Add check for uptake rate

[leplatrem]

Check that X% of clients get update after Y minutes

Would require https://bugzilla.mozilla.org/show_bug.cgi?id=1612712

[leplatrem]

I started to work on this, using this query.

Since timestamps go forward, we can assume that if a client received timestamp X, then it also received all X - N.
I was thinking of accumulating the number of received events by timestamp, and hacked quickly something like this:

    cumulated = {}

    for row in rows:
        total = row["total"]
        etag = row["received_timestamp"][1: -1]
        if etag not in cumulated:
        	cumulated[etag] = {
        		"published": utcfromtimestamp(int(etag)).isoformat(),
        		"first_seen": row["min_timestamp"],
        		"total": 0,
        	}

        cumulated[etag]["duration"] = (
            datetime.fromisoformat(row["max_timestamp"])
            - datetime.fromisoformat(cumulated[etag]["first_seen"])
        ).seconds / 60

        for e in cumulated.keys():
        	if e <= etag:
        		cumulated[etag]["total"] += total

Which would give:

  "1582329230774": {
    "published": "2020-02-21T23:53:50.774000+00:00",
    "start": "2020-02-21T23:50:00",
    "total": 9677877,
    "duration": 1010.0
  },
  "1582416087079": {
    "published": "2020-02-23T00:01:27.079000+00:00",
    "start": "2020-02-23T00:00:00",
    "total": 7711960,
    "duration": 1000.0
  },
  "1582549284992": {
    "published": "2020-02-24T13:01:24.992000+00:00",
    "start": "2020-02-24T13:00:00",
    "total": 3140898,
    "duration": 220.0
  },
  ...

Some questions arose, so I thought sharing them here (most likely with myself) would be helpful:

• how to properly pick the etag to be studied over the period? The closest after beginning of period? The one with the highest uptake?
• what does represent better the global health of the pipeline? Percentage of uptake after X minutes or number of minutes to reach X percentage?
• how to handle the variations of connected clients during nights and week-ends?
• if accumulation is the way to go, how to treat the same client reporting several updates over the studied period of time?