Files
dockwatch/internal/monitor/groups.go
T
jbergner 45ca18b74e
release-tag / release-image (push) Failing after 1m20s
init
2026-08-31 17:09:21 +02:00

482 lines
13 KiB
Go

package monitor
import (
"context"
"database/sql"
"errors"
"fmt"
"regexp"
"strings"
"time"
)
type ProbeGroup struct {
ID int64 `json:"id"`
Name string `json:"name"`
Description string `json:"description"`
Status string `json:"status"`
Monitors []Monitor `json:"monitors,omitempty"`
CreatedAt int64 `json:"created_at"`
UpdatedAt int64 `json:"updated_at"`
}
type ProbeGroupInput struct {
Name string `json:"name"`
Description string `json:"description"`
MonitorIDs []int64 `json:"monitor_ids"`
}
type StatusPage struct {
ID int64 `json:"id"`
Name string `json:"name"`
Slug string `json:"slug"`
Description string `json:"description"`
Status string `json:"status"`
Enabled bool `json:"enabled"`
ServiceIDs []int64 `json:"service_ids"`
Services []ProbeGroup `json:"services,omitempty"`
CreatedAt int64 `json:"created_at"`
UpdatedAt int64 `json:"updated_at"`
}
type StatusPageInput struct {
Name string `json:"name"`
Slug string `json:"slug"`
Description string `json:"description"`
Enabled *bool `json:"enabled"`
ServiceIDs []int64 `json:"service_ids"`
}
var slugRx = regexp.MustCompile(`^[a-z0-9][a-z0-9-]{0,62}$`)
func aggregateStatus(ms []Monitor) string {
if len(ms) == 0 {
return "unknown"
}
hasMaint, hasPending, hasPaused := false, false, false
for _, m := range ms {
switch m.Status {
case "down":
return "down"
case "maintenance":
hasMaint = true
case "pending":
hasPending = true
case "paused":
hasPaused = true
}
}
if hasMaint {
return "maintenance"
}
if hasPending {
return "pending"
}
if hasPaused {
return "paused"
}
return "up"
}
func (s *Service) ListGroups(ctx context.Context) ([]ProbeGroup, error) {
rows, err := s.db.QueryContext(ctx, `SELECT id,name,description,created_at,updated_at FROM monitor_services ORDER BY name`)
if err != nil {
return nil, err
}
out := []ProbeGroup{}
for rows.Next() {
var g ProbeGroup
if err := rows.Scan(&g.ID, &g.Name, &g.Description, &g.CreatedAt, &g.UpdatedAt); err != nil {
_ = rows.Close()
return nil, err
}
out = append(out, g)
}
if err := rows.Err(); err != nil {
_ = rows.Close()
return nil, err
}
if err := rows.Close(); err != nil {
return nil, err
}
// Load monitors once and group in memory. This avoids an N+1 query pattern
// on the dashboard and keeps refresh cost predictable with many services.
all, err := s.List(ctx)
if err != nil {
return nil, err
}
byService := make(map[int64][]Monitor, len(out))
for _, m := range all {
if m.ServiceID != nil {
byService[*m.ServiceID] = append(byService[*m.ServiceID], m)
}
}
for i := range out {
out[i].Monitors = byService[out[i].ID]
if out[i].Monitors == nil {
out[i].Monitors = []Monitor{}
}
out[i].Status = aggregateStatus(out[i].Monitors)
}
return out, nil
}
func (s *Service) GetGroup(ctx context.Context, id int64) (ProbeGroup, error) {
var g ProbeGroup
if err := s.db.QueryRowContext(ctx, `SELECT id,name,description,created_at,updated_at FROM monitor_services WHERE id=?`, id).Scan(&g.ID, &g.Name, &g.Description, &g.CreatedAt, &g.UpdatedAt); err != nil {
return g, err
}
ms, err := s.listGroupMonitors(ctx, id)
if err != nil {
return g, err
}
g.Monitors = ms
g.Status = aggregateStatus(ms)
return g, nil
}
func (s *Service) listGroupMonitors(ctx context.Context, id int64) ([]Monitor, error) {
rows, err := s.db.QueryContext(ctx, selectMonitor+` WHERE m.service_id=? ORDER BY m.name`, time.Now().Add(-24*time.Hour).Unix(), id)
if err != nil {
return nil, err
}
defer rows.Close()
out := []Monitor{}
for rows.Next() {
m, e := scanMonitor(rows)
if e != nil {
return nil, e
}
out = append(out, m)
}
return out, rows.Err()
}
func normalizeGroupInput(in *ProbeGroupInput) error {
in.Name = strings.TrimSpace(in.Name)
in.Description = strings.TrimSpace(in.Description)
if in.Name == "" || len(in.Name) > 120 || strings.ContainsAny(in.Name, "\r\n") {
return errors.New("valid service name required")
}
if len(in.Description) > 2000 {
return errors.New("service description too long")
}
if in.MonitorIDs != nil {
seen := map[int64]bool{}
ids := make([]int64, 0, len(in.MonitorIDs))
for _, id := range in.MonitorIDs {
if id < 1 {
return errors.New("monitor_ids must contain positive IDs")
}
if !seen[id] {
seen[id] = true
ids = append(ids, id)
}
}
in.MonitorIDs = ids
}
return nil
}
func assignGroupMonitors(ctx context.Context, tx *sql.Tx, groupID int64, monitorIDs []int64, clearExisting bool) error {
if clearExisting {
if _, err := tx.ExecContext(ctx, `UPDATE monitors SET service_id=NULL,updated_at=? WHERE service_id=?`, time.Now().Unix(), groupID); err != nil {
return err
}
}
for _, monitorID := range monitorIDs {
r, err := tx.ExecContext(ctx, `UPDATE monitors SET service_id=?,updated_at=? WHERE id=?`, groupID, time.Now().Unix(), monitorID)
if err != nil {
return err
}
n, _ := r.RowsAffected()
if n == 0 {
return fmt.Errorf("monitor %d not found", monitorID)
}
}
return nil
}
func (s *Service) CreateGroup(ctx context.Context, in ProbeGroupInput) (ProbeGroup, error) {
if err := normalizeGroupInput(&in); err != nil {
return ProbeGroup{}, err
}
tx, err := s.db.BeginTx(ctx, nil)
if err != nil {
return ProbeGroup{}, err
}
defer tx.Rollback()
now := time.Now().Unix()
r, err := tx.ExecContext(ctx, `INSERT INTO monitor_services(name,description,created_at,updated_at) VALUES(?,?,?,?)`, in.Name, in.Description, now, now)
if err != nil {
return ProbeGroup{}, err
}
id, _ := r.LastInsertId()
if in.MonitorIDs != nil {
if err := assignGroupMonitors(ctx, tx, id, in.MonitorIDs, false); err != nil {
return ProbeGroup{}, err
}
}
if err := tx.Commit(); err != nil {
return ProbeGroup{}, err
}
return s.GetGroup(ctx, id)
}
func (s *Service) UpdateGroup(ctx context.Context, id int64, in ProbeGroupInput) (ProbeGroup, error) {
if err := normalizeGroupInput(&in); err != nil {
return ProbeGroup{}, err
}
tx, err := s.db.BeginTx(ctx, nil)
if err != nil {
return ProbeGroup{}, err
}
defer tx.Rollback()
r, err := tx.ExecContext(ctx, `UPDATE monitor_services SET name=?,description=?,updated_at=? WHERE id=?`, in.Name, in.Description, time.Now().Unix(), id)
if err != nil {
return ProbeGroup{}, err
}
n, _ := r.RowsAffected()
if n == 0 {
return ProbeGroup{}, sql.ErrNoRows
}
if in.MonitorIDs != nil {
if err := assignGroupMonitors(ctx, tx, id, in.MonitorIDs, true); err != nil {
return ProbeGroup{}, err
}
}
if err := tx.Commit(); err != nil {
return ProbeGroup{}, err
}
return s.GetGroup(ctx, id)
}
func (s *Service) DeleteGroup(ctx context.Context, id int64) error {
_, e := s.db.ExecContext(ctx, `DELETE FROM monitor_services WHERE id=?`, id)
return e
}
func normalizeStatusPage(in *StatusPageInput) error {
in.Name = strings.TrimSpace(in.Name)
in.Slug = strings.ToLower(strings.TrimSpace(in.Slug))
in.Description = strings.TrimSpace(in.Description)
if in.Name == "" || len(in.Name) > 120 || strings.ContainsAny(in.Name, "\r\n") || !slugRx.MatchString(in.Slug) {
return errors.New("name and valid slug required (lowercase letters, numbers, hyphens)")
}
if len(in.Description) > 4000 {
return errors.New("status page description too long")
}
seen := map[int64]bool{}
var ids []int64
for _, id := range in.ServiceIDs {
if id > 0 && !seen[id] {
seen[id] = true
ids = append(ids, id)
}
}
in.ServiceIDs = ids
return nil
}
func (s *Service) ListStatusPages(ctx context.Context) ([]StatusPage, error) {
rows, e := s.db.QueryContext(ctx, `SELECT id,name,slug,description,enabled,created_at,updated_at FROM status_pages ORDER BY name`)
if e != nil {
return nil, e
}
out := []StatusPage{}
for rows.Next() {
var p StatusPage
if e = rows.Scan(&p.ID, &p.Name, &p.Slug, &p.Description, &p.Enabled, &p.CreatedAt, &p.UpdatedAt); e != nil {
_ = rows.Close()
return nil, e
}
p.ServiceIDs = []int64{}
out = append(out, p)
}
if e = rows.Err(); e != nil {
_ = rows.Close()
return nil, e
}
if e = rows.Close(); e != nil {
return nil, e
}
// Resolve page memberships only after releasing the one SQLite connection.
for i := range out {
out[i].ServiceIDs, e = s.statusPageServiceIDs(ctx, out[i].ID)
if e != nil {
return nil, e
}
}
return out, nil
}
func (s *Service) GetStatusPage(ctx context.Context, id int64) (StatusPage, error) {
var p StatusPage
e := s.db.QueryRowContext(ctx, `SELECT id,name,slug,description,enabled,created_at,updated_at FROM status_pages WHERE id=?`, id).Scan(&p.ID, &p.Name, &p.Slug, &p.Description, &p.Enabled, &p.CreatedAt, &p.UpdatedAt)
if e != nil {
return p, e
}
p.ServiceIDs, e = s.statusPageServiceIDs(ctx, id)
return p, e
}
func (s *Service) statusPageServiceIDs(ctx context.Context, id int64) ([]int64, error) {
rows, e := s.db.QueryContext(ctx, `SELECT service_id FROM status_page_services WHERE page_id=? ORDER BY sort_order,service_id`, id)
if e != nil {
return nil, e
}
defer rows.Close()
out := []int64{}
for rows.Next() {
var x int64
if e = rows.Scan(&x); e != nil {
return nil, e
}
out = append(out, x)
}
return out, rows.Err()
}
func (s *Service) savePageServices(ctx context.Context, tx *sql.Tx, id int64, ids []int64) error {
if _, e := tx.ExecContext(ctx, `DELETE FROM status_page_services WHERE page_id=?`, id); e != nil {
return e
}
for i, sid := range ids {
if _, e := tx.ExecContext(ctx, `INSERT INTO status_page_services(page_id,service_id,sort_order) VALUES(?,?,?)`, id, sid, i); e != nil {
return e
}
}
return nil
}
func (s *Service) CreateStatusPage(ctx context.Context, in StatusPageInput) (StatusPage, error) {
if e := normalizeStatusPage(&in); e != nil {
return StatusPage{}, e
}
en := true
if in.Enabled != nil {
en = *in.Enabled
}
tx, e := s.db.BeginTx(ctx, nil)
if e != nil {
return StatusPage{}, e
}
defer tx.Rollback()
now := time.Now().Unix()
r, e := tx.ExecContext(ctx, `INSERT INTO status_pages(name,slug,description,enabled,created_at,updated_at) VALUES(?,?,?,?,?,?)`, in.Name, in.Slug, in.Description, en, now, now)
if e != nil {
return StatusPage{}, e
}
id, _ := r.LastInsertId()
if e = s.savePageServices(ctx, tx, id, in.ServiceIDs); e != nil {
return StatusPage{}, e
}
if e = tx.Commit(); e != nil {
return StatusPage{}, e
}
return s.GetStatusPage(ctx, id)
}
func (s *Service) UpdateStatusPage(ctx context.Context, id int64, in StatusPageInput) (StatusPage, error) {
if e := normalizeStatusPage(&in); e != nil {
return StatusPage{}, e
}
old, e := s.GetStatusPage(ctx, id)
if e != nil {
return StatusPage{}, e
}
en := old.Enabled
if in.Enabled != nil {
en = *in.Enabled
}
tx, e := s.db.BeginTx(ctx, nil)
if e != nil {
return StatusPage{}, e
}
defer tx.Rollback()
if _, e = tx.ExecContext(ctx, `UPDATE status_pages SET name=?,slug=?,description=?,enabled=?,updated_at=? WHERE id=?`, in.Name, in.Slug, in.Description, en, time.Now().Unix(), id); e != nil {
return StatusPage{}, e
}
if e = s.savePageServices(ctx, tx, id, in.ServiceIDs); e != nil {
return StatusPage{}, e
}
if e = tx.Commit(); e != nil {
return StatusPage{}, e
}
return s.GetStatusPage(ctx, id)
}
func (s *Service) DeleteStatusPage(ctx context.Context, id int64) error {
_, e := s.db.ExecContext(ctx, `DELETE FROM status_pages WHERE id=?`, id)
return e
}
func (s *Service) PublicStatusPage(ctx context.Context, slug string) (StatusPage, error) {
p := StatusPage{Services: []ProbeGroup{}, ServiceIDs: []int64{}}
e := s.db.QueryRowContext(ctx, `SELECT id,name,slug,description,enabled,created_at,updated_at FROM status_pages WHERE slug=? AND enabled=1`, slug).Scan(&p.ID, &p.Name, &p.Slug, &p.Description, &p.Enabled, &p.CreatedAt, &p.UpdatedAt)
if e != nil {
return p, e
}
rows, e := s.db.QueryContext(ctx, `SELECT ms.id,ms.name,ms.description,ms.created_at,ms.updated_at FROM monitor_services ms JOIN status_page_services ps ON ps.service_id=ms.id WHERE ps.page_id=? ORDER BY ps.sort_order,ms.name`, p.ID)
if e != nil {
return p, e
}
for rows.Next() {
var g ProbeGroup
if e = rows.Scan(&g.ID, &g.Name, &g.Description, &g.CreatedAt, &g.UpdatedAt); e != nil {
_ = rows.Close()
return p, e
}
g.Monitors = []Monitor{}
p.Services = append(p.Services, g)
p.ServiceIDs = append(p.ServiceIDs, g.ID)
}
if e = rows.Err(); e != nil {
_ = rows.Close()
return p, e
}
if e = rows.Close(); e != nil {
return p, e
}
all, e := s.List(ctx)
if e != nil {
return p, e
}
byService := map[int64][]Monitor{}
for _, m := range all {
if m.ServiceID != nil {
byService[*m.ServiceID] = append(byService[*m.ServiceID], m)
}
}
for gi := range p.Services {
p.Services[gi].Monitors = byService[p.Services[gi].ID]
if p.Services[gi].Monitors == nil {
p.Services[gi].Monitors = []Monitor{}
}
p.Services[gi].Status = aggregateStatus(p.Services[gi].Monitors)
}
p.Status = aggregateServiceStatuses(p.Services)
return p, nil
}
func aggregateServiceStatuses(groups []ProbeGroup) string {
if len(groups) == 0 {
return "unknown"
}
hasMaint, hasPending, hasPaused := false, false, false
for _, g := range groups {
switch g.Status {
case "down":
return "down"
case "maintenance":
hasMaint = true
case "pending", "unknown":
hasPending = true
case "paused":
hasPaused = true
}
}
if hasMaint {
return "maintenance"
}
if hasPending {
return "pending"
}
if hasPaused {
return "paused"
}
return "up"
}