code-graph-context 2.10.0 → 2.10.2

package/dist/core/embeddings/embedding-sidecar.js

@@ -174,6 +174,13 @@ export class EmbeddingSidecar {
             });
             if (!res.ok) {
                 const detail = await res.text();
+                const isOOM = detail.toLowerCase().includes('out of memory');
+                if (res.status === 500 && isOOM) {
+                    // OOM leaves GPU memory in a corrupted state — kill the sidecar
+                    // so the next request spawns a fresh process with clean memory
+                    console.error('[embedding-sidecar] OOM detected, restarting sidecar to reclaim GPU memory');
+                    await this.stop();
+                }
                 throw new Error(`Sidecar embed failed (${res.status}): ${detail}`);
             }
             const data = (await res.json());
@@ -183,6 +190,9 @@ export class EmbeddingSidecar {
         }
         catch (err) {
             if (err instanceof Error && err.name === 'AbortError') {
+                // Timeout likely means the sidecar is stuck — kill it
+                console.error('[embedding-sidecar] Request timed out, restarting sidecar');
+                await this.stop();
                 throw new Error(`Embedding request timed out after ${this.config.requestTimeoutMs}ms`);
             }
             throw err;

package/dist/core/embeddings/local-embeddings.service.js

@@ -6,7 +6,7 @@
 import { debugLog } from '../../mcp/utils.js';
 import { getEmbeddingSidecar } from './embedding-sidecar.js';
 const BATCH_CONFIG = {
-    maxBatchSize: 16, // Small batches — 1.5B model on MPS has limited VRAM
+    maxBatchSize: 8, // Small batches — 1.5B model on MPS OOMs at higher values on 16GB machines
 };
 export class LocalEmbeddingsService {
     async embedText(text) {
@@ -20,12 +20,14 @@ export class LocalEmbeddingsService {
         return sidecar.embed(texts);
     }
     async embedTextsInBatches(texts, batchSize = BATCH_CONFIG.maxBatchSize) {
+        // Cap batch size — callers (e.g. graph-generator) may pass 100 which OOMs the local model
+        const safeBatchSize = Math.min(batchSize, BATCH_CONFIG.maxBatchSize);
         await debugLog('Batch embedding started', { provider: 'local', textCount: texts.length });
         const sidecar = getEmbeddingSidecar();
         const results = [];
-        const totalBatches = Math.ceil(texts.length / batchSize);
-        for (let i = 0; i < texts.length; i += batchSize) {
-            const batch = texts.slice(i, i + batchSize);
+        const totalBatches = Math.ceil(texts.length / safeBatchSize);
+        for (let i = 0; i < texts.length; i += safeBatchSize) {
+            const batch = texts.slice(i, i + safeBatchSize);
             const batchIndex = Math.floor(i / batchSize) + 1;
             await debugLog('Embedding batch progress', {
                 provider: 'local',

package/dist/mcp/mcp.server.js

@@ -124,11 +124,13 @@ process.on('uncaughtException', async (error) => {
     console.error(JSON.stringify({ level: 'error', message: 'Uncaught exception', error: String(error), stack: error.stack }));
     await debugLog('Uncaught exception', { error: String(error), stack: error.stack });
     await logServerStats('uncaught-exception');
+    await stopEmbeddingSidecar();
 });
 process.on('unhandledRejection', async (reason) => {
     console.error(JSON.stringify({ level: 'error', message: 'Unhandled rejection', reason: String(reason) }));
     await debugLog('Unhandled rejection', { reason: String(reason) });
     await logServerStats('unhandled-rejection');
+    await stopEmbeddingSidecar();
 });
 // Log other process events that might indicate issues
 process.on('warning', async (warning) => {
@@ -137,6 +139,7 @@ process.on('warning', async (warning) => {
 process.on('beforeExit', async (code) => {
     await debugLog('Process beforeExit', { code });
     await logServerStats('before-exit');
+    await stopEmbeddingSidecar();
 });
 process.on('exit', (code) => {
     // Note: Can't use async here, exit is synchronous

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "code-graph-context",
-  "version": "2.10.0",
+  "version": "2.10.2",
   "description": "MCP server that builds code graphs to provide rich context to LLMs",
   "type": "module",
   "homepage": "https://github.com/drewdrewH/code-graph-context#readme",