1

Using AWS S3 C++ SDK for uploading .jpg images to a certain IAM user introduce huge time delays that in any case are caused due to network traffic and latency issues. I am using free-tier S3 version and MSVC 2017 64bit for my application (on Windows 10 PC). Here is a sample code:

Aws::SDKOptions options;
Aws::InitAPI(options);

Aws::Client::ClientConfiguration config;
config.region = Aws::Region::US_EAST_2;

Aws::S3::S3Client s3_client(Aws::Auth::AWSCredentials(KEY,ACCESS_KEY), config);

const Aws::String bucket_name = BUCKET;
const Aws::String object_name = "image.jpg";

Aws::S3::Model::PutObjectRequest put_object_request;

put_object_request.SetBucket(bucket_name);
put_object_request.SetKey(object_name);

std::shared_ptr<Aws::IOStream> input_data =
        Aws::MakeShared<Aws::FStream>("PutObjectInputStream",
                                      "../image.jpg",
                                      std::ios_base::in | std::ios::binary);

put_object_request.SetBody(input_data);
put_object_request.SetContentType("image/jpeg");

input_data->seekg(0LL, input_data->end);
put_object_request.SetContentLength(static_cast<long>(input_data->tellg()));

auto put_object_outcome = s3_client.PutObject(put_object_request);

When I upload images bigger than 100KB the total

PutObject(put_object_request);

time of execution exceeds 2min for a 520KB image.

I have tried the same example using Python boto3 and the total upload time for the same image is around 25s.

Have anyone faced the same issue?

thanasispap
  • 43
  • 10
  • I think there is some SDK method to break the image and upload all parts simultaneously – Ojasvi Monga Jul 11 '19 at 13:04
  • There is UploadPartAsync (I think it refers to objects greater of 5MB). Nevertheless the total time for one such small file and also the time difference between Python and C++ implementation makes no sense for me. – thanasispap Jul 11 '19 at 13:10
  • I dont think comparing across languages is a good idea. – Ojasvi Monga Jul 11 '19 at 13:13
  • Try to write a program to calculate 1000 primes. The time taken by a python and a CPP program will probably be quite different. Some languages are just slower than others. It is a simple fact. – Ojasvi Monga Jul 11 '19 at 13:16
  • I agree, but the reason I indicate that is that I am using the same SDK. – thanasispap Jul 11 '19 at 13:17
  • 1
    The multipart Upload object API is [available](https://docs.aws.amazon.com/AmazonS3/latest/dev/sdksupportformpu.html) only for Java, PHP, .NET. The functions tho seems to be available on C++ API but there is no example of such use. – thanasispap Jul 15 '19 at 08:28

1 Answers1

0

After a better look to AWS github repo I figure out the issue.

The problem was that WinHttpSyncHttpClient was making timouts and reset upload activity internally thus not exiting the upload thread and finally aborting the transaction. By adding a custom timeout value the problem solved.

I used multipart Upload to re-implement the example as it seems more robust and manageable. Although I thought that it is unavailable for C++ SDK, its not the case as the TransferManager does the same job for C++ (not by using S3 headers as Java, .NET and PHP does).

Thanks KaibaLopez and SCalwas from AWS github repo who help me solve the issues (issue1, issue2). I am pasting an example code is case anyone face the same issue:

#include "pch.h"
#include <iostream>
#include <fstream>
#include <filesystem>

#include <aws/core/Aws.h>
#include <aws/core/auth/AWSCredentials.h>
#include <aws/s3/S3Client.h>
#include <aws/s3/model/Bucket.h>
#include <aws/transfer/TransferManager.h>
#include <aws/transfer/TransferHandle.h>

static const char* KEY = "KEY";
static const char* BUCKET = "BUCKET_NAME";
static const char* ACCESS_KEY = "AKEY";
static const char* OBJ_NAME = "img.jpg";
static const char* const ALLOCATION_TAG = "S3_SINGLE_OBJ_TEST";

int main()
{
    Aws::SDKOptions options;
    Aws::InitAPI(options);
    {
        Aws::Client::ClientConfiguration config;
        config.region = Aws::Region::US_EAST_2;
        config.requestTimeoutMs = 20000;

        auto s3_client = std::make_shared<Aws::S3::S3Client>(Aws::Auth::AWSCredentials(KEY, ACCESS_KEY), config);

        const Aws::String bucket_name = BUCKET;
        const Aws::String object_name = OBJ_NAME;
        const Aws::String key_name = OBJ_NAME;

        auto s3_client_executor = Aws::MakeShared<Aws::Utils::Threading::DefaultExecutor>(ALLOCATION_TAG);
        Aws::Transfer::TransferManagerConfiguration trConfig(s3_client_executor.get());
        trConfig.s3Client = s3_client;


        trConfig.uploadProgressCallback =
            [](const Aws::Transfer::TransferManager*, const std::shared_ptr<const Aws::Transfer::TransferHandle>&transferHandle)
        { std::cout << "Upload Progress: " << transferHandle->GetBytesTransferred() <<
            " of " << transferHandle->GetBytesTotalSize() << " bytes" << std::endl;};

        std::cout << "File start upload" << std::endl;

        auto tranfer_manager = Aws::Transfer::TransferManager::Create(trConfig);
        auto transferHandle = tranfer_manager->UploadFile(object_name.c_str(),
            bucket_name.c_str(), key_name.c_str(), "multipart/form-data", Aws::Map<Aws::String, Aws::String>());

        transferHandle->WaitUntilFinished();

        if(transferHandle->GetStatus() == Aws::Transfer::TransferStatus::COMPLETED)
            std::cout << "File up" << std::endl;
        else
            std::cout << "Error uploading: " << transferHandle->GetLastError() << std::endl;
    }
    Aws::ShutdownAPI(options);
    return 0;
}
thanasispap
  • 43
  • 10