fix(bench): resolve query benchmark CI failure and increase embedding timeout (#749)

* fix(bench): resolve doubled file paths in query benchmark on CI

The query benchmark's benchDiffImpact reads row.file from the DB and
joins it with the repo root, but on CI the npm-installed buildGraph
stores paths that include the root directory prefix without a leading
slash.  This produces doubled paths like /root/root/src/... causing
ENOENT.

Add resolveDbFile() helper that detects absolute-like paths and falls
back to prepending '/' when the joined path doesn't exist.

Also increase embedding benchmark per-model timeout from 10→20 minutes
and add explicit workflow timeout (180 min) so model downloads on cold
CI caches don't cause silent failures.

* fix(bench): return null from resolveDbFile and increase job timeout (#749)

- resolveDbFile now returns null when no candidate exists, eliminating
  redundant existsSync calls in the caller
- Bump embedding-benchmark job timeout from 180 to 195 min to provide
  adequate headroom for 7 cold-cache model downloads at 20 min each

Author: carlos-alm

.github/workflows/benchmark.yml

@@ -156,6 +156,8 @@ jobs:
 
   embedding-benchmark:
     runs-on: ubuntu-latest
+    # 7 models x 20 min each = 140 min worst-case + ~30 min setup/npm-wait headroom
+    timeout-minutes: 195
     if: >-
       github.event_name == 'workflow_dispatch' ||
       (github.event.workflow_run.conclusion == 'success' &&
@@ -240,6 +242,7 @@ jobs:
 
       - name: Run embedding benchmark
         if: steps.existing.outputs.skip != 'true'
+        timeout-minutes: 160
         env:
           HF_TOKEN: ${{ secrets.HF_TOKEN }}
         run: |

scripts/embedding-benchmark.ts

@@ -125,7 +125,7 @@ const dbPath = path.join(root, '.codegraph', 'graph.db');
 
 const { MODELS } = await import(srcImport(srcDir, 'domain/search/index.js'));
 
-const TIMEOUT_MS = 600_000;
+const TIMEOUT_MS = 1_200_000; // 20 min — model download + embedding can be slow on CI
 const hasHfToken = !!process.env.HF_TOKEN;
 const modelKeys = Object.keys(MODELS);
 const results = {};

scripts/query-benchmark.ts

@@ -176,6 +176,21 @@ function benchDepths(fn, name, depths) {
 	return result;
 }
 
+/**
+ * Resolve a file path from the DB to an absolute path.
+ * Handles relative paths (normal) and absolute-like paths without leading '/'
+ * (observed on CI when the npm-installed buildGraph stores full paths).
+ */
+function resolveDbFile(rootDir: string, dbFile: string): string | null {
+	if (path.isAbsolute(dbFile)) return fs.existsSync(dbFile) ? dbFile : null;
+	const joined = path.join(rootDir, dbFile);
+	if (fs.existsSync(joined)) return joined;
+	// DB may store an absolute path without the leading '/'
+	const withSlash = '/' + dbFile;
+	if (fs.existsSync(withSlash)) return withSlash;
+	return null;
+}
+
 function benchDiffImpact(hubName) {
 	const db = new Database(dbPath, { readonly: true });
 	const row = db
@@ -185,7 +200,14 @@ function benchDiffImpact(hubName) {
 
 	if (!row) return { latencyMs: 0, affectedFunctions: 0, affectedFiles: 0 };
 
-	const hubFile = path.join(root, row.file);
+	// row.file is normally relative (e.g. 'src/domain/builder.ts'), but some
+	// environments store absolute-like paths without the leading '/'.  Handle
+	// both cases so the benchmark works regardless of DB path format.
+	const hubFile = resolveDbFile(root, row.file);
+	if (!hubFile) {
+		console.error(`[benchDiffImpact] Cannot find hub file for row.file=${row.file}`);
+		return { latencyMs: 0, affectedFunctions: 0, affectedFiles: 0 };
+	}
 	const original = fs.readFileSync(hubFile, 'utf8');
 
 	try {