AWS instrumentation for SDK v3, S3

[estolfo]

Provide instrumentation for AWS SDK version 3 S3 client

See #2124 for draft work.

[david-luna]

Reviewed @aws-sdk/node-http-handler package which is used by the clients (S3, SNS, SQS) and instrumented it. The approach is to see if there is a way to identify the client from the request and only instrument node-http-handler module for all AWS clients

After instrumenting it I saw the requests have the following shapes

For ListBucketsCommand request

HttpRequest {
  method: 'GET',
  hostname: 's3.us-east-2.amazonaws.com',
  port: undefined,
  query: {},
  headers: {
    'content-type': 'application/xml',
    host: 's3.us-east-2.amazonaws.com',
    'x-amz-user-agent': 'aws-sdk-js/3.319.0',
    'user-agent': 'aws-sdk-js/3.319.0 os/darwin/22.4.0 lang/js md/nodejs/18.16.0 api/s3/3.319.0',
    'amz-sdk-invocation-id': '001e0d75-xxx',
    'amz-sdk-request': 'attempt=1; max=3',
    'x-amz-date': '20230425T171036Z',
    'x-amz-content-sha256': 'e3b...',
    authorization: 'AWS4-HMAC-SHA256 Credential=...'
  },
  body: '',
  protocol: 'https:',
  path: '/'
}

For CreateBucketCommand request

HttpRequest {
  method: 'PUT',
  hostname: 's3.us-east-2.amazonaws.com',
  port: undefined,
  query: {},
  headers: {
    'content-type': 'application/xml',
    'content-length': '497',
    Expect: '100-continue',
    host: 's3.us-east-2.amazonaws.com',
    'x-amz-user-agent': 'aws-sdk-js/3.319.0',
    'user-agent': 'aws-sdk-js/3.319.0 os/darwin/22.4.0 lang/js md/nodejs/18.16.0 api/s3/3.319.0',
    'amz-sdk-invocation-id': '50a4183f-xxx',
    'amz-sdk-request': 'attempt=1; max=3',
    'x-amz-date': '20230425T171037Z',
    'x-amz-content-sha256': '544f...',
    authorization: 'AWS4-HMAC-SHA256 Credential=...'
  },
  body: '<?xml version="1.0" encoding="UTF-8"?><CreateBucketConfiguration xmlns="http://s3.amazonaws.com/doc/2006-03-01/"><LocationConstraint>async () =&gt; {&#x0A;            if (typeof region === &quot;string&quot;) {&#x0A;                return (0, getRealRegion_1.getRealRegion)(region);&#x0A;            }&#x0A;            const providedRegion = await region();&#x0A;            return (0, getRealRegion_1.getRealRegion)(providedRegion);&#x0A;        }</LocationConstraint></CreateBucketConfiguration>',
  protocol: 'https:',
  path: '/elasticapmtest-bucket-20230425171036/'
}

For DeleteBucketCommand request

HttpRequest {
  method: 'DELETE',
  hostname: 's3.us-east-2.amazonaws.com',
  port: undefined,
  query: {},
  headers: {
    host: 's3.us-east-2.amazonaws.com',
    'x-amz-user-agent': 'aws-sdk-js/3.319.0',
    'user-agent': 'aws-sdk-js/3.319.0 os/darwin/22.4.0 lang/js md/nodejs/18.16.0 api/s3/3.319.0',
    'amz-sdk-invocation-id': '9a5fa4a9-xxx',
    'amz-sdk-request': 'attempt=1; max=3',
    'x-amz-date': '20230425T171819Z',
    'x-amz-content-sha256': 'e3b...',
    authorization: 'AWS4-HMAC-SHA256 Credential=...'
  },
  body: undefined,
  protocol: 'https:',
  path: '/elasticapmtest-bucket-20230425171819/'
}

In this examples we can see Bucket is used as part of the request path so IMO it's going to be tricky to differentiate. We see some references of S3 in the hostname and host header but I think is not enough

[trentm]

IMO it's going to be tricky to differentiate.

I agree. That looks like it will be too low-level a place to do our instrumentation. We don't want to have to reverse map from the HTTP request shape to the S3 (or SQS, SNS, etc.) API call name.