fix(worker): Add tmp shard cleanup on indexing failure (#805)

* fix: cleanup temporary shard files on failed indexing

When zoekt-git-index fails during repository indexing, it can leave behind
.tmp shard files that accumulate over time and fill up disk space. This is
especially problematic for large repos that repeatedly fail to index.

Changes:
- Add cleanupTempShards() function to zoekt.ts that removes temporary shard
  files (files with .tmp in their name) for a specific repository
- Call cleanupTempShards() in repoIndexManager.ts when indexGitRepository
  fails, before re-throwing the error

This ensures that even if a repository consistently fails to index, the
temporary files created during each attempt are cleaned up.

Co-authored-by: michael <michael@sourcebot.dev>

* fix merge issue

* changelog

* changelog

---------

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Brendan Kellam <bshizzle1234@gmail.com>

Author: 3 people

CHANGELOG.md

@@ -16,6 +16,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Fixed
 - Fixed issue where certain file and folder names would cause type errors. [#862](https://github.com/sourcebot-dev/sourcebot/pull/862)
 - Fixed token refresh error "Provider config not found or invalid for: x" when a sso is configured using deprecated env vars. [#841](https://github.com/sourcebot-dev/sourcebot/pull/841)
+- Fixed issue where temporary shard files created on index failure were not being cleaned up. [#805](https://github.com/sourcebot-dev/sourcebot/pull/805)
 
 ## [4.10.27] - 2026-02-05
 

packages/backend/src/repoIndexManager.ts

@@ -1,20 +1,19 @@
 import * as Sentry from '@sentry/node';
 import { PrismaClient, Repo, RepoIndexingJobStatus, RepoIndexingJobType } from "@sourcebot/db";
-import { createLogger, Logger } from "@sourcebot/shared";
-import { env, RepoIndexingJobMetadata, repoIndexingJobMetadataSchema, RepoMetadata, repoMetadataSchema, getRepoPath } from '@sourcebot/shared';
+import { createLogger, env, getRepoPath, Logger, RepoIndexingJobMetadata, repoIndexingJobMetadataSchema, RepoMetadata, repoMetadataSchema } from "@sourcebot/shared";
+import { DelayedError, Job, Queue, Worker } from "bullmq";
 import { existsSync } from 'fs';
 import { readdir, rm } from 'fs/promises';
-import { DelayedError, Job, Queue, Worker } from "bullmq";
 import { Redis } from 'ioredis';
-import Redlock, { ExecutionError } from 'redlock';
 import micromatch from 'micromatch';
-import { WORKER_STOP_GRACEFUL_TIMEOUT_MS, INDEX_CACHE_DIR } from './constants.js';
+import Redlock, { ExecutionError } from 'redlock';
+import { INDEX_CACHE_DIR, WORKER_STOP_GRACEFUL_TIMEOUT_MS } from './constants.js';
 import { cloneRepository, fetchRepository, getBranches, getCommitHashForRefName, getLatestCommitTimestamp, getLocalDefaultBranch, getTags, isPathAValidGitRepoRoot, unsetGitConfig, upsertGitConfig } from './git.js';
 import { captureEvent } from './posthog.js';
 import { PromClient } from './promClient.js';
 import { RepoWithConnections, Settings } from "./types.js";
 import { getAuthCredentialsForRepo, getShardPrefix, measure, setIntervalAsync } from './utils.js';
-import { indexGitRepository } from './zoekt.js';
+import { cleanupTempShards, indexGitRepository } from './zoekt.js';
 
 const LOG_TAG = 'repo-index-manager';
 const logger = createLogger(LOG_TAG);
@@ -478,9 +477,17 @@ export class RepoIndexManager {
         }
 
         logger.info(`Indexing ${repo.name} (id: ${repo.id})...`);
-        const { durationMs } = await measure(() => indexGitRepository(repo, this.settings, revisions, signal));
-        const indexDuration_s = durationMs / 1000;
-        logger.info(`Indexed ${repo.name} (id: ${repo.id}) in ${indexDuration_s}s`);
+        try {
+            const { durationMs } = await measure(() => indexGitRepository(repo, this.settings, revisions, signal));
+            const indexDuration_s = durationMs / 1000;
+            logger.info(`Indexed ${repo.name} (id: ${repo.id}) in ${indexDuration_s}s`);
+        } catch (error) {
+            // Clean up any temporary shard files left behind by the failed indexing operation.
+            // Zoekt creates .tmp files during indexing which can accumulate if indexing fails repeatedly.
+            logger.warn(`Indexing failed for ${repo.name} (id: ${repo.id}), cleaning up temp shard files...`);
+            await cleanupTempShards(repo);
+            throw error;
+        }
 
         return revisions;
     }

packages/backend/src/zoekt.ts

@@ -1,6 +1,7 @@
 import { Repo } from "@sourcebot/db";
 import { createLogger, env, getRepoPath } from "@sourcebot/shared";
 import { exec } from "child_process";
+import { readdir, rm } from "fs/promises";
 import { INDEX_CACHE_DIR } from "./constants.js";
 import { Settings } from "./types.js";
 import { getShardPrefix } from "./utils.js";
@@ -54,3 +55,34 @@ export const indexGitRepository = async (repo: Repo, settings: Settings, revisio
         })
     });
 }
+
+/**
+ * Cleans up temporary shard files left behind by a failed indexing operation.
+ * Zoekt creates temporary files (with `.tmp` suffix) during indexing, which
+ * can be left behind if the indexing process fails or is interrupted.
+ * 
+ * @param repo - The repository whose temp shards should be cleaned up
+ */
+export const cleanupTempShards = async (repo: Repo) => {
+    const shardPrefix = getShardPrefix(repo.orgId, repo.id);
+    
+    try {
+        const files = await readdir(INDEX_CACHE_DIR);
+        const tempFiles = files.filter(file => 
+            file.startsWith(shardPrefix) && file.includes('.tmp')
+        );
+        
+        for (const file of tempFiles) {
+            const filePath = `${INDEX_CACHE_DIR}/${file}`;
+            logger.info(`Cleaning up temp shard file: ${filePath}`);
+            await rm(filePath, { force: true });
+        }
+        
+        if (tempFiles.length > 0) {
+            logger.info(`Cleaned up ${tempFiles.length} temp shard file(s) for repo ${repo.id}`);
+        }
+    } catch (error) {
+        // Log but don't throw - cleanup is best effort
+        logger.warn(`Failed to cleanup temp shards for repo ${repo.id}:`, error);
+    }
+}