Failed setting TCP keep-alive interval with error code: 12018

[azuraw10]

Describe the bug

Creating an instance of Aws::Client::ClientConfiguration causes Failed setting TCP keep-alive interval with error code: 12018 warning message when region is fetched from the EC2 Instance Metadata Service. There is an option to get rid of the warnings by setting enableTcpKeepAlive to false once the object is created, but the problem is that the warnings are generated in the constructor and when executing the constructor enableTcpKeepAlive is always set to true. This causes problems on Windows, since the tcp keep alive setting on WinHttp is not available.

Expected Behavior

No warnings Failed setting TCP keep-alive interval with error code: 12018 should be generated. It should be possible to set enableTcpKeepAlive to false when creating the object (eg. pass enableTcpKeepAlive as a constructor parameter).

Current Behavior

Warning messages Failed setting TCP keep-alive interval with error code: 12018 are generated when Aws::Client::ClientConfiguration is created.

Reproduction Steps

int main()
{
    Aws::SDKOptions options;
    options.loggingOptions.logLevel = Aws::Utils::Logging::LogLevel::Warn;

    Aws::InitAPI(options);

    Aws::Client::ClientConfiguration config;
    config.enableTcpKeepAlive = false;

    ShutdownAPI(options);
}

Note: You need to make sure that the region will be fetched from the EC2 Instance Metadata Service.

Possible Solution

Since the tcp keep alive setting on WinHttp is not available, we can just set enableTcpKeepAlive to false by default on Windows:

@@ -45,7 +45,11 @@ void setLegacyClientConfigurationParameters(ClientConfiguration& clientConfig)
     clientConfig.httpRequestTimeoutMs = 0;
     clientConfig.requestTimeoutMs = 3000;
     clientConfig.connectTimeoutMs = 1000;
+    #ifdef _WIN32
+    clientConfig.enableTcpKeepAlive = false;
+    #else
     clientConfig.enableTcpKeepAlive = true;
+    #endif // _WIN32
     clientConfig.tcpKeepAliveIntervalMs = 30000;
     clientConfig.lowSpeedLimit = 1;
     clientConfig.proxyScheme = Aws::Http::Scheme::HTTP;

Another possibility is to pass enableTcpKeepAlive as a Aws::Client::ClientConfiguration constructor parameter.

Additional Information/Context

No response

AWS CPP SDK version used

1.9.234

Compiler and Version used

Microsoft Visual C++ Compiler 16.11.32228.343

Operating System and version

Windows Server 2022 20348.1607

[yasminetalby]

Hello @azuraw10 ,

Thank you very much for your submission.
The behavior you are experiencing is expected. As Tcp keep alive setting on WinHttp is not available, the team has previously decided to communicate that we can't turn off keep-alive on Windows as this is not a fatal error. You can see this in previous issue submission : #1062
We have added documentation for this in the following PR : b045497

Other general points of guidance:

1. All AWS SDK for C++ calls performed between Aws::InitAPI and Aws::ShutdownAPI should either to be contained within a pair of curly braces or should be invoked by functions called between the two methods.

A basic sample example is :

#include <aws/core/Aws.h>
int main(int argc, char** argv)
{
   Aws::SDKOptions options;
   Aws::InitAPI(options);
   {
      // make your SDK calls here.
   }
   Aws::ShutdownAPI(options);
   return 0;
}

You can checkout basic use recommendations in our developer guide: https://docs.aws.amazon.com/sdk-for-cpp/v1/developer-guide/basic-use.html

2. It seems that you are using an old version of the SDK (1.9.234). I would recommend upgrading to the latest version as we regularly release bug fixes and new features. The latest SDK version is : 1.11.113

Are you experiencing any issues or limitation caused by the current SDK behavior?

Thank you very much for your collaboration and feedback,

Sincerely,

Yasmine

[azuraw10]

Hi @yasminetalby
Thank you for the quick response.

Are you experiencing any issues or limitation caused by the current SDK behavior?

Unfortunately yes, we are storing the logs and this particular warning is flooding our storage systems and monitoring systems - Failed setting TCP keep-alive interval with error code: 12018 warning message occurred 126000 times in the last 30 days in our system. We cannot allow ourselves just ignore the warning, but we need to handle this one as it is the most frequent warning in our system. Since it does not make sense to turn on keepAlive on Windows, why the solution I suggest is not acceptable?

[yasminetalby]

Hello @azuraw10 ,

Thank you very much for providing feedback and the limitation you are experiencing.
We are currently prioritizing backlog items for the team based on user impact, as this is not a fatal error but rather a warning and had not caused any issue in the past I wanted to check if you were experiencing limitations to know how to best prioritize this item. This is definitely something we can change as an improvement to the current behavior to accommodate your needs. I will relabel this as a feature-request.

Thank you very much for your time and collaboration.
Sincerely,

Yasmine

[Mike4Online]

@yasminetalby ,

I concur with the original poster that noisy warnings in log files are a burden as they often require effort to investigate. Such warnings should be eliminated if possible, or demoted to [INFO] messages if there is nothing significant to warn about,

You previously wrote:

As Tcp keep alive setting on WinHttp is not available, the team has previously decided to communicate that we can't turn off keep-alive on Windows

Just to clarify, WinHTTP does provide a TCP keepalive option, WINHTTP_OPTION_WEB_SOCKET_KEEPALIVE_INTERVAL, and the AWS SDK for C++ uses this option, but as you wrote it cannot be disabled.

The Microsoft documentation for WINHTTP_OPTION_WEB_SOCKET_KEEPALIVE_INTERVAL states:

The default value for WINHTTP_OPTION_WEB_SOCKET_KEEPALIVE_INTERVAL is read from HKLM:\SOFTWARE\Microsoft\WebSocket\KeepaliveInterval. If a value is not set, the default value of 30000 will be used.

It is not possible to have a lower keepalive interval than 15000 milliseconds. Using WinHttpSetOption to set a value lower than 15000 will return with ERROR_INVALID_PARAMETER.

It should be possible to modify the AWS SDK for C++ so it can only pass timeout values to WinHttpSetOption that are within a valid range, i.e.:

1. Revise the log message in aws-cpp-sdk-core/source/http/windows/WinHttpSyncHttpClient.cpp to also log the value of the keepAliveIntervalMs parameter, so any logged warnings would indicate what keepalive timeout value that WinHttpSetOption call is attempting to set (see PR b045497). We could expect to see warnings in the logs showing a keepalive timeout value less than 15000 (the minimum), or perhaps a value greater than some undocumented maximum value. But if we saw a valid value such as 30000, then we could investigate a possible bug in WinHTTP itself.
2. Determine a practical upper limit for WINHTTP_OPTION_WEB_SOCKET_KEEPALIVE_INTERVAL.
3. Add code to alter (or "coerce") the specified timeout value, if necessary, to ensure it falls within the valid range of values for WINHTTP_OPTION_WEB_SOCKET_KEEPALIVE_INTERVAL, immediately prior to calling WinHttpSetOption. The alteration of this timeout value could be logged as a warning.

[gahr]

I am also being hit by this, with my logs piling up. The weird thing is that I'm using the default value of 30000. The WinHttpSetOption call fails even if I change it to 15000, 0, or 20000.