-client
 
       - name: Checkout
         uses: actions/checkout@v4
@@ -29,10 +33,26 @@ jobs:
           dotnet-version: '8.0.x'
 
       - name: Restore
-        run: dotnet restore
+        run: dotnet restore JoshHeaps.Net/JoshHeaps.Net.csproj
 
+      - name: Build native chess engine (libchess_engine.so)
+        run: |
+          cmake -S native/chess_engine -B native/chess_engine/build -DCMAKE_BUILD_TYPE=Release
+          cmake --build native/chess_engine/build
+          cp native/chess_engine/build/libchess_engine.so JoshHeaps.Net/Resources/
+
+      # Publishing the project rather than the solution keeps chess_engine.vcxproj
+      # (Windows-only MSBuild C++ targets) and the UiTests assemblies out of it.
       - name: Publish
-        run: dotnet publish -c Release -o ./publish
+        run: dotnet publish JoshHeaps.Net/JoshHeaps.Net.csproj -c Release -o ./publish
+
+      - name: Verify payload
+        run: |
+          set -euo pipefail
+          test -f publish/Resources/libchess_engine.so \
+            || { echo "libchess_engine.so missing from publish output"; exit 1; }
+          test -x publish/Resources/stockfish-ubuntu-x86-64-sse41-popcnt \
+            || { echo "stockfish is not executable"; exit 1; }
 
       - name: Prepare SSH
         env:
@@ -47,6 +67,8 @@ jobs:
           chmod 600 ~/.ssh/deploy_key
           ssh-keyscan -p "$PORT" "$SSH_HOST" >> ~/.ssh/known_hosts 2>/dev/null
 
+      # --exclude chess-data: never let --delete remove the learned-engine training
+      # data, which lives in the deploy dir unless ChessEngine__WeightsPath is set.
       - name: Rsync to server
         env:
           SSH_HOST:   ${{ secrets.SSH_HOST }}
@@ -56,11 +78,11 @@ jobs:
         run: |
           set -euo pipefail
           PORT="${SSH_PORT:-22}"
-          rsync -az --delete \
+          rsync -az --delete --exclude 'chess-data' \
             -e "ssh -p $PORT -i ~/.ssh/deploy_key -o StrictHostKeyChecking=yes" \
             publish/ "$SSH_USER@$SSH_HOST:$TARGET_DIR/"
 
-      - name: Restart service
+      - name: Reload and restart service
         env:
           SSH_HOST:     ${{ secrets.SSH_HOST }}
           SSH_PORT:     ${{ secrets.SSH_PORT }}

JoshHeaps.Net/Services/Implementations/BlogService.cs

@@ -1,4 +1,4 @@
-using System.Collections.Concurrent;
+using System.Collections.Concurrent;
 using System.Text.Json;
 using JoshHeaps.Net.Models;
 using JoshHeaps.Net.Services.Interfaces;
@@ -10,6 +10,7 @@ public class BlogService : IBlogService
     private readonly HttpClient _httpClient;
     private readonly ILogger<BlogService> _logger;
     private readonly TimeSpan _cacheTtl = TimeSpan.FromMinutes(5);
+    private readonly TimeSpan _staleGrace = TimeSpan.FromHours(1);
 
     private readonly ConcurrentDictionary<string, CacheEntry> _cache = new();
 
@@ -65,7 +66,9 @@ public class BlogService : IBlogService
 
     private async Task<T?> GetCachedAsync<T>(string key, Func<Task<T?>> factory) where T : class
     {
-        if (_cache.TryGetValue(key, out var entry) && entry.ExpiresAt > DateTime.UtcNow)
+        _cache.TryGetValue(key, out var entry);
+
+        if (entry is not null && entry.ExpiresAt > DateTime.UtcNow)
             return (T?)entry.Value;
 
         var result = await factory();
@@ -73,15 +76,26 @@ public class BlogService : IBlogService
         if (result is not null)
         {
             _cache[key] = new CacheEntry(result, DateTime.UtcNow.Add(_cacheTtl));
+            return result;
         }
-        else if (entry is not null)
+
+        return entry is null ? null : ServeStale<T>(key, entry);
+    }
+
+    // Stale entries are dropped once the grace window closes so that a dead API fails the same
+    // way for every key. Serving them indefinitely let a cached post list render next to 404s on
+    // the posts themselves, which reads as a site bug rather than an outage.
+    private T? ServeStale<T>(string key, CacheEntry entry) where T : class
+    {
+        if (entry.ExpiresAt.Add(_staleGrace) > DateTime.UtcNow)
         {
-            // API unreachable — serve stale cache
-            _logger.LogWarning("Serving stale cache for key {Key}", key);
+            _logger.LogWarning("Blog API unreachable, serving stale cache for key {Key}", key);
             return (T?)entry.Value;
         }
 
-        return result;
+        _cache.TryRemove(key, out _);
+        _logger.LogError("Blog API unreachable for over {StaleGrace}, dropping stale cache for key {Key}", _staleGrace, key);
+        return null;
     }
 
     public void ClearCache()