Feature request: additional metrics

[tjcelaya]

There are three areas identified where we want more enhancements for metrics:

Response Codes

Referenced from this comment in #418, we should also track metrics around unexpected response codes. This will require some thought, questions to answer include (but are not limited to):

• does the expected response code factor into the metric at all?
  • e.g. expecting-200-got-403 vs expecting-206-got-403
• does the request method factor into the metric?
  • e.g. 500 on PUT vs 500 on GET

Server-side response time

From the same comment, we should see if responses include a header for the server's perceived response time and expose that in a similar fashion as the response-$METHOD metrics. minfo includes the following two headers, though we need to know exactly what they mean and which we want to expose:

• x-response-time
• x-request-processing-time

Bandwidth tracking

Similarly to the way COSBench tracks data throughput we might want to track the number of bytes read from or written to Manta.

Detailed timeouts

We need to clarify what timeouts apply to which scenarios and determine if it's possible to tell a TCP timeout from an HTTP timeout based on the exception thrown. I suspect this will really just amount to a SocketTimeoutException being thrown instead of a NoHttpResponseException but regardless we should document the relationship between different kinds of timeouts and the resulting exceptions.