⚡ Bolt: Optimize ArtifactStore GC performance with batched I/O operations

.jules/bolt.md

@@ -0,0 +1,3 @@
+## 2024-06-11 - Optimize ArtifactStore garbage collection performance
+**Learning:** Sequential fs.stat and fs.rm operations in large directories (like the artifact store after extended sessions) caused significant Event Loop blocking and O(N) execution time, dramatically slowing down `maybeGc()`.
+**Action:** Replaced sequential awaits in `ArtifactStore.gc` loops with concurrent batching (Promise.all) in chunks of 10. Chunked iteration is essential because it avoids opening too many file descriptors (EMFILE) while vastly improving throughput over sequential await.

src/core/sub-agent/artifacts/store.ts

@@ -93,13 +93,26 @@ export class ArtifactStore {
     const entries = await fs.readdir(root, { withFileTypes: true }).catch(() => []);
 
     const files: Array<{ name: string; path: string; mtimeMs: number; size: number }> = [];
-    for (const entry of entries) {
-      if (!entry.isFile()) continue;
-      const filePath = path.join(root, entry.name);
-      if (!isWithinDir(root, filePath)) continue;
-      const stat = await fs.stat(filePath).catch(() => null);
-      if (!stat) continue;
-      files.push({ name: entry.name, path: filePath, mtimeMs: stat.mtimeMs, size: stat.size });
+    const validEntries = entries.filter(
+      (e) => e.isFile() && isWithinDir(root, path.join(root, e.name)),
+    );
+
+    for (let i = 0; i < validEntries.length; i += 10) {
+      const chunk = validEntries.slice(i, i + 10);
+      await Promise.all(
+        chunk.map(async (entry) => {
+          const filePath = path.join(root, entry.name);
+          const stat = await fs.stat(filePath).catch(() => null);
+          if (stat) {
+            files.push({
+              name: entry.name,
+              path: filePath,
+              mtimeMs: stat.mtimeMs,
+              size: stat.size,
+            });
+          }
+        }),
+      );
     }
 
     const maxAgeMs = options?.maxAgeMs ?? LIMITS.artifactTtlMs;
@@ -118,8 +131,9 @@ export class ArtifactStore {
       removedBytes += file.size;
     };
 
-    for (const file of expired) {
-      await removeFile(file);
+    for (let i = 0; i < expired.length; i += 10) {
+      const chunk = expired.slice(i, i + 10);
+      await Promise.all(chunk.map((file) => removeFile(file)));
     }
 
     // Recompute remaining after TTL removal (newest first).
@@ -130,17 +144,23 @@ export class ArtifactStore {
     let currentFiles = remaining.length;
     let currentBytes = remaining.reduce((acc, f) => acc + f.size, 0);
 
+    const toRemove: Array<{ path: string; size: number }> = [];
     for (let i = remaining.length - 1; i >= 0; i--) {
       const tooManyFiles = currentFiles > maxFiles;
       const tooManyBytes = currentBytes > maxTotalBytes;
       if (!tooManyFiles && !tooManyBytes) break;
 
       const oldest = remaining[i];
-      await removeFile(oldest);
+      toRemove.push(oldest);
       currentFiles -= 1;
       currentBytes -= oldest.size;
     }
 
+    for (let i = 0; i < toRemove.length; i += 10) {
+      const chunk = toRemove.slice(i, i + 10);
+      await Promise.all(chunk.map((file) => removeFile(file)));
+    }
+
     return { removedFiles, removedBytes };
   }
 

tests/integration/prompt_templates.test.ts

@@ -36,7 +36,9 @@ describe('Prompt templates', () => {
 
     expect(planSystem).toContain('You are SalmonLoop.');
     expect(patchSystem).toContain('You are PATCH, a phase-native diff compiler.');
-    expect(autopilotSystem).toContain('You are a senior software engineer running in "autopilot" mode.');
+    expect(autopilotSystem).toContain(
+      'You are a senior software engineer running in "autopilot" mode.',
+    );
     expect(answerSystem).toContain('You are a coding assistant in "answer" mode.');
     expect(researchSystem).toContain('You are a research assistant.');
   });