backend/instances.go: fix instance stats query

The instance stats query contained a bug in which it only returned
instances that have _last_ checked in within the given time window
instead of all _active_ instances during the window.

As a result, instance count was skewed and much lower than the actual
amount of active instances.

This change updates the quey from:
  last_check_for_updates > (ts - window_size)
  AND last_check_for_updates < ts
to
  last_check_for_updates > (ts - window_size)
  AND instance_created_ts < ts
.

The change also updates the default window size from 2 hours to one day
to better match our legacy instance count metrics.

Signed-off-by: Thilo Fromm <thilofromm@microsoft.com>

Author: t-lo

backend/pkg/api/instances.go

@@ -48,8 +48,8 @@ const (
 )
 
 const (
-	validityInterval postgresDuration = "1 days"
-	defaultInterval  time.Duration    = 2 * time.Hour
+	validityInterval     postgresDuration = "1 days"
+	defaultStatsInterval time.Duration    = 24 * time.Hour
 )
 
 // Instance represents an instance running one or more applications for which
@@ -646,19 +646,24 @@ func (api *API) instanceStatusHistoryQuery(instanceID, appID, groupID string, li
 
 // GetDefaultInterval returns the default interval used for instance stats queries.
 func (api *API) GetDefaultInterval() time.Duration {
-	return defaultInterval
+	return defaultStatsInterval
 }
 
-// instanceStatsQuery returns a SelectDataset prepared to return all instances
-// that have been checked in during a given duration from a given time.
+// instanceStatsQuery returns a SelectDataset prepared to return all active instances within a time period.
+//
+//	This is somewhat complicated by the fact that an instance created _before_ or _during_ our window
+//	can have checked in _after_ the window we're supposed to check. last_check_for_updates gets updated every
+//	time an instance checks for an update. Hence we filter for instances that:
+//	 1. have been created before timestamp 't'
+//	 2. have checked in _during_ or _after_ the time period
 func (api *API) instanceStatsQuery(t *time.Time, duration *time.Duration) *goqu.SelectDataset {
 	if t == nil {
 		now := time.Now().UTC()
 		t = &now
 	}
 
 	if duration == nil {
-		d := defaultInterval
+		d := defaultStatsInterval
 		duration = &d
 	}
 
@@ -710,12 +715,13 @@ func (api *API) instanceStatsQuery(t *time.Time, duration *time.Duration) *goqu.
 				Else("").
 				As("arch"),
 			goqu.C("version").As("version"),
-			goqu.COUNT("*").As("instances")).
+			goqu.COUNT("*").As("instances")).Distinct().
 		Join(goqu.T("groups"), goqu.On(goqu.C("group_id").Eq(goqu.T("groups").Col("id")))).
 		Join(goqu.T("channel"), goqu.On(goqu.T("groups").Col("channel_id").Eq(goqu.T("channel").Col("id")))).
+		Join(goqu.T("instance"), goqu.On(goqu.T("instance_application").Col("instance_id").Eq(goqu.T("instance").Col("id")))).
 		Where(
 			goqu.C("last_check_for_updates").Gt(timestampMinusDuration),
-			goqu.C("last_check_for_updates").Lte(timestamp)).
+			goqu.T("instance").Col("created_ts").Lte(timestamp)).
 		GroupBy(timestamp,
 			goqu.T("channel").Col("name"),
 			goqu.T("channel").Col("arch"),