feat: cached public status snapshot with licence gate

server/internal/services/statussnapshot.go

@@ -1,10 +1,15 @@
 package services
 
 import (
+	"encoding/json"
+	"log"
 	"sort"
 	"time"
 
+	"gitea.hostxtra.co.uk/mrhid6/vantage/server/internal/db"
 	"gitea.hostxtra.co.uk/mrhid6/vantage/server/internal/models"
+	"gitea.hostxtra.co.uk/mrhid6/vantage/shared/license"
+	"go.mongodb.org/mongo-driver/v2/bson"
 )
 
 // HistoryDays is the width of the public history bar. It is also the window the
@@ -368,6 +373,166 @@ func componentStatus(mon models.Monitor, known, inMaintenance bool, now time.Tim
 	}
 }
 
+const statusCacheTTL = 30 * time.Second
+
+// PublicStatusSnapshot is the whole public read path.
+//
+// A missing page, an unpublished page and a page belonging to another instance
+// all return ErrPageNotFound, identically. A distinct error for "exists but
+// unpublished" would confirm it exists.
+func PublicStatusSnapshot(instanceID, pageID string) (*StatusSnapshot, error) {
+	if err := ValidatePageID(pageID); err != nil {
+		return nil, ErrPageNotFound
+	}
+	if snap := cachedSnapshot(instanceID, pageID); snap != nil {
+		return snap, nil
+	}
+
+	page, err := GetStatusPage(instanceID, pageID)
+	if err != nil {
+		return nil, err
+	}
+	if !page.Published {
+		return nil, ErrPageNotFound
+	}
+
+	// The licence check answers 200 with available:false rather than 403,
+	// because the reader is a member of the public who can do nothing about it
+	// and deserves an explanation rather than a browser error.
+	st := GetLicenseState(instanceID)
+	if !st.Active() {
+		return &StatusSnapshot{Available: false, Reason: "licence_inactive", Title: page.Title}, nil
+	}
+	if !st.Feature(license.FeatureStatusPages) {
+		return &StatusSnapshot{Available: false, Reason: "feature_unavailable", Title: page.Title}, nil
+	}
+
+	in, err := loadSnapshotInput(instanceID, *page)
+	if err != nil {
+		return nil, err
+	}
+	snap := assembleSnapshot(in)
+	storeSnapshot(instanceID, pageID, snap)
+	return &snap, nil
+}
+
+func loadSnapshotInput(instanceID string, page models.StatusPage) (snapshotInput, error) {
+	in := snapshotInput{
+		Page:     page,
+		Monitors: map[string]models.Monitor{},
+		Rollups:  map[string][]models.Rollup{},
+		Now:      time.Now().UTC(),
+	}
+	since := in.Now.AddDate(0, 0, -HistoryDays)
+
+	ids := []string{}
+	for _, sec := range page.Sections {
+		for _, e := range sec.Entries {
+			ids = append(ids, e.MonitorID)
+		}
+	}
+	if len(ids) == 0 {
+		// A page with no components still renders: title, banner and any
+		// authored incidents. Skipping the monitor queries avoids three
+		// unbounded $in lookups on an empty list.
+		authored, err := ListStatusIncidentsForPage(instanceID, page.PageID, since)
+		if err != nil {
+			return in, err
+		}
+		in.Authored = authored
+		return in, nil
+	}
+
+	ctx, cancel := spCtx()
+	defer cancel()
+
+	cur, err := db.Col("monitors").Find(ctx, bson.M{
+		"instance_id": instanceID,
+		"monitor_id":  bson.M{"$in": ids},
+	})
+	if err != nil {
+		return in, err
+	}
+	mons := []models.Monitor{}
+	if err := cur.All(ctx, &mons); err != nil {
+		return in, err
+	}
+	for _, m := range mons {
+		in.Monitors[m.MonitorID] = m
+	}
+
+	rc, err := db.Col("monitor_rollups").Find(ctx, bson.M{
+		"instance_id":  instanceID,
+		"monitor_id":   bson.M{"$in": ids},
+		"period_start": bson.M{"$gte": since},
+	})
+	if err != nil {
+		return in, err
+	}
+	rollups := []models.Rollup{}
+	if err := rc.All(ctx, &rollups); err != nil {
+		return in, err
+	}
+	for _, r := range rollups {
+		in.Rollups[r.MonitorID] = append(in.Rollups[r.MonitorID], r)
+	}
+
+	ic, err := db.Col("incidents").Find(ctx, bson.M{
+		"instance_id": instanceID,
+		"monitor_id":  bson.M{"$in": ids},
+		"started_at":  bson.M{"$gte": since},
+	})
+	if err != nil {
+		return in, err
+	}
+	auto := []models.Incident{}
+	if err := ic.All(ctx, &auto); err != nil {
+		return in, err
+	}
+	in.AutoIncidents = auto
+
+	authored, err := ListStatusIncidentsForPage(instanceID, page.PageID, since)
+	if err != nil {
+		return in, err
+	}
+	in.Authored = authored
+	return in, nil
+}
+
+// A cache miss on Redis is a cache miss, never an error: the status page must
+// survive the outage it exists to report.
+func cachedSnapshot(instanceID, pageID string) *StatusSnapshot {
+	if statusRedis == nil {
+		return nil
+	}
+	ctx, cancel := spCtx()
+	defer cancel()
+	raw, err := statusRedis.Get(ctx, statusCacheKey(instanceID, pageID)).Bytes()
+	if err != nil || len(raw) == 0 {
+		return nil
+	}
+	var snap StatusSnapshot
+	if err := json.Unmarshal(raw, &snap); err != nil {
+		return nil
+	}
+	return &snap
+}
+
+func storeSnapshot(instanceID, pageID string, snap StatusSnapshot) {
+	if statusRedis == nil {
+		return
+	}
+	raw, err := json.Marshal(snap)
+	if err != nil {
+		return
+	}
+	ctx, cancel := spCtx()
+	defer cancel()
+	if err := statusRedis.Set(ctx, statusCacheKey(instanceID, pageID), raw, statusCacheTTL).Err(); err != nil {
+		log.Printf("status cache store %s/%s: %v", instanceID, pageID, err)
+	}
+}
+
 func overallState(sections []PublicSection) string {
 	worst := PublicUp
 	anyDown, anyMaint, anyOther := false, false, false