fix: S3→Telegram upload pipeline — OOM, queue limits, shutdown drain, timeouts
Deploy FileDrop / deploy (push) Successful in 46s

CRITICAL:
- Content-MD5 no longer loads entire file via arrayBuffer() — MD5 computed
  incrementally in streamBodyToTemp alongside SHA-256 (fixes OOM for GB files)

HIGH:
- Add 120s timeout to Telegraf API calls via Promise.race in executeWithBotRetry
  (prevents queue slot exhaustion from hung Telegram connections)
- Add queue size limit (1000 pending max) — reject new tasks when full
- Add graceful shutdown drain — waitForQueue with 30s timeout before exit
- Fix temp file leak when findFileByBucketAndKey throws (wrap in try-catch)
- Fix createReadStream fd leak — destroy stream on forwardToStorage error
- writer.end() wrapped in silent try-catch to prevent error swallowing
- writer.end() result ignored, writerFailed flag prevents double-end

MEDIUM:
- Remove 'retry after' from isTransientError patterns to stop double-retry
  layering (was causing up to 96 bot attempts per chunk)

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Claude
2026-07-29 08:40:50 +07:00
parent de7d276245
commit d8da2044b2
4 changed files with 110 additions and 42 deletions
+15 -1
View File
@@ -1,6 +1,7 @@
import { serve } from 'bun';
import { config } from './config/index';
import { fileInfoCache } from './infrastructure/cache/index';
import { clearQueue, getQueueStats, waitForQueue } from './infrastructure/telegram/upload-queue';
import { startBot } from './interfaces/bot/handler';
import { handleS3Request } from './interfaces/http/controllers/s3-controller';
import { cleanupRateLimitCache } from './interfaces/http/middleware/rate-limit';
@@ -63,9 +64,22 @@ logger.info('Server started', { port: config.port, url: config.baseUrl });
const gracefulShutdown = async (signal: string): Promise<void> => {
logger.info('Graceful shutdown signal received', { signal });
logger.info('Closing HTTP server');
logger.info('Closing HTTP server — no new requests accepted');
server.stop();
// Drain pending upload queue with a timeout
const { pending, size } = getQueueStats();
if (pending > 0 || size > 0) {
logger.info('Draining upload queue', { pending, size });
const drainTimeout = setTimeout(() => {
logger.warn('Upload queue drain timeout — clearing remaining tasks');
clearQueue();
}, 30_000);
await waitForQueue();
clearTimeout(drainTimeout);
logger.info('Upload queue drained');
}
logger.info('Stopping Telegram bot');
bot.stop(signal);
+20 -2
View File
@@ -42,7 +42,9 @@ const sleep = (ms: number): Promise<void> => {
const isTransientError = (error: unknown): boolean => {
const str = error instanceof Error ? error.message : String(error);
const transientPatterns = [
'retry after',
// 'retry after' is deliberately omitted — 429 is handled by
// executeWithBotRetry at a deeper layer. Including it here would
// cause double-retry (up to 96 attempts per chunk).
'timeout',
'Timed out',
'etimedout',
@@ -76,6 +78,12 @@ const isTransientError = (error: unknown): boolean => {
*/
const MAX_TRANSIENT_RETRIES = 3;
/**
* Timeout in milliseconds for individual Telegram API calls.
* 120 seconds to accommodate large document uploads.
*/
const TELEGRAM_API_TIMEOUT_MS = 120_000;
/**
* Manages a pool of Telegram bots with automatic rotation and rate-limit handling.
*
@@ -129,7 +137,17 @@ export class BotPool implements ITelegramService {
const currentBot = this.bots[botIndex];
const currentToken = this.botTokens[botIndex];
try {
return await action(currentBot, currentToken);
// Add timeout to prevent hung API calls from occupying queue slots
const result = await Promise.race([
action(currentBot, currentToken),
new Promise<never>((_, reject) =>
setTimeout(
() => reject(new Error(`Telegram API timeout after ${TELEGRAM_API_TIMEOUT_MS}ms`)),
TELEGRAM_API_TIMEOUT_MS,
),
),
]);
return result;
} catch (error: unknown) {
const errorStr = error instanceof Error ? error.message : String(error);
const match = errorStr.match(/retry after (\d+)/i);
+20 -1
View File
@@ -2,6 +2,13 @@ import PQueue from 'p-queue';
import { config } from '../../env';
import logger from '../../shared/logger/index';
/**
* Maximum number of pending (queued + in-flight) upload tasks before
* new submissions are rejected. Prevents unbounded memory growth when
* Telegram is slow or unavailable.
*/
const MAX_QUEUE_PENDING = 1000;
/**
* P-queue instance for serialising Telegram upload tasks.
*
@@ -16,7 +23,11 @@ const uploadQueue = new PQueue({
uploadQueue.on('add', () => {
const stats = getQueueStats();
if (stats.size > 5) {
logger.warn('Upload queue building up', { pending: stats.pending, size: stats.size });
logger.warn('Upload queue building up', {
pending: stats.pending,
size: stats.size,
max: MAX_QUEUE_PENDING,
});
}
});
@@ -34,6 +45,14 @@ uploadQueue.on('next', () => {
* @returns A promise that resolves with the task's result.
*/
export const enqueueUpload = <T>(task: () => Promise<T>): Promise<T> => {
const stats = getQueueStats();
if (stats.pending + stats.size > MAX_QUEUE_PENDING) {
return Promise.reject(
new Error(
`Upload queue full (${stats.pending + stats.size} pending, max ${MAX_QUEUE_PENDING})`,
),
);
}
return uploadQueue.add(task);
};
@@ -778,25 +778,34 @@ const handleHeadObject = async (
};
/**
* Streams the request body to a temporary file while computing its SHA-256 hash.
* Streams the request body to a temporary file while computing its SHA-256
* and MD5 hashes.
*
* Unlike `req.arrayBuffer()`, this approach uses O(1) memory regardless of
* file size, making it safe for multi-GB Docker registry layer blobs.
*
* MD5 is computed alongside SHA-256 so that Content-MD5 verification (when
* the header is present) does not need to re-read the entire file.
*
* @param body - The ReadableStream from the HTTP request body.
* @returns The temp file path, SHA-256 hash, total size, and signature bytes.
* @returns The temp file path, SHA-256 hash, MD5 hash (base64), total size, and signature bytes.
*/
const streamBodyToTemp = async (
body: ReadableStream<Uint8Array> | null,
): Promise<{
tempPath: string;
fileHash: string;
md5Hash: string;
sizeBytes: number;
signatureBuffer: Buffer;
}> => {
const tempPath = `/tmp/filedrop-s3-${nanoid()}`;
const writer = Bun.file(tempPath).writer();
const hasher = new Bun.CryptoHasher('sha256');
const sha256 = new Bun.CryptoHasher('sha256');
const md5 = new Bun.CryptoHasher('md5');
let writerFailed = false;
// Handle body being null (GET/HEAD/DELETE or empty PUT)
const reader = (
body ??
new ReadableStream({
@@ -816,7 +825,8 @@ const streamBodyToTemp = async (
if (done) break;
const chunk = Buffer.from(value);
sizeBytes += chunk.byteLength;
hasher.update(chunk);
sha256.update(chunk);
md5.update(chunk);
writer.write(chunk);
if (signatureBytes < SIGNATURE_BYTES) {
@@ -827,16 +837,27 @@ const streamBodyToTemp = async (
}
}
writer.end();
try {
writer.end();
} catch {
writerFailed = true;
}
return {
tempPath,
fileHash: hasher.digest('hex'),
fileHash: sha256.digest('hex'),
md5Hash: md5.digest('base64'),
sizeBytes,
signatureBuffer: Buffer.concat(signatureChunks, signatureBytes),
};
} catch (error) {
writer.end();
if (!writerFailed) {
try {
writer.end();
} catch {
/* writer may already be errored */
}
}
await cleanupTempFile(tempPath);
throw error;
} finally {
@@ -926,25 +947,17 @@ const handlePutObject = async (
}
}
// Content-MD5 validation: verify MD5 when Content-MD5 header is present
// Content-MD5 validation: use pre-computed MD5 from streaming (no OOM re-read)
const contentMd5 = headers['content-md5'];
if (contentMd5) {
const computedMd5 = Buffer.from(
await crypto.subtle.digest(
'MD5',
new Uint8Array(await Bun.file(streamed.tempPath).arrayBuffer()),
),
).toString('base64');
if (contentMd5 !== computedMd5) {
await cleanupTempFile(streamed.tempPath);
return s3ErrorResponse(
'BadDigest',
'The Content-MD5 you specified did not match what we received.',
`/${bucket}/${key}`,
400,
reqId,
);
}
if (contentMd5 && contentMd5 !== streamed.md5Hash) {
await cleanupTempFile(streamed.tempPath);
return s3ErrorResponse(
'BadDigest',
'The Content-MD5 you specified did not match what we received.',
`/${bucket}/${key}`,
400,
reqId,
);
}
// M12: Reject oversized bodies
@@ -960,17 +973,17 @@ const handlePutObject = async (
}
// Idempotent PUT: if the object already exists, skip upload
const existing = await findFileByBucketAndKey(bucketRecord.id, key);
if (existing) {
await cleanupTempFile(streamed.tempPath);
return s3Response(null, 200, reqId, { etag: `"${streamed.fileHash}"` });
}
try {
const existing = await findFileByBucketAndKey(bucketRecord.id, key);
if (existing) {
await cleanupTempFile(streamed.tempPath);
return s3Response(null, 200, reqId, { etag: `"${streamed.fileHash}"` });
}
return await storeFileFromTemp(streamed, key, bucketRecord, contentType, reqId);
} catch (uploadError) {
} catch (error) {
await cleanupTempFile(streamed.tempPath);
throw uploadError;
throw error;
}
};
@@ -1022,11 +1035,15 @@ const storeFileFromTemp = async (
return s3Response(null, 200, reqId, { etag: `"${file.fileHash}"` });
}
const forwardResult = await botPool.forwardToStorage(
createReadStream(streamed.tempPath),
partFileNamePrefix,
'document',
);
const fileStream = createReadStream(streamed.tempPath);
let forwardResult: ForwardResult;
try {
forwardResult = await botPool.forwardToStorage(fileStream, partFileNamePrefix, 'document');
} catch (error) {
fileStream.destroy();
throw error;
}
fileStream.destroy();
const publicId = nanoid();
const { db, files: fileSchema } = await import('../../../db/index');