What exactly is the purpose of allocationTag when allocating memory

[specious]

The comment for Aws::MakeShared() in aws/core/utils/memory/stl/AWSAllocator.h says:

aws-sdk-cpp/aws-cpp-sdk-core/include/aws/core/utils/memory/stl/AWSAllocator.h

Line 44 in 3f2a797

* allocationTag is for memory tracking purposes.

What exactly is the use of this allocation tag, which is a somewhat conspicuous parameter in the various memory allocation functions?

https://github.com/aws/aws-sdk-cpp/search?q=allocationTag

In Aws::MakeShared() it's actually marked as an unused parameter:

aws-sdk-cpp/aws-cpp-sdk-core/include/aws/core/utils/memory/stl/AWSAllocator.h

Line 49 in 3f2a797

AWS_UNREFERENCED_PARAM(allocationTag);

Does that mean we might as well just pass nullptr as the allocationTag?

[jmklix]

This is just used for logging purposes. Whatever value that you assign it will show up in the debug logs. I would recommend using something unique and recognizable rather than a nullptr.

[jmklix]

Here is how to turn logging on:

SDKOptions options;
options.loggingOptions.logLevel = Aws::Utils::Logging::LogLevel::Info;
Aws::InitAPI(options);
//do SDK stuff;
Aws::ShutdownAPI(options);

[specious]

I see that there are multiple log levels of one which can be selected:

aws-sdk-cpp/aws-cpp-sdk-core/include/aws/core/utils/logging/LogLevel.h

Lines 22 to 31 in 7030834

	enum class LogLevel : int
	{
	Off = 0,
	Fatal = 1,
	Error = 2,
	Warn = 3,
	Info = 4,
	Debug = 5,
	Trace = 6
	};

However, even when I set:

Aws::SDKOptions options;
options.loggingOptions.logLevel = Aws::Utils::Logging::LogLevel::Trace;
Aws::InitAPI(options);

I still don't see anything from this in the CloudWatch logs:

[jmklix]

Those logs only show up in the terminal while cloudwatch only logs specific events. You can add more info to the cloudwatch logs by following this

[specious]

Thank you for all the excellent information.

I have found the solution to having AWS SDK log messages in CloudWatch.

In the developer guide it says that by default:

Once logging is enabled in your application, the SDK will generate log files in your executable directory following the default naming pattern of aws_sdk_<date>.log. The log file generated by the prefix-naming option rolls over once per hour to allow for archiving or deleting log files.

This, however, is of little use on an AWS Lambda server, as the instance is ephemeral.

While I was perusing the docs, I noticed that the default logger is derived from Aws::Utils::Logging::FormattedLogSystem, and in the documentation for its ProcessFormattedStatement() member function, it is mentioned that it is possible to derive your own custom logger from it which can do whatever you like with the log messages, and it also mentioned there that there is another very simple logger, also provided by the SDK, that is derived from it called Aws::Utils::Logging::ConsoleLogSystem which in fact writes the log messages directly to stdout.

So I tried it similar to this example:

#include <aws/core/utils/logging/ConsoleLogSystem.h>

// ...

//
// Main entry point on Lambda instantiation (cold start)
//

int main() {
  std::cout << ":: cold starting lambda function\n";

  Aws::SDKOptions options;

  //
  // Set logging level
  //
  // Levels: 0ff | Fatal | Error | Warn | Info | Debug | Trace
  //
  // See: aws-cpp-sdk-core/include/aws/core/utils/logging/LogLevel.h
  //
  Aws::Utils::Logging::LogLevel logLevel = Aws::Utils::Logging::LogLevel::Info;

  //
  // Change to the console logger to see the log messages in the CloudWatch console
  //
  options.loggingOptions.logger_create_fn = [&logLevel] {
    return Aws::MakeShared<Aws::Utils::Logging::ConsoleLogSystem>(
      "console_logger", logLevel
    ); 
  };

  //
  // This is the master setting for the log level
  //
  // (but due to what appears to be a bug in AWS SDK 1.9.339 it currently works just as a toggle switch)
  //
  options.loggingOptions.logLevel = logLevel;

  Aws::InitAPI(options);

  // ...
}

And voilà, I see AWS SDK messages in my CloudWatch logs.

[specious]

Actually, at least as of the current latest SDK version (1.9.339), if you are passing a custom factory function to create the logger, then it doesn't matter what loggingOptions.logLevel is set to, as long as it's not set to Off (it essentially becomes a toggle switch that takes effect when you call Aws::InitAPI()):

aws-sdk-cpp/aws-cpp-sdk-core/source/Aws.cpp

Lines 33 to 43 in 367f0d2

	if(options.loggingOptions.logLevel != Aws::Utils::Logging::LogLevel::Off)
	{
	if(options.loggingOptions.logger_create_fn)
	{
	Aws::Utils::Logging::InitializeAWSLogging(options.loggingOptions.logger_create_fn());
	}
	else
	{
	Aws::Utils::Logging::InitializeAWSLogging(
	Aws::MakeShared<Aws::Utils::Logging::DefaultLogSystem>(ALLOCATION_TAG, options.loggingOptions.logLevel, options.loggingOptions.defaultLogPrefix));
	}

This means that (with the current SDK version) the following code works just as well:

  Aws::SDKOptions options;

  options.loggingOptions.logLevel = Aws::Utils::Logging::LogLevel::Fatal; // <-- this just needs to be not `Off`, otherwise logging will never be initialized

  options.loggingOptions.logger_create_fn = [] {
    return Aws::MakeShared<Aws::Utils::Logging::ConsoleLogSystem>(
      "console_logger", // <-- tag of your choice
      Aws::Utils::Logging::LogLevel::Debug // <-- this sets the level
    ); 
  };

  Aws::InitAPI(options); // <-- logging will be turned on here if `loggingOptions.logLevel` is not `Off`

[github-actions[bot]]

Hello! Reopening this discussion to make it searchable.