Report duration of failed retrievals

[bajtos]

At the moment, most retrievals (>99.99%) fail. We are not measuring the duration of failed retrievals, and therefore we don't know how many tasks can an honest checker node complete every round.

Let's start collecting that data.

• Modify spark checker to report duration for failed retrievals too.

  Note: we should be collecting this data, but apparently, some measurements come with an invalid end_at value. See Report retrieval network errors #43 (comment)

  Let's ensure end_at is always set correctly.

• Modify spark-evaluate to produce two retrieval stats - duration of successful requests and duration of all requests.

• Figure out how to handle measurements with end_at set to Date(0) - they are clearly invalid, but why are we receiving them? Are they produced by fraudulent nodes?

[bajtos]

Oh! We don’t have any catch block, that’s why we are not setting end_at. And I am guessing that PG or node-pg converts null to Date(0).

spark/lib/spark.js

Lines 97 to 99 in a609849

	} finally {
	clearTimeout(timeout)
	}

See also #43 (comment)

[bajtos]

Figure out how to handle measurements with end_at set to Date(0) - they are clearly invalid, but why are we receiving them? Are they produced by fraudulent nodes?

This will be fixed by filecoin-station/spark-api#160