将大型koa请求主体上传到AWS S3的正确方法是什么？

Question

我正在构建一个应用程序后端。客户端将文件作为请求主体发布到服务器，然后服务器将文件上载到AWS S3。服务器正在使用NodeJS和koa Web框架。

如果我使用raw-body来使post体缓冲，当文件很大时，缓冲区很大，并导致内存不足错误。 如果我直接将ctx.req（一个IncomingMessage对象）传递给S3.putObject，AWS SDK会抛出一个错误，说Cannot determine length of [object Object]，看起来像AWS SDK尝试获取流的长度然后开始多部分上传。

AWS开发工具包版本2.383.0（目前最新） NodeJS 10.14.2

这时，我写了一个函数，从IncomingMessage中读取一个流，等待数据事件填满一个大缓冲区（16MB），然后将多部分上传到S3，这很好地解决了问题，但我仍然在寻找为了更好的解决方

Answer 1

经过几个月的运行，我认为我的最终解决方案是稳定可靠的。

主要概念是从IncomingMessage流存储到缓冲区，缓冲区达到一个大小后，然后将当前部分放到S3，然后继续读取流直到结束。

const uploaderLogger = Log4js.getLogger('customUploader');
function customMultiPartUpload(s3, bucket, key, incomingMessage, partSizeInByte) {
    return new Promise((resolve) => {
        partSizeInByte = partSizeInByte || uploadBufferMB * 1024 * 1024;
        uploaderLogger.debug(`part size is ${partSizeInByte}`);

        let uploadId = null;
        let partNumber = 0;
        let parts = [];
        let fileSize = 0;
        let reserveBuffer = Buffer.alloc(0);
        const sendBuffer = Buffer.alloc(partSizeInByte);
        const md5Hash = Crypto.createHash('md5');

        const doUpload = async (uploadBuffer) => {
            if (!uploadId) {
                uploaderLogger.debug('multipart upload not initialized');
                const createData = await s3.createMultipartUpload({
                    Bucket: bucket,
                    Key: key
                }).promise();
                uploadId = createData.UploadId;
                uploaderLogger.debug(`uploadId ${uploadId}`);

                partNumber = 0;
            }
            fileSize += uploadBuffer.length;
            uploaderLogger.debug(`buffer length ${uploadBuffer.length}, total ${fileSize}`);

            partNumber += 1;
            uploaderLogger.debug(`part number ${partNumber}`);

            md5Hash.update(uploadBuffer);

            const partData = await s3.uploadPart({
                Bucket: bucket,
                Key: key,
                PartNumber: partNumber,
                UploadId: uploadId,
                Body: uploadBuffer
            }).promise();
            parts.push({
                PartNumber: partNumber,
                ETag: partData.ETag
            });
            uploaderLogger.debug(`etag ${partData.ETag}`);
        };

        incomingMessage.on('data', async (chunkBuffer) => {
            incomingMessage.pause();

            reserveBuffer = Buffer.concat([ reserveBuffer, chunkBuffer ]);
            if (reserveBuffer.length > partSizeInByte) {
                do {
                    reserveBuffer.copy(sendBuffer, 0, 0, partSizeInByte);
                    reserveBuffer = reserveBuffer.slice(partSizeInByte);
                    await doUpload(sendBuffer);
                } while (reserveBuffer.length > partSizeInByte);
            }

            incomingMessage.resume();
        });

        incomingMessage.on('end', async () => {
            uploaderLogger.debug('stream end');

            if (reserveBuffer.length > 0) {
                await doUpload(reserveBuffer);
            }

            if (uploadId) {
                uploaderLogger.debug('uploadId not null');
                await s3.completeMultipartUpload({
                    Bucket: bucket,
                    Key: key,
                    UploadId: uploadId,
                    MultipartUpload: {
                        Parts: parts
                    }
                }).promise();
                uploaderLogger.debug('multipart upload complete');
            }

            const hash = md5Hash.digest('hex');

            resolve({
                size: fileSize,
                hash: hash
            });
            uploaderLogger.debug(`return file size ${fileSize}, hash ${hash}`);
        });
    });
}

调整partSizeInByte以适应您的服务器内存使用量，当服务器处理许多请求时，太大的部件大小可能导致OOM，太小的部件尺寸可能小于S3部件限制。