Allow skipping the first 24 hours in relative release view

The first 24 hours of a release are extremely noisy, it's generally
best to ignore it when comparing releases after a sufficient amount
of time has passed.

docs/API.md

@@ -58,12 +58,12 @@ Query parameters:
 * `platform` (required): Platform for measure (e.g. `windows`)
 * `measure` (required): Measure identifier (e.g. `main_crashes`)
 * `interval` (required): Interval of data to gather, in seconds
-* `start` (optional): Starting point to gather measure data from. If
-  not specified, will return `interval` worth of data, counting back
-  from the time of the query. This parameter is ignored if `relative`
-  is specified (see below)
 * `relative` (optional): If true (specified and non-zero), return results
   *from* the time of release of the latest version
+* `start` (optional): Starting point to gather measure data from. If
+  not specified, will return `interval` worth of data, counting either back
+  from the time of the query (non-relative) or up to the specified time
+  interval (relative).
 * `version` (optional): Retrieve only data particular to a specific version.
   May be specified multiple times.
 

missioncontrol/api/views.py

@@ -179,13 +179,17 @@ def measure(request):
                 'version': version,
                 'data': []
             }
+            if start:
+                start_timestamp = base_timestamp + datetime.timedelta(seconds=int(start))
+            else:
+                start_timestamp = base_timestamp
             ret[build_id]['data'] = [
                 [int((timestamp - base_timestamp).total_seconds()), value, usage_hours] for
                 (timestamp, value, usage_hours) in datums.filter(
                     series__build__version=version,
                     series__build__build_id=build_id,
-                    timestamp__range=(base_timestamp,
-                                      base_timestamp + datetime.timedelta(seconds=int(interval)))
+                    timestamp__range=(start_timestamp,
+                                      start_timestamp + datetime.timedelta(seconds=int(interval)))
                 ).order_by('timestamp').values_list('timestamp', 'value', 'usage_hours')]
 
     return JsonResponse(data={'measure_data': ret})

tests/test_api.py

@@ -207,24 +207,30 @@ def test_get_measure(fake_measure_data, client):
 
 
 @pytest.mark.parametrize('interval', [86400, 300, 0])
+@pytest.mark.parametrize('start', [None, 0, 250, 301])
 @freeze_time('2017-07-01 13:00')
-def test_compare(fake_measure_data_offset, client, interval):
+def test_compare(fake_measure_data_offset, client, interval, start):
     (platform, channel, measure) = ('linux', 'release', 'main_crashes')
 
-    resp = client.get(reverse('measure'), {
+    params = {
         'platform': platform,
         'channel': channel,
         'measure': measure,
         'interval': interval,
         'relative': 1
-    })
+    }
+    if start is not None:
+        params.update({'start': start})
+    resp = client.get(reverse('measure'), params)
     # despite the samples being captured at different times, they should
     # return the same relative value for compare
+    min_time = start or 0
     expected_data = [
         datum for datum in
         [[0, 100.0, 20.0],
          [300, 10.0, 16.0],
-         [600, 10.0, 20.0]] if not interval or datum[0] <= interval
+         [600, 10.0, 20.0]] if datum[0] >= min_time and
+        (not interval or datum[0] <= (min_time + interval))
     ]
     assert resp.json()['measure_data'] == {
         '20170620075044': {