feat: authored status incidents and maintenance windows

2026-08-24 14:18:06 +00:00
parent 6f998ff506
commit 161835802d
1 changed files with 257 additions and 0 deletions
@@ -0,0 +1,257 @@
+package services
+
+import (
+	"errors"
+	"fmt"
+	"time"
+
+	"gitea.hostxtra.co.uk/mrhid6/vantage/server/internal/db"
+	"gitea.hostxtra.co.uk/mrhid6/vantage/server/internal/models"
+	"github.com/google/uuid"
+	"go.mongodb.org/mongo-driver/v2/bson"
+	"go.mongodb.org/mongo-driver/v2/mongo"
+	"go.mongodb.org/mongo-driver/v2/mongo/options"
+)
+
+var ErrIncidentNotFound = errors.New("status incident not found")
+
+var incidentStatuses = map[string]bool{
+	models.IncidentInvestigating: true,
+	models.IncidentIdentified:    true,
+	models.IncidentMonitoring:    true,
+	models.IncidentResolved:      true,
+}
+
+var maintenanceStatuses = map[string]bool{
+	models.MaintenanceScheduled:  true,
+	models.MaintenanceInProgress: true,
+	models.MaintenanceCompleted:  true,
+}
+
+var impacts = map[string]bool{
+	models.ImpactNone: true, models.ImpactMinor: true,
+	models.ImpactMajor: true, models.ImpactCritical: true,
+}
+
+func validateIncident(inc *models.StatusIncident) error {
+	if inc.Title == "" {
+		return errors.New("title is required")
+	}
+	if len(inc.PageIDs) == 0 {
+		return errors.New("at least one page is required")
+	}
+	switch inc.Kind {
+	case models.StatusKindIncident:
+		if !incidentStatuses[inc.Status] {
+			return fmt.Errorf("invalid incident status %q", inc.Status)
+		}
+	case models.StatusKindMaintenance:
+		if !maintenanceStatuses[inc.Status] {
+			return fmt.Errorf("invalid maintenance status %q", inc.Status)
+		}
+		if inc.ScheduledStart == nil || inc.ScheduledEnd == nil {
+			return errors.New("maintenance needs a scheduled start and end")
+		}
+		if !inc.ScheduledEnd.After(*inc.ScheduledStart) {
+			return errors.New("maintenance must end after it starts")
+		}
+	default:
+		return fmt.Errorf("invalid kind %q", inc.Kind)
+	}
+	if inc.Impact == "" {
+		inc.Impact = models.ImpactNone
+	}
+	if !impacts[inc.Impact] {
+		return fmt.Errorf("invalid impact %q", inc.Impact)
+	}
+	return nil
+}
+
+func CreateStatusIncident(instanceID string, inc *models.StatusIncident) (*models.StatusIncident, error) {
+	if err := validateIncident(inc); err != nil {
+		return nil, err
+	}
+	inc.ID = bson.ObjectID{}
+	inc.InstanceID = instanceID
+	inc.IncidentID = uuid.NewString()
+	now := time.Now()
+	inc.CreatedAt = now
+	inc.UpdatedAt = now
+	if inc.StartedAt.IsZero() {
+		if inc.Kind == models.StatusKindMaintenance && inc.ScheduledStart != nil {
+			inc.StartedAt = *inc.ScheduledStart
+		} else {
+			inc.StartedAt = now
+		}
+	}
+	if inc.Updates == nil {
+		inc.Updates = []models.StatusIncidentUpdate{}
+	}
+
+	ctx, cancel := spCtx()
+	defer cancel()
+	if _, err := db.Col("status_incidents").InsertOne(ctx, inc); err != nil {
+		return nil, err
+	}
+	invalidatePages(instanceID, inc.PageIDs)
+	return inc, nil
+}
+
+func invalidatePages(instanceID string, pageIDs []string) {
+	for _, p := range pageIDs {
+		InvalidateStatusCache(instanceID, p)
+	}
+}
+
+func getIncident(instanceID, incidentID string) (*models.StatusIncident, error) {
+	ctx, cancel := spCtx()
+	defer cancel()
+	var inc models.StatusIncident
+	err := db.Col("status_incidents").
+		FindOne(ctx, bson.M{"instance_id": instanceID, "incident_id": incidentID}).
+		Decode(&inc)
+	if errors.Is(err, mongo.ErrNoDocuments) {
+		return nil, ErrIncidentNotFound
+	}
+	if err != nil {
+		return nil, err
+	}
+	return &inc, nil
+}
+
+func ListStatusIncidents(instanceID, pageID string) ([]models.StatusIncident, error) {
+	filter := bson.M{"instance_id": instanceID}
+	if pageID != "" {
+		filter["page_ids"] = pageID
+	}
+	ctx, cancel := spCtx()
+	defer cancel()
+	cur, err := db.Col("status_incidents").Find(ctx, filter,
+		options.Find().SetSort(bson.M{"started_at": -1}))
+	if err != nil {
+		return nil, err
+	}
+	out := []models.StatusIncident{}
+	if err := cur.All(ctx, &out); err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+// ListStatusIncidentsForPage is the public read's query: one page, bounded by
+// the history window, so a five-year-old instance does not assemble five years
+// of incidents on every cache miss.
+func ListStatusIncidentsForPage(instanceID, pageID string, since time.Time) ([]models.StatusIncident, error) {
+	ctx, cancel := spCtx()
+	defer cancel()
+	cur, err := db.Col("status_incidents").Find(ctx, bson.M{
+		"instance_id": instanceID,
+		"page_ids":    pageID,
+		"$or": []bson.M{
+			{"started_at": bson.M{"$gte": since}},
+			{"resolved_at": nil},
+			{"status": models.MaintenanceScheduled},
+		},
+	}, options.Find().SetSort(bson.M{"started_at": -1}))
+	if err != nil {
+		return nil, err
+	}
+	out := []models.StatusIncident{}
+	if err := cur.All(ctx, &out); err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func UpdateStatusIncident(instanceID, incidentID string, inc *models.StatusIncident) (*models.StatusIncident, error) {
+	existing, err := getIncident(instanceID, incidentID)
+	if err != nil {
+		return nil, err
+	}
+	inc.Kind = existing.Kind // kind is fixed at creation
+	if err := validateIncident(inc); err != nil {
+		return nil, err
+	}
+
+	set := bson.M{
+		"page_ids":          inc.PageIDs,
+		"title":             inc.Title,
+		"impact":            inc.Impact,
+		"affected_monitors": inc.AffectedMonitors,
+		"status":            inc.Status,
+		"scheduled_start":   inc.ScheduledStart,
+		"scheduled_end":     inc.ScheduledEnd,
+		"updated_at":        time.Now(),
+	}
+	if inc.Status == models.IncidentResolved || inc.Status == models.MaintenanceCompleted {
+		if existing.ResolvedAt == nil {
+			now := time.Now()
+			set["resolved_at"] = now
+		}
+	} else {
+		// Reopening clears it, so a mistakenly resolved incident does not keep
+		// a resolution time it no longer has.
+		set["resolved_at"] = nil
+	}
+
+	ctx, cancel := spCtx()
+	defer cancel()
+	if _, err := db.Col("status_incidents").UpdateOne(ctx,
+		bson.M{"instance_id": instanceID, "incident_id": incidentID},
+		bson.M{"$set": set}); err != nil {
+		return nil, err
+	}
+	// Both old and new page sets, or a page the incident was just removed from
+	// keeps showing it for up to 30 seconds.
+	invalidatePages(instanceID, existing.PageIDs)
+	invalidatePages(instanceID, inc.PageIDs)
+	return getIncident(instanceID, incidentID)
+}
+
+func AppendStatusIncidentUpdate(instanceID, incidentID, status, body, author string) (*models.StatusIncident, error) {
+	existing, err := getIncident(instanceID, incidentID)
+	if err != nil {
+		return nil, err
+	}
+	if body == "" {
+		return nil, errors.New("update body is required")
+	}
+	valid := incidentStatuses
+	if existing.Kind == models.StatusKindMaintenance {
+		valid = maintenanceStatuses
+	}
+	if !valid[status] {
+		return nil, fmt.Errorf("invalid status %q for a %s", status, existing.Kind)
+	}
+
+	upd := models.StatusIncidentUpdate{At: time.Now(), Status: status, Body: body, Author: author}
+	set := bson.M{"status": status, "updated_at": upd.At}
+	if status == models.IncidentResolved || status == models.MaintenanceCompleted {
+		set["resolved_at"] = upd.At
+	}
+
+	ctx, cancel := spCtx()
+	defer cancel()
+	if _, err := db.Col("status_incidents").UpdateOne(ctx,
+		bson.M{"instance_id": instanceID, "incident_id": incidentID},
+		bson.M{"$push": bson.M{"updates": upd}, "$set": set}); err != nil {
+		return nil, err
+	}
+	invalidatePages(instanceID, existing.PageIDs)
+	return getIncident(instanceID, incidentID)
+}
+
+func DeleteStatusIncident(instanceID, incidentID string) error {
+	existing, err := getIncident(instanceID, incidentID)
+	if err != nil {
+		return err
+	}
+	ctx, cancel := spCtx()
+	defer cancel()
+	if _, err := db.Col("status_incidents").DeleteOne(ctx,
+		bson.M{"instance_id": instanceID, "incident_id": incidentID}); err != nil {
+		return err
+	}
+	invalidatePages(instanceID, existing.PageIDs)
+	return nil
+}