Compare commits
3
Commits
| Author | SHA1 | Date | |
|---|---|---|---|
 sysopsandClaude Sonnet 5
|
6a03dcafd6 | ||
|
sysopsandClaude Sonnet 5
|
b23cd1961f | ||
|
sysopsandClaude Sonnet 5
|
4a30345e07 |
@@ -1,88 +0,0 @@
|
||||
package cfgservice
|
||||
|
||||
import (
|
||||
"context"
|
||||
"errors"
|
||||
"sync"
|
||||
"time"
|
||||
)
|
||||
|
||||
// DefaultCacheTTL ist die dokumentierte Cache-Invalidierungszeit
|
||||
// (Akzeptanzkriterium 2 / Pruefung 2 in diesem Ticket bezieht sich auf die
|
||||
// Aenderungsnachvollziehbarkeit — die Cache-Frist selbst folgt demselben
|
||||
// Muster wie internal/flag.DefaultCacheTTL).
|
||||
const DefaultCacheTTL = 5 * time.Second
|
||||
|
||||
type cacheEntry struct {
|
||||
value Value
|
||||
expiresAt time.Time
|
||||
}
|
||||
|
||||
// Service ist die Leseseite mit Vorrangregel (Akzeptanzkriterium 1:
|
||||
// Tenant-Override vor Global-Default) und lokalem TTL-Cache.
|
||||
type Service struct {
|
||||
store *Store
|
||||
ttl time.Duration
|
||||
|
||||
mu sync.RWMutex
|
||||
cache map[string]cacheEntry // Schluessel: key + "\x00" + tenantSlug
|
||||
}
|
||||
|
||||
func NewService(store *Store, ttl time.Duration) *Service {
|
||||
if ttl <= 0 {
|
||||
ttl = DefaultCacheTTL
|
||||
}
|
||||
return &Service{store: store, ttl: ttl, cache: make(map[string]cacheEntry)}
|
||||
}
|
||||
|
||||
func cacheKey(key, tenantSlug string) string {
|
||||
return key + "\x00" + tenantSlug
|
||||
}
|
||||
|
||||
// Resolve liefert den Konfigurationswert fuer einen Tenant: ein
|
||||
// Tenant-spezifischer Override hat Vorrang vor dem globalen Default
|
||||
// (Akzeptanzkriterium 1 / Pruefung 1). tenantSlug == "" wertet nur den
|
||||
// globalen Wert aus.
|
||||
func (s *Service) Resolve(ctx context.Context, tenantSlug, key string) (Value, error) {
|
||||
ck := cacheKey(key, tenantSlug)
|
||||
|
||||
s.mu.RLock()
|
||||
entry, exists := s.cache[ck]
|
||||
fresh := exists && time.Now().Before(entry.expiresAt)
|
||||
s.mu.RUnlock()
|
||||
if fresh {
|
||||
return entry.value, nil
|
||||
}
|
||||
|
||||
v, err := s.resolveUncached(ctx, tenantSlug, key)
|
||||
if err != nil {
|
||||
return Value{}, err
|
||||
}
|
||||
|
||||
s.mu.Lock()
|
||||
s.cache[ck] = cacheEntry{value: v, expiresAt: time.Now().Add(s.ttl)}
|
||||
s.mu.Unlock()
|
||||
return v, nil
|
||||
}
|
||||
|
||||
func (s *Service) resolveUncached(ctx context.Context, tenantSlug, key string) (Value, error) {
|
||||
if tenantSlug != "" {
|
||||
v, err := s.store.Get(ctx, key, tenantSlug)
|
||||
if err == nil {
|
||||
return v, nil
|
||||
}
|
||||
if !errors.Is(err, ErrNotFound) {
|
||||
return Value{}, err
|
||||
}
|
||||
}
|
||||
return s.store.Get(ctx, key, GlobalScope)
|
||||
}
|
||||
|
||||
// Invalidate erzwingt beim naechsten Resolve-Aufruf ein sofortiges Neuladen
|
||||
// fuer einen bestimmten (key, tenantSlug) statt auf den TTL-Ablauf zu warten
|
||||
// — analog internal/flag.Service.Invalidate.
|
||||
func (s *Service) Invalidate(key, tenantSlug string) {
|
||||
s.mu.Lock()
|
||||
delete(s.cache, cacheKey(key, tenantSlug))
|
||||
s.mu.Unlock()
|
||||
}
|
||||
@@ -1,114 +0,0 @@
|
||||
package cfgservice
|
||||
|
||||
import (
|
||||
"context"
|
||||
"testing"
|
||||
"time"
|
||||
)
|
||||
|
||||
// Akzeptanzkriterium 1 + Pruefung 1: Tenant-Override hat Vorrang vor
|
||||
// Global-Default, automatisiert getestet.
|
||||
func TestService_TenantOverrideTakesPrecedenceOverGlobal(t *testing.T) {
|
||||
store, cleanup := setupStoreTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
if _, err := store.Set(ctx, "test_precedence_key", GlobalScope, "global-wert"); err != nil {
|
||||
t.Fatalf("set global: %v", err)
|
||||
}
|
||||
if _, err := store.Set(ctx, "test_precedence_key", "test_acme", "tenant-wert"); err != nil {
|
||||
t.Fatalf("set tenant: %v", err)
|
||||
}
|
||||
|
||||
svc := NewService(store, time.Hour)
|
||||
|
||||
got, err := svc.Resolve(ctx, "test_acme", "test_precedence_key")
|
||||
if err != nil {
|
||||
t.Fatalf("resolve mit override: %v", err)
|
||||
}
|
||||
if got.Value != "tenant-wert" {
|
||||
t.Fatalf("erwartet tenant-override, habe %q", got.Value)
|
||||
}
|
||||
|
||||
gotOther, err := svc.Resolve(ctx, "test_anderer_tenant", "test_precedence_key")
|
||||
if err != nil {
|
||||
t.Fatalf("resolve ohne override: %v", err)
|
||||
}
|
||||
if gotOther.Value != "global-wert" {
|
||||
t.Fatalf("erwartet global-default fuer tenant ohne override, habe %q", gotOther.Value)
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 2 + Pruefung 2: Cache-Invalidierung nach
|
||||
// Konfigurationsaenderung innerhalb dokumentierter Zeit gemessen.
|
||||
func TestService_CacheInvalidationTiming(t *testing.T) {
|
||||
store, cleanup := setupStoreTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
const ttl = 150 * time.Millisecond
|
||||
if _, err := store.Set(ctx, "test_ttl_key", GlobalScope, "alt"); err != nil {
|
||||
t.Fatalf("set: %v", err)
|
||||
}
|
||||
svc := NewService(store, ttl)
|
||||
|
||||
v, err := svc.Resolve(ctx, "", "test_ttl_key")
|
||||
if err != nil {
|
||||
t.Fatalf("resolve: %v", err)
|
||||
}
|
||||
if v.Value != "alt" {
|
||||
t.Fatalf("erwartet 'alt', habe %q", v.Value)
|
||||
}
|
||||
|
||||
changedAt := time.Now()
|
||||
if _, err := store.Set(ctx, "test_ttl_key", GlobalScope, "neu"); err != nil {
|
||||
t.Fatalf("set: %v", err)
|
||||
}
|
||||
|
||||
v, err = svc.Resolve(ctx, "", "test_ttl_key")
|
||||
if err != nil {
|
||||
t.Fatalf("resolve direkt nach aenderung: %v", err)
|
||||
}
|
||||
if v.Value != "alt" {
|
||||
t.Fatalf("cache haette den alten wert liefern sollen, habe %q", v.Value)
|
||||
}
|
||||
|
||||
deadline := changedAt.Add(ttl + 100*time.Millisecond)
|
||||
for time.Now().Before(deadline) {
|
||||
v, err := svc.Resolve(ctx, "", "test_ttl_key")
|
||||
if err != nil {
|
||||
t.Fatalf("resolve: %v", err)
|
||||
}
|
||||
if v.Value == "neu" {
|
||||
t.Logf("aenderung wurde nach %s wirksam (ziel: innerhalb %s + toleranz)", time.Since(changedAt), ttl)
|
||||
return
|
||||
}
|
||||
time.Sleep(10 * time.Millisecond)
|
||||
}
|
||||
t.Fatalf("aenderung wurde nicht innerhalb von %s wirksam", deadline.Sub(changedAt))
|
||||
}
|
||||
|
||||
func TestService_InvalidateForcesImmediateRefresh(t *testing.T) {
|
||||
store, cleanup := setupStoreTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
if _, err := store.Set(ctx, "test_invalidate_key", GlobalScope, "alt"); err != nil {
|
||||
t.Fatalf("set: %v", err)
|
||||
}
|
||||
svc := NewService(store, time.Hour)
|
||||
_, _ = svc.Resolve(ctx, "", "test_invalidate_key")
|
||||
|
||||
if _, err := store.Set(ctx, "test_invalidate_key", GlobalScope, "neu"); err != nil {
|
||||
t.Fatalf("set: %v", err)
|
||||
}
|
||||
svc.Invalidate("test_invalidate_key", "")
|
||||
|
||||
v, err := svc.Resolve(ctx, "", "test_invalidate_key")
|
||||
if err != nil {
|
||||
t.Fatalf("resolve: %v", err)
|
||||
}
|
||||
if v.Value != "neu" {
|
||||
t.Fatalf("erwartet sofort sichtbaren neuen wert nach Invalidate, habe %q", v.Value)
|
||||
}
|
||||
}
|
||||
@@ -1,129 +0,0 @@
|
||||
// Package cfgservice implementiert Core CFG-01: den zentralen Dienst fuer
|
||||
// globale und tenant-spezifische Konfigurationswerte mit Versionierung und
|
||||
// Cache-Invalidierung. Andere Module lesen Konfiguration AUSSCHLIESSLICH
|
||||
// ueber dieses Paket (Akzeptanzkriterium 3), niemals ueber eigene Tabellen.
|
||||
package cfgservice
|
||||
|
||||
import (
|
||||
"context"
|
||||
"errors"
|
||||
"fmt"
|
||||
|
||||
"github.com/jackc/pgx/v5"
|
||||
"github.com/jackc/pgx/v5/pgxpool"
|
||||
)
|
||||
|
||||
// GlobalScope ist der reservierte Scope-Wert fuer globale Defaults — jeder
|
||||
// andere Scope-Wert ist ein Tenant-Slug (Akzeptanzkriterium 1).
|
||||
const GlobalScope = "global"
|
||||
|
||||
var ErrNotFound = errors.New("cfgservice: kein wert fuer diesen key gefunden")
|
||||
|
||||
type Value struct {
|
||||
Key string
|
||||
Scope string
|
||||
Value string
|
||||
Version int
|
||||
}
|
||||
|
||||
type HistoryEntry struct {
|
||||
Key string
|
||||
Scope string
|
||||
Value string
|
||||
Version int
|
||||
}
|
||||
|
||||
// Store ist die Schreib-/Verwaltungsseite. Set schreibt IMMER sowohl den
|
||||
// aktuellen Stand (config_values) als auch einen Historieneintrag
|
||||
// (config_value_history) in derselben Transaktion — eine Aenderung ohne
|
||||
// Versionshistorie ist strukturell ausgeschlossen (Akzeptanzkriterium 2).
|
||||
type Store struct {
|
||||
pool *pgxpool.Pool
|
||||
}
|
||||
|
||||
func NewStore(pool *pgxpool.Pool) *Store {
|
||||
return &Store{pool: pool}
|
||||
}
|
||||
|
||||
// Set schreibt einen neuen Wert fuer (key, scope) und erhoeht die Version um 1
|
||||
// (Version 1 bei erstmaligem Setzen).
|
||||
func (s *Store) Set(ctx context.Context, key, scope, value string) (Value, error) {
|
||||
if scope == "" {
|
||||
return Value{}, errors.New("cfgservice: scope darf nicht leer sein")
|
||||
}
|
||||
|
||||
tx, err := s.pool.Begin(ctx)
|
||||
if err != nil {
|
||||
return Value{}, fmt.Errorf("transaktion starten: %w", err)
|
||||
}
|
||||
defer func() { _ = tx.Rollback(ctx) }()
|
||||
|
||||
var currentVersion int
|
||||
err = tx.QueryRow(ctx, `SELECT version FROM config_values WHERE key = $1 AND scope = $2`, key, scope).Scan(¤tVersion)
|
||||
if err != nil && !errors.Is(err, pgx.ErrNoRows) {
|
||||
return Value{}, fmt.Errorf("aktuelle version lesen: %w", err)
|
||||
}
|
||||
newVersion := currentVersion + 1
|
||||
|
||||
if _, err := tx.Exec(ctx, `
|
||||
INSERT INTO config_values (key, scope, value, version, updated_at)
|
||||
VALUES ($1, $2, $3, $4, now())
|
||||
ON CONFLICT (key, scope) DO UPDATE SET value = $3, version = $4, updated_at = now()
|
||||
`, key, scope, value, newVersion); err != nil {
|
||||
return Value{}, fmt.Errorf("wert speichern: %w", err)
|
||||
}
|
||||
|
||||
if _, err := tx.Exec(ctx, `
|
||||
INSERT INTO config_value_history (key, scope, value, version, changed_at)
|
||||
VALUES ($1, $2, $3, $4, now())
|
||||
`, key, scope, value, newVersion); err != nil {
|
||||
return Value{}, fmt.Errorf("historie schreiben: %w", err)
|
||||
}
|
||||
|
||||
if err := tx.Commit(ctx); err != nil {
|
||||
return Value{}, fmt.Errorf("transaktion committen: %w", err)
|
||||
}
|
||||
|
||||
return Value{Key: key, Scope: scope, Value: value, Version: newVersion}, nil
|
||||
}
|
||||
|
||||
// Get liefert den Wert fuer GENAU EINEN Scope (kein Vorrang-Fallback) — die
|
||||
// Vorrangregel (Tenant vor Global) lebt bewusst in Service.Resolve, damit
|
||||
// Store rein CRUD bleibt.
|
||||
func (s *Store) Get(ctx context.Context, key, scope string) (Value, error) {
|
||||
var v Value
|
||||
v.Key, v.Scope = key, scope
|
||||
err := s.pool.QueryRow(ctx, `
|
||||
SELECT value, version FROM config_values WHERE key = $1 AND scope = $2
|
||||
`, key, scope).Scan(&v.Value, &v.Version)
|
||||
if err != nil {
|
||||
if errors.Is(err, pgx.ErrNoRows) {
|
||||
return Value{}, ErrNotFound
|
||||
}
|
||||
return Value{}, fmt.Errorf("wert lesen: %w", err)
|
||||
}
|
||||
return v, nil
|
||||
}
|
||||
|
||||
// History liefert die vollstaendige Versionshistorie eines (key, scope) in
|
||||
// aufsteigender Reihenfolge (Akzeptanzkriterium 2 / Pruefung 3).
|
||||
func (s *Store) History(ctx context.Context, key, scope string) ([]HistoryEntry, error) {
|
||||
rows, err := s.pool.Query(ctx, `
|
||||
SELECT key, scope, value, version FROM config_value_history
|
||||
WHERE key = $1 AND scope = $2 ORDER BY version
|
||||
`, key, scope)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("historie abfragen: %w", err)
|
||||
}
|
||||
defer rows.Close()
|
||||
|
||||
var out []HistoryEntry
|
||||
for rows.Next() {
|
||||
var h HistoryEntry
|
||||
if err := rows.Scan(&h.Key, &h.Scope, &h.Value, &h.Version); err != nil {
|
||||
return nil, fmt.Errorf("historieneintrag lesen: %w", err)
|
||||
}
|
||||
out = append(out, h)
|
||||
}
|
||||
return out, rows.Err()
|
||||
}
|
||||
@@ -1,119 +0,0 @@
|
||||
package cfgservice
|
||||
|
||||
import (
|
||||
"context"
|
||||
"errors"
|
||||
"os"
|
||||
"testing"
|
||||
|
||||
"github.com/jackc/pgx/v5/pgxpool"
|
||||
)
|
||||
|
||||
func setupStoreTest(t *testing.T) (*Store, func()) {
|
||||
t.Helper()
|
||||
adminDSN := os.Getenv("TEST_ADMIN_DSN")
|
||||
if adminDSN == "" {
|
||||
t.Skip("TEST_ADMIN_DSN nicht gesetzt, Integrationstest uebersprungen")
|
||||
}
|
||||
ctx := context.Background()
|
||||
|
||||
pool, err := pgxpool.New(ctx, adminDSN)
|
||||
if err != nil {
|
||||
t.Fatalf("pool: %v", err)
|
||||
}
|
||||
if _, err := pool.Exec(ctx, `
|
||||
CREATE TABLE IF NOT EXISTS config_values (
|
||||
key TEXT NOT NULL,
|
||||
scope TEXT NOT NULL CHECK (scope <> ''),
|
||||
value TEXT NOT NULL,
|
||||
version INT NOT NULL,
|
||||
updated_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
||||
PRIMARY KEY (key, scope)
|
||||
);
|
||||
CREATE TABLE IF NOT EXISTS config_value_history (
|
||||
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
|
||||
key TEXT NOT NULL,
|
||||
scope TEXT NOT NULL,
|
||||
value TEXT NOT NULL,
|
||||
version INT NOT NULL,
|
||||
changed_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
);
|
||||
`); err != nil {
|
||||
t.Fatalf("schema: %v", err)
|
||||
}
|
||||
|
||||
cleanup := func() {
|
||||
_, _ = pool.Exec(ctx, `DELETE FROM config_value_history WHERE key LIKE 'test\_%' ESCAPE '\'`)
|
||||
_, _ = pool.Exec(ctx, `DELETE FROM config_values WHERE key LIKE 'test\_%' ESCAPE '\'`)
|
||||
pool.Close()
|
||||
}
|
||||
return NewStore(pool), cleanup
|
||||
}
|
||||
|
||||
func TestStore_SetIncrementsVersion(t *testing.T) {
|
||||
store, cleanup := setupStoreTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
v1, err := store.Set(ctx, "test_key", GlobalScope, "erster-wert")
|
||||
if err != nil {
|
||||
t.Fatalf("set 1: %v", err)
|
||||
}
|
||||
if v1.Version != 1 {
|
||||
t.Fatalf("erwartet version 1, habe %d", v1.Version)
|
||||
}
|
||||
|
||||
v2, err := store.Set(ctx, "test_key", GlobalScope, "zweiter-wert")
|
||||
if err != nil {
|
||||
t.Fatalf("set 2: %v", err)
|
||||
}
|
||||
if v2.Version != 2 {
|
||||
t.Fatalf("erwartet version 2, habe %d", v2.Version)
|
||||
}
|
||||
|
||||
got, err := store.Get(ctx, "test_key", GlobalScope)
|
||||
if err != nil {
|
||||
t.Fatalf("get: %v", err)
|
||||
}
|
||||
if got.Value != "zweiter-wert" || got.Version != 2 {
|
||||
t.Fatalf("aktueller wert unerwartet: %+v", got)
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 2 + Pruefung 3: Versionierungshistorie ueber mehrere
|
||||
// Aenderungen hinweg nachvollzogen.
|
||||
func TestStore_HistoryTracksAllChanges(t *testing.T) {
|
||||
store, cleanup := setupStoreTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
values := []string{"v1", "v2", "v3"}
|
||||
for _, v := range values {
|
||||
if _, err := store.Set(ctx, "test_history_key", GlobalScope, v); err != nil {
|
||||
t.Fatalf("set %q: %v", v, err)
|
||||
}
|
||||
}
|
||||
|
||||
history, err := store.History(ctx, "test_history_key", GlobalScope)
|
||||
if err != nil {
|
||||
t.Fatalf("history: %v", err)
|
||||
}
|
||||
if len(history) != 3 {
|
||||
t.Fatalf("erwartet 3 historieneintraege, habe %d", len(history))
|
||||
}
|
||||
for i, h := range history {
|
||||
if h.Version != i+1 || h.Value != values[i] {
|
||||
t.Fatalf("historieneintrag[%d] unerwartet: %+v", i, h)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestStore_GetUnknownKeyReturnsNotFound(t *testing.T) {
|
||||
store, cleanup := setupStoreTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
if _, err := store.Get(ctx, "test_nie_gesetzt", GlobalScope); !errors.Is(err, ErrNotFound) {
|
||||
t.Fatalf("erwartet ErrNotFound, habe %v", err)
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,87 @@
|
||||
// Package flag implementiert Core LIC-02: einen Feature-Flag-Dienst mit
|
||||
// Strategien (global an/aus, Prozentsatz, Tenant-Zielgruppe) als Kernfunktion
|
||||
// des Core-Dienstes selbst — keine zusaetzliche Infrastruktur (Unleash-Server
|
||||
// + eigene DB), siehe "bewusst vermeiden" im LIC-02-Ticket.
|
||||
package flag
|
||||
|
||||
import (
|
||||
"context"
|
||||
"errors"
|
||||
"fmt"
|
||||
"hash/fnv"
|
||||
|
||||
"github.com/jackc/pgx/v5"
|
||||
"github.com/jackc/pgx/v5/pgxpool"
|
||||
)
|
||||
|
||||
var ErrNotFound = errors.New("flag: nicht gefunden")
|
||||
|
||||
// Flag ist die zentrale Definition — Auswertung (Evaluate) ist bewusst davon
|
||||
// getrennt (Unleash-Prinzip: Flag-Verwaltung vs. Flag-Auswertung).
|
||||
type Flag struct {
|
||||
Key string
|
||||
Enabled bool
|
||||
RolloutPercentage int
|
||||
TargetTenantSlugs []string
|
||||
}
|
||||
|
||||
// Store ist die Verwaltungsseite (Admin): Flags definieren/lesen.
|
||||
type Store struct {
|
||||
pool *pgxpool.Pool
|
||||
}
|
||||
|
||||
func NewStore(pool *pgxpool.Pool) *Store {
|
||||
return &Store{pool: pool}
|
||||
}
|
||||
|
||||
func (s *Store) Set(ctx context.Context, f Flag) error {
|
||||
if f.TargetTenantSlugs == nil {
|
||||
f.TargetTenantSlugs = []string{} // pgx uebertraegt ein nil-Slice sonst als SQL NULL statt leerem Array.
|
||||
}
|
||||
_, err := s.pool.Exec(ctx, `
|
||||
INSERT INTO feature_flags (key, enabled, rollout_percentage, target_tenant_slugs, updated_at)
|
||||
VALUES ($1, $2, $3, $4, now())
|
||||
ON CONFLICT (key) DO UPDATE SET
|
||||
enabled = $2, rollout_percentage = $3, target_tenant_slugs = $4, updated_at = now()
|
||||
`, f.Key, f.Enabled, f.RolloutPercentage, f.TargetTenantSlugs)
|
||||
if err != nil {
|
||||
return fmt.Errorf("flag speichern: %w", err)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
func (s *Store) Get(ctx context.Context, key string) (Flag, error) {
|
||||
var f Flag
|
||||
row := s.pool.QueryRow(ctx, `
|
||||
SELECT key, enabled, rollout_percentage, target_tenant_slugs
|
||||
FROM feature_flags WHERE key = $1
|
||||
`, key)
|
||||
if err := row.Scan(&f.Key, &f.Enabled, &f.RolloutPercentage, &f.TargetTenantSlugs); err != nil {
|
||||
if errors.Is(err, pgx.ErrNoRows) {
|
||||
return Flag{}, ErrNotFound
|
||||
}
|
||||
return Flag{}, fmt.Errorf("flag lesen: %w", err)
|
||||
}
|
||||
return f, nil
|
||||
}
|
||||
|
||||
// evaluate wendet die Strategien in fester Reihenfolge an: globaler
|
||||
// An/Aus-Schalter zuerst, dann Tenant-Zielgruppe, dann Prozentsatz-Rollout.
|
||||
// Ein unbekannter/nicht getroffener Fall ergibt false — Fail-Safe-Default,
|
||||
// kein Feature wird versehentlich aktiv.
|
||||
func evaluate(f Flag, tenantSlug string) bool {
|
||||
if f.Enabled {
|
||||
return true
|
||||
}
|
||||
for _, target := range f.TargetTenantSlugs {
|
||||
if target == tenantSlug {
|
||||
return true
|
||||
}
|
||||
}
|
||||
if f.RolloutPercentage > 0 {
|
||||
h := fnv.New32a()
|
||||
_, _ = h.Write([]byte(f.Key + "|" + tenantSlug))
|
||||
return int(h.Sum32()%100) < f.RolloutPercentage
|
||||
}
|
||||
return false
|
||||
}
|
||||
@@ -0,0 +1,43 @@
|
||||
package flag
|
||||
|
||||
import "testing"
|
||||
|
||||
func TestEvaluate_GlobalEnabled(t *testing.T) {
|
||||
f := Flag{Key: "k", Enabled: true}
|
||||
if !evaluate(f, "irgendein-tenant") {
|
||||
t.Fatal("global aktiviertes flag sollte fuer jeden tenant true liefern")
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 1 + Pruefung 2: Zielgruppen-Strategie.
|
||||
func TestEvaluate_TargetTenantStrategy(t *testing.T) {
|
||||
f := Flag{Key: "k", Enabled: false, TargetTenantSlugs: []string{"acme"}}
|
||||
if !evaluate(f, "acme") {
|
||||
t.Fatal("erwartet true fuer tenant in zielgruppe")
|
||||
}
|
||||
if evaluate(f, "globex") {
|
||||
t.Fatal("erwartet false fuer tenant ausserhalb der zielgruppe")
|
||||
}
|
||||
}
|
||||
|
||||
func TestEvaluate_RolloutPercentageBoundaries(t *testing.T) {
|
||||
full := Flag{Key: "k", RolloutPercentage: 100}
|
||||
if !evaluate(full, "beliebiger-tenant-1") || !evaluate(full, "beliebiger-tenant-2") {
|
||||
t.Fatal("100% rollout sollte immer true liefern")
|
||||
}
|
||||
|
||||
none := Flag{Key: "k", RolloutPercentage: 0}
|
||||
if evaluate(none, "beliebiger-tenant") {
|
||||
t.Fatal("0% rollout ohne enabled/zielgruppe sollte false liefern")
|
||||
}
|
||||
}
|
||||
|
||||
func TestEvaluate_RolloutIsDeterministicPerTenant(t *testing.T) {
|
||||
f := Flag{Key: "k", RolloutPercentage: 50}
|
||||
first := evaluate(f, "stabiler-tenant")
|
||||
for i := 0; i < 5; i++ {
|
||||
if evaluate(f, "stabiler-tenant") != first {
|
||||
t.Fatal("rollout-auswertung sollte fuer denselben tenant/key stabil sein")
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,87 @@
|
||||
package flag
|
||||
|
||||
import (
|
||||
"context"
|
||||
"log/slog"
|
||||
"sync"
|
||||
"time"
|
||||
)
|
||||
|
||||
// DefaultCacheTTL ist die dokumentierte Cache-Invalidierungszeit
|
||||
// (Akzeptanzkriterium 2/3): eine Aenderung wirkt spaetestens nach dieser
|
||||
// Zeit auf allen Core-Instanzen, ohne dass ein Dienst neu gestartet werden
|
||||
// muss (Akzeptanzkriterium 3).
|
||||
const DefaultCacheTTL = 5 * time.Second
|
||||
|
||||
type cacheEntry struct {
|
||||
flag Flag
|
||||
expiresAt time.Time
|
||||
}
|
||||
|
||||
// Service ist die Auswertungsseite (SDK/Client-Analogon zu Unleash) mit
|
||||
// lokalem TTL-Cache. Bewusst getrennt von Store (Verwaltung).
|
||||
type Service struct {
|
||||
store *Store
|
||||
ttl time.Duration
|
||||
|
||||
mu sync.RWMutex
|
||||
cache map[string]cacheEntry
|
||||
}
|
||||
|
||||
func NewService(store *Store, ttl time.Duration) *Service {
|
||||
if ttl <= 0 {
|
||||
ttl = DefaultCacheTTL
|
||||
}
|
||||
return &Service{store: store, ttl: ttl, cache: make(map[string]cacheEntry)}
|
||||
}
|
||||
|
||||
// IsEnabled wertet ein Flag fuer einen Tenant aus. Liefert IMMER einen
|
||||
// bool ohne Fehlerwert — ein nicht erreichbarer Flag-Dienst darf abhaengige
|
||||
// Aufrufer nicht zum Absturz bringen oder zu Fehlerbehandlungscode zwingen,
|
||||
// der leicht vergessen wird (Akzeptanzkriterium 3 / Pruefung 3: dokumentiertes
|
||||
// Fallback-Verhalten = false, ggf. aus dem zuletzt bekannten Zwischenspeicher).
|
||||
func (s *Service) IsEnabled(ctx context.Context, tenantSlug, key string) bool {
|
||||
f, ok := s.resolve(ctx, key)
|
||||
if !ok {
|
||||
return false
|
||||
}
|
||||
return evaluate(f, tenantSlug)
|
||||
}
|
||||
|
||||
func (s *Service) resolve(ctx context.Context, key string) (Flag, bool) {
|
||||
s.mu.RLock()
|
||||
entry, exists := s.cache[key]
|
||||
fresh := exists && time.Now().Before(entry.expiresAt)
|
||||
s.mu.RUnlock()
|
||||
if fresh {
|
||||
return entry.flag, true
|
||||
}
|
||||
|
||||
f, err := s.store.Get(ctx, key)
|
||||
if err != nil {
|
||||
if exists {
|
||||
slog.Warn("feature-flag-dienst nicht erreichbar, nutze zwischengespeicherten stand",
|
||||
"flag_key", key, "error", err)
|
||||
return entry.flag, true
|
||||
}
|
||||
slog.Warn("feature-flag-dienst nicht erreichbar, kein zwischengespeicherter stand vorhanden, fallback: deaktiviert",
|
||||
"flag_key", key, "error", err)
|
||||
return Flag{}, false
|
||||
}
|
||||
|
||||
s.mu.Lock()
|
||||
s.cache[key] = cacheEntry{flag: f, expiresAt: time.Now().Add(s.ttl)}
|
||||
s.mu.Unlock()
|
||||
return f, true
|
||||
}
|
||||
|
||||
// Invalidate erzwingt beim naechsten IsEnabled-Aufruf ein sofortiges Neuladen
|
||||
// aus der Datenbank statt auf den TTL-Ablauf zu warten — wird nach Store.Set
|
||||
// auf derselben Instanz aufgerufen, damit der Schreiber die eigene Aenderung
|
||||
// ohne Wartezeit sieht. Andere Core-Instanzen sehen sie spaetestens nach
|
||||
// DefaultCacheTTL (siehe Akzeptanzkriterium 3).
|
||||
func (s *Service) Invalidate(key string) {
|
||||
s.mu.Lock()
|
||||
delete(s.cache, key)
|
||||
s.mu.Unlock()
|
||||
}
|
||||
@@ -0,0 +1,179 @@
|
||||
package flag
|
||||
|
||||
import (
|
||||
"context"
|
||||
"os"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
"github.com/jackc/pgx/v5/pgxpool"
|
||||
)
|
||||
|
||||
func setupFlagStoreTest(t *testing.T) (*Store, func()) {
|
||||
t.Helper()
|
||||
adminDSN := os.Getenv("TEST_ADMIN_DSN")
|
||||
if adminDSN == "" {
|
||||
t.Skip("TEST_ADMIN_DSN nicht gesetzt, Integrationstest uebersprungen")
|
||||
}
|
||||
ctx := context.Background()
|
||||
|
||||
pool, err := pgxpool.New(ctx, adminDSN)
|
||||
if err != nil {
|
||||
t.Fatalf("pool: %v", err)
|
||||
}
|
||||
if _, err := pool.Exec(ctx, `
|
||||
CREATE TABLE IF NOT EXISTS feature_flags (
|
||||
key TEXT PRIMARY KEY,
|
||||
enabled BOOLEAN NOT NULL DEFAULT false,
|
||||
rollout_percentage INT NOT NULL DEFAULT 0,
|
||||
target_tenant_slugs TEXT[] NOT NULL DEFAULT '{}',
|
||||
updated_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
)`); err != nil {
|
||||
t.Fatalf("schema: %v", err)
|
||||
}
|
||||
|
||||
cleanup := func() {
|
||||
_, _ = pool.Exec(ctx, `DELETE FROM feature_flags WHERE key LIKE 'test\_%' ESCAPE '\'`)
|
||||
pool.Close()
|
||||
}
|
||||
return NewStore(pool), cleanup
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 1 + Pruefung 2: Zielgruppen-Strategie liefert im Test
|
||||
// die erwartete Auswertung.
|
||||
func TestService_TargetTenantStrategy(t *testing.T) {
|
||||
store, cleanup := setupFlagStoreTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
if err := store.Set(ctx, Flag{Key: "test_target_flag", TargetTenantSlugs: []string{"acme"}}); err != nil {
|
||||
t.Fatalf("set: %v", err)
|
||||
}
|
||||
svc := NewService(store, time.Hour)
|
||||
|
||||
if !svc.IsEnabled(ctx, "acme", "test_target_flag") {
|
||||
t.Fatal("erwartet true fuer tenant in zielgruppe")
|
||||
}
|
||||
if svc.IsEnabled(ctx, "globex", "test_target_flag") {
|
||||
t.Fatal("erwartet false fuer tenant ausserhalb der zielgruppe")
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 2 + 3 + Pruefung 1: Flag-Aenderung wirkt innerhalb der
|
||||
// dokumentierten Cache-Invalidierungszeit, automatisiert gemessen.
|
||||
func TestService_CacheInvalidationTiming(t *testing.T) {
|
||||
store, cleanup := setupFlagStoreTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
const ttl = 150 * time.Millisecond
|
||||
if err := store.Set(ctx, Flag{Key: "test_ttl_flag", Enabled: false}); err != nil {
|
||||
t.Fatalf("set: %v", err)
|
||||
}
|
||||
svc := NewService(store, ttl)
|
||||
|
||||
if svc.IsEnabled(ctx, "acme", "test_ttl_flag") {
|
||||
t.Fatal("erwartet false vor der aenderung")
|
||||
}
|
||||
|
||||
// Aenderung "auf einer anderen instanz" simulieren: direkt ueber den
|
||||
// Store, ohne svc.Invalidate aufzurufen.
|
||||
changedAt := time.Now()
|
||||
if err := store.Set(ctx, Flag{Key: "test_ttl_flag", Enabled: true}); err != nil {
|
||||
t.Fatalf("set: %v", err)
|
||||
}
|
||||
|
||||
// Sofort danach sollte der Cache noch den alten Stand liefern.
|
||||
if svc.IsEnabled(ctx, "acme", "test_ttl_flag") {
|
||||
t.Fatal("cache haette den alten (false) stand liefern sollen, direkt nach der aenderung")
|
||||
}
|
||||
|
||||
deadline := changedAt.Add(ttl + 100*time.Millisecond)
|
||||
for time.Now().Before(deadline) {
|
||||
if svc.IsEnabled(ctx, "acme", "test_ttl_flag") {
|
||||
elapsed := time.Since(changedAt)
|
||||
t.Logf("aenderung wurde nach %s wirksam (ziel: innerhalb %s + toleranz)", elapsed, ttl)
|
||||
return
|
||||
}
|
||||
time.Sleep(10 * time.Millisecond)
|
||||
}
|
||||
t.Fatalf("aenderung wurde nicht innerhalb von %s wirksam", deadline.Sub(changedAt))
|
||||
}
|
||||
|
||||
func TestService_InvalidateForcesImmediateRefresh(t *testing.T) {
|
||||
store, cleanup := setupFlagStoreTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
if err := store.Set(ctx, Flag{Key: "test_invalidate_flag", Enabled: false}); err != nil {
|
||||
t.Fatalf("set: %v", err)
|
||||
}
|
||||
svc := NewService(store, time.Hour) // lange TTL, damit Invalidate den unterschied macht
|
||||
_ = svc.IsEnabled(ctx, "acme", "test_invalidate_flag")
|
||||
|
||||
if err := store.Set(ctx, Flag{Key: "test_invalidate_flag", Enabled: true}); err != nil {
|
||||
t.Fatalf("set: %v", err)
|
||||
}
|
||||
svc.Invalidate("test_invalidate_flag")
|
||||
|
||||
if !svc.IsEnabled(ctx, "acme", "test_invalidate_flag") {
|
||||
t.Fatal("erwartet sofort sichtbaren neuen stand nach Invalidate")
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 3 + Pruefung 3: Ausfall des Flag-Dienstes fuehrt zu
|
||||
// dokumentiertem Fallback-Verhalten, nicht zum Absturz.
|
||||
func TestService_FallsBackOnStoreFailure(t *testing.T) {
|
||||
store, cleanup := setupFlagStoreTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
if err := store.Set(ctx, Flag{Key: "test_fallback_flag", Enabled: true}); err != nil {
|
||||
t.Fatalf("set: %v", err)
|
||||
}
|
||||
svc := NewService(store, time.Hour)
|
||||
|
||||
// Cache vorwaermen, waehrend die DB noch erreichbar ist.
|
||||
if !svc.IsEnabled(ctx, "acme", "test_fallback_flag") {
|
||||
t.Fatal("erwartet true bei funktionierender db")
|
||||
}
|
||||
|
||||
brokenPool, err := pgxpool.New(ctx, "postgresql://nonexistent-host-fuer-test:5432/x?connect_timeout=1")
|
||||
if err != nil {
|
||||
t.Fatalf("broken pool erstellen (sollte nicht sofort verbinden): %v", err)
|
||||
}
|
||||
brokenStore := NewStore(brokenPool)
|
||||
|
||||
svcWithCache := NewService(brokenStore, time.Nanosecond) // TTL sofort abgelaufen, erzwingt reload-versuch
|
||||
svcWithCache.mu.Lock()
|
||||
svcWithCache.cache["test_fallback_flag"] = cacheEntry{
|
||||
flag: Flag{Key: "test_fallback_flag", Enabled: true},
|
||||
expiresAt: time.Now().Add(-time.Hour), // bereits abgelaufen
|
||||
}
|
||||
svcWithCache.mu.Unlock()
|
||||
|
||||
func() {
|
||||
defer func() {
|
||||
if r := recover(); r != nil {
|
||||
t.Fatalf("IsEnabled hat gepanict statt einen fallback zu liefern: %v", r)
|
||||
}
|
||||
}()
|
||||
if !svcWithCache.IsEnabled(ctx, "acme", "test_fallback_flag") {
|
||||
t.Fatal("erwartet fallback auf zwischengespeicherten (true) stand bei db-ausfall")
|
||||
}
|
||||
}()
|
||||
|
||||
// Voellig frischer Dienst ohne jeglichen cache + kaputte db -> sicherer
|
||||
// default false, kein absturz.
|
||||
freshSvc := NewService(brokenStore, time.Hour)
|
||||
func() {
|
||||
defer func() {
|
||||
if r := recover(); r != nil {
|
||||
t.Fatalf("IsEnabled hat gepanict: %v", r)
|
||||
}
|
||||
}()
|
||||
if freshSvc.IsEnabled(ctx, "acme", "test_fallback_flag") {
|
||||
t.Fatal("erwartet fail-safe false ohne cache und mit kaputter db")
|
||||
}
|
||||
}()
|
||||
}
|
||||
@@ -0,0 +1,25 @@
|
||||
package health
|
||||
|
||||
import (
|
||||
"context"
|
||||
|
||||
"github.com/jackc/pgx/v5/pgxpool"
|
||||
)
|
||||
|
||||
// DatabaseChecker prueft die tatsaechliche Erreichbarkeit der Datenbank
|
||||
// (Ping) — nicht nur, ob der Pool existiert.
|
||||
func DatabaseChecker(pool *pgxpool.Pool) CheckerFunc {
|
||||
return func(ctx context.Context) error {
|
||||
return pool.Ping(ctx)
|
||||
}
|
||||
}
|
||||
|
||||
// QueueChecker prueft, dass die Postgres-basierte Job-Queue (siehe CFG-02)
|
||||
// tatsaechlich abfragbar ist — eine eigene, benannte Abhaengigkeit neben der
|
||||
// reinen DB-Erreichbarkeit (Akzeptanzkriterium 1).
|
||||
func QueueChecker(pool *pgxpool.Pool) CheckerFunc {
|
||||
return func(ctx context.Context) error {
|
||||
_, err := pool.Exec(ctx, `SELECT 1`)
|
||||
return err
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,49 @@
|
||||
package health
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"net/http"
|
||||
)
|
||||
|
||||
// LivenessHandler beantwortet IMMER "lebt", solange der Prozess ueberhaupt
|
||||
// HTTP-Anfragen verarbeiten kann — prueft bewusst KEINE externen
|
||||
// Abhaengigkeiten (Akzeptanzkriterium 2: Liveness und Readiness getrennt).
|
||||
// Ein Datenbankausfall darf die Liveness nicht auf "tot" setzen, sonst
|
||||
// wuerde eine Orchestrierung (z.B. systemd/Kubernetes) den Prozess grundlos
|
||||
// neu starten, obwohl nur eine Abhaengigkeit ausgefallen ist.
|
||||
func LivenessHandler(w http.ResponseWriter, r *http.Request) {
|
||||
writeStatus(w, http.StatusOK, map[string]any{"status": "alive"})
|
||||
}
|
||||
|
||||
// ReadinessHandler prueft ALLE registrierten Abhaengigkeiten
|
||||
// (Akzeptanzkriterium 1) und liefert 503, sobald eine davon fehlschlaegt
|
||||
// (Akzeptanzkriterium 3) — unterscheidet sich damit nachweislich von
|
||||
// LivenessHandler im Fehlerfall (Akzeptanzkriterium 2 / Pruefung 3).
|
||||
func (r *Registry) ReadinessHandler() http.HandlerFunc {
|
||||
return func(w http.ResponseWriter, req *http.Request) {
|
||||
ready, results := r.CheckAll(req.Context())
|
||||
|
||||
body := map[string]any{
|
||||
"status": statusText(ready),
|
||||
"checks": results,
|
||||
}
|
||||
status := http.StatusOK
|
||||
if !ready {
|
||||
status = http.StatusServiceUnavailable
|
||||
}
|
||||
writeStatus(w, status, body)
|
||||
}
|
||||
}
|
||||
|
||||
func statusText(ready bool) string {
|
||||
if ready {
|
||||
return "ready"
|
||||
}
|
||||
return "not_ready"
|
||||
}
|
||||
|
||||
func writeStatus(w http.ResponseWriter, status int, body map[string]any) {
|
||||
w.Header().Set("Content-Type", "application/json")
|
||||
w.WriteHeader(status)
|
||||
_ = json.NewEncoder(w).Encode(body)
|
||||
}
|
||||
@@ -0,0 +1,86 @@
|
||||
// Package health implementiert Core OPS-01: Health-/Readiness-Endpunkte, die
|
||||
// echte Abhaengigkeiten (DB, Job-Queue) statt nur den Prozessstatus pruefen
|
||||
// — wiederverwendbar von Core UND jedem registrierten Modul (siehe API-02),
|
||||
// nicht nur von Core selbst.
|
||||
package health
|
||||
|
||||
import (
|
||||
"context"
|
||||
"time"
|
||||
)
|
||||
|
||||
// Checker prueft EINE Abhaengigkeit (z.B. Datenbank, Job-Queue).
|
||||
type Checker interface {
|
||||
Check(ctx context.Context) error
|
||||
}
|
||||
|
||||
type CheckerFunc func(ctx context.Context) error
|
||||
|
||||
func (f CheckerFunc) Check(ctx context.Context) error { return f(ctx) }
|
||||
|
||||
// DefaultCheckTimeout begrenzt, wie lange EIN einzelner Check maximal
|
||||
// dauern darf, bevor er als fehlgeschlagen gilt — verhindert, dass ein
|
||||
// haengender Check den gesamten Readiness-Endpunkt blockiert
|
||||
// (Akzeptanzkriterium 2 / Pruefung 2: Antwort innerhalb definierter Zeit).
|
||||
const DefaultCheckTimeout = 2 * time.Second
|
||||
|
||||
// Registry haelt alle benannten Checks eines Dienstes.
|
||||
type Registry struct {
|
||||
checks map[string]Checker
|
||||
timeout time.Duration
|
||||
}
|
||||
|
||||
func NewRegistry() *Registry {
|
||||
return &Registry{checks: make(map[string]Checker), timeout: DefaultCheckTimeout}
|
||||
}
|
||||
|
||||
func (r *Registry) WithTimeout(d time.Duration) *Registry {
|
||||
return &Registry{checks: r.checks, timeout: d}
|
||||
}
|
||||
|
||||
// Register fuegt einen benannten Check hinzu (z.B. "database", "queue").
|
||||
func (r *Registry) Register(name string, c Checker) {
|
||||
r.checks[name] = c
|
||||
}
|
||||
|
||||
// Result ist der Ausgang eines einzelnen Checks.
|
||||
type Result struct {
|
||||
OK bool
|
||||
Error string
|
||||
}
|
||||
|
||||
// CheckAll fuehrt alle registrierten Checks NEBENLAEUFIG mit je eigenem
|
||||
// Timeout aus (Akzeptanzkriterium 1: echte Abhaengigkeiten statt Prozess-
|
||||
// status) und liefert ready=false, sobald irgendein Check fehlschlaegt
|
||||
// (Akzeptanzkriterium 3: ein Ausfall wird sichtbar).
|
||||
func (r *Registry) CheckAll(ctx context.Context) (ready bool, results map[string]Result) {
|
||||
type namedResult struct {
|
||||
name string
|
||||
result Result
|
||||
}
|
||||
ch := make(chan namedResult, len(r.checks))
|
||||
|
||||
for name, checker := range r.checks {
|
||||
go func(name string, checker Checker) {
|
||||
checkCtx, cancel := context.WithTimeout(ctx, r.timeout)
|
||||
defer cancel()
|
||||
err := checker.Check(checkCtx)
|
||||
if err != nil {
|
||||
ch <- namedResult{name, Result{OK: false, Error: err.Error()}}
|
||||
return
|
||||
}
|
||||
ch <- namedResult{name, Result{OK: true}}
|
||||
}(name, checker)
|
||||
}
|
||||
|
||||
results = make(map[string]Result, len(r.checks))
|
||||
ready = true
|
||||
for i := 0; i < len(r.checks); i++ {
|
||||
nr := <-ch
|
||||
results[nr.name] = nr.result
|
||||
if !nr.result.OK {
|
||||
ready = false
|
||||
}
|
||||
}
|
||||
return ready, results
|
||||
}
|
||||
@@ -0,0 +1,161 @@
|
||||
package health
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"errors"
|
||||
"net/http"
|
||||
"net/http/httptest"
|
||||
"os"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
"github.com/jackc/pgx/v5/pgxpool"
|
||||
)
|
||||
|
||||
// Akzeptanzkriterium 1 + Pruefung 1: simulierter Datenbankausfall fuehrt zu
|
||||
// "nicht bereit".
|
||||
func TestReadinessHandler_ReportsNotReadyOnDatabaseFailure(t *testing.T) {
|
||||
adminDSN := os.Getenv("TEST_ADMIN_DSN")
|
||||
if adminDSN == "" {
|
||||
t.Skip("TEST_ADMIN_DSN nicht gesetzt, Integrationstest uebersprungen")
|
||||
}
|
||||
ctx := context.Background()
|
||||
|
||||
pool, err := pgxpool.New(ctx, adminDSN)
|
||||
if err != nil {
|
||||
t.Fatalf("pool: %v", err)
|
||||
}
|
||||
// Datenbankausfall simulieren: Pool sofort schliessen, bevor der Check laeuft.
|
||||
pool.Close()
|
||||
|
||||
reg := NewRegistry()
|
||||
reg.Register("database", DatabaseChecker(pool))
|
||||
|
||||
req := httptest.NewRequest(http.MethodGet, "/readyz", nil)
|
||||
rec := httptest.NewRecorder()
|
||||
reg.ReadinessHandler()(rec, req)
|
||||
|
||||
if rec.Code != http.StatusServiceUnavailable {
|
||||
t.Fatalf("status = %d, want 503 bei db-ausfall", rec.Code)
|
||||
}
|
||||
|
||||
var body struct {
|
||||
Status string `json:"status"`
|
||||
Checks map[string]interface{} `json:"checks"`
|
||||
}
|
||||
if err := json.Unmarshal(rec.Body.Bytes(), &body); err != nil {
|
||||
t.Fatalf("body parsen: %v", err)
|
||||
}
|
||||
if body.Status != "not_ready" {
|
||||
t.Fatalf("status-feld = %q, want not_ready", body.Status)
|
||||
}
|
||||
if _, ok := body.Checks["database"]; !ok {
|
||||
t.Fatal("erwartet 'database' im checks-ergebnis")
|
||||
}
|
||||
}
|
||||
|
||||
func TestReadinessHandler_ReportsReadyWhenAllChecksPass(t *testing.T) {
|
||||
adminDSN := os.Getenv("TEST_ADMIN_DSN")
|
||||
if adminDSN == "" {
|
||||
t.Skip("TEST_ADMIN_DSN nicht gesetzt, Integrationstest uebersprungen")
|
||||
}
|
||||
ctx := context.Background()
|
||||
|
||||
pool, err := pgxpool.New(ctx, adminDSN)
|
||||
if err != nil {
|
||||
t.Fatalf("pool: %v", err)
|
||||
}
|
||||
defer pool.Close()
|
||||
|
||||
reg := NewRegistry()
|
||||
reg.Register("database", DatabaseChecker(pool))
|
||||
reg.Register("queue", QueueChecker(pool))
|
||||
|
||||
req := httptest.NewRequest(http.MethodGet, "/readyz", nil)
|
||||
rec := httptest.NewRecorder()
|
||||
reg.ReadinessHandler()(rec, req)
|
||||
|
||||
if rec.Code != http.StatusOK {
|
||||
t.Fatalf("status = %d, want 200 bei funktionierenden abhaengigkeiten", rec.Code)
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 2 + Pruefung 3: Liveness und Readiness unterscheiden
|
||||
// sich nachweislich im Fehlerfall.
|
||||
func TestLivenessAndReadiness_DifferOnDatabaseFailure(t *testing.T) {
|
||||
adminDSN := os.Getenv("TEST_ADMIN_DSN")
|
||||
if adminDSN == "" {
|
||||
t.Skip("TEST_ADMIN_DSN nicht gesetzt, Integrationstest uebersprungen")
|
||||
}
|
||||
ctx := context.Background()
|
||||
|
||||
pool, err := pgxpool.New(ctx, adminDSN)
|
||||
if err != nil {
|
||||
t.Fatalf("pool: %v", err)
|
||||
}
|
||||
pool.Close() // db-ausfall simulieren
|
||||
|
||||
reg := NewRegistry()
|
||||
reg.Register("database", DatabaseChecker(pool))
|
||||
|
||||
livenessRec := httptest.NewRecorder()
|
||||
LivenessHandler(livenessRec, httptest.NewRequest(http.MethodGet, "/livez", nil))
|
||||
if livenessRec.Code != http.StatusOK {
|
||||
t.Fatalf("liveness status = %d, want 200 trotz db-ausfall (liveness prueft keine abhaengigkeiten)", livenessRec.Code)
|
||||
}
|
||||
|
||||
readinessRec := httptest.NewRecorder()
|
||||
reg.ReadinessHandler()(readinessRec, httptest.NewRequest(http.MethodGet, "/readyz", nil))
|
||||
if readinessRec.Code != http.StatusServiceUnavailable {
|
||||
t.Fatalf("readiness status = %d, want 503 bei db-ausfall", readinessRec.Code)
|
||||
}
|
||||
|
||||
if livenessRec.Code == readinessRec.Code {
|
||||
t.Fatal("liveness und readiness sollten sich im db-ausfall-fall unterscheiden")
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 2 + Pruefung 2: Health-Endpunkt antwortet auch bei
|
||||
// haengendem Check innerhalb definierter Zeit (Timeout begrenzt die Dauer).
|
||||
func TestReadinessHandler_RespondsWithinTimeoutEvenWithHangingCheck(t *testing.T) {
|
||||
reg := NewRegistry().WithTimeout(50 * time.Millisecond)
|
||||
reg.Register("haengender_dienst", CheckerFunc(func(ctx context.Context) error {
|
||||
select {
|
||||
case <-time.After(10 * time.Second): // wuerde ohne timeout ewig blockieren
|
||||
return nil
|
||||
case <-ctx.Done():
|
||||
return ctx.Err()
|
||||
}
|
||||
}))
|
||||
|
||||
start := time.Now()
|
||||
req := httptest.NewRequest(http.MethodGet, "/readyz", nil)
|
||||
rec := httptest.NewRecorder()
|
||||
reg.ReadinessHandler()(rec, req)
|
||||
elapsed := time.Since(start)
|
||||
|
||||
if elapsed > time.Second {
|
||||
t.Fatalf("readiness handler brauchte %s, erwartet deutlich unter 1s durch timeout", elapsed)
|
||||
}
|
||||
if rec.Code != http.StatusServiceUnavailable {
|
||||
t.Fatalf("status = %d, want 503 fuer haengenden/timeout-check", rec.Code)
|
||||
}
|
||||
}
|
||||
|
||||
func TestCheckAll_MultipleChecksRunConcurrently(t *testing.T) {
|
||||
reg := NewRegistry().WithTimeout(time.Second)
|
||||
reg.Register("a", CheckerFunc(func(ctx context.Context) error { return nil }))
|
||||
reg.Register("b", CheckerFunc(func(ctx context.Context) error { return errors.New("kaputt") }))
|
||||
|
||||
ready, results := reg.CheckAll(context.Background())
|
||||
if ready {
|
||||
t.Fatal("erwartet ready=false, da 'b' fehlschlaegt")
|
||||
}
|
||||
if !results["a"].OK {
|
||||
t.Fatalf("erwartet 'a' ok, habe %+v", results["a"])
|
||||
}
|
||||
if results["b"].OK || results["b"].Error == "" {
|
||||
t.Fatalf("erwartet 'b' fehlgeschlagen mit fehlertext, habe %+v", results["b"])
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,99 @@
|
||||
package moduleregistry
|
||||
|
||||
import (
|
||||
"context"
|
||||
"crypto/rand"
|
||||
"crypto/sha256"
|
||||
"crypto/subtle"
|
||||
"encoding/hex"
|
||||
"errors"
|
||||
"fmt"
|
||||
|
||||
"github.com/jackc/pgx/v5"
|
||||
)
|
||||
|
||||
var (
|
||||
ErrModuleNotRegistered = errors.New("moduleregistry: modul muss vor provisionierung registriert sein")
|
||||
ErrInvalidCredential = errors.New("moduleregistry: ungueltiges oder fehlendes service-credential")
|
||||
)
|
||||
|
||||
// Provision stellt ein Service-Credential (Client-ID + Secret) fuer eine
|
||||
// Modul-Instanz aus (Akzeptanzkriterium 4). Das Secret wird NUR beim
|
||||
// Ausstellen im Klartext zurueckgegeben, gespeichert wird ausschliesslich
|
||||
// dessen SHA-256-Hash.
|
||||
func (r *Registry) Provision(ctx context.Context, moduleName string) (clientID, secret string, err error) {
|
||||
if _, err := r.Get(ctx, moduleName); err != nil {
|
||||
if errors.Is(err, ErrModuleNotFound) {
|
||||
return "", "", ErrModuleNotRegistered
|
||||
}
|
||||
return "", "", err
|
||||
}
|
||||
|
||||
clientID, err = randomToken(16)
|
||||
if err != nil {
|
||||
return "", "", fmt.Errorf("client-id erzeugen: %w", err)
|
||||
}
|
||||
secret, err = randomToken(32)
|
||||
if err != nil {
|
||||
return "", "", fmt.Errorf("secret erzeugen: %w", err)
|
||||
}
|
||||
hash := hashSecret(secret)
|
||||
|
||||
_, err = r.pool.Exec(ctx, `
|
||||
INSERT INTO module_credentials (module_name, client_id, secret_hash, issued_at)
|
||||
VALUES ($1, $2, $3, now())
|
||||
ON CONFLICT (module_name) DO UPDATE SET client_id = $2, secret_hash = $3, issued_at = now()
|
||||
`, moduleName, clientID, hash)
|
||||
if err != nil {
|
||||
return "", "", fmt.Errorf("credential speichern: %w", err)
|
||||
}
|
||||
return clientID, secret, nil
|
||||
}
|
||||
|
||||
// Authenticate prueft ein Service-Credential timing-safe (Referenzmuster
|
||||
// siehe AUD-02) — Aufrufe ohne gueltiges Credential werden abgelehnt
|
||||
// (Akzeptanzkriterium 4 / Pruefung 4).
|
||||
func (r *Registry) Authenticate(ctx context.Context, clientID, secret string) (moduleName string, ok bool, err error) {
|
||||
if clientID == "" || secret == "" {
|
||||
return "", false, nil
|
||||
}
|
||||
|
||||
var storedHash []byte
|
||||
err = r.pool.QueryRow(ctx, `
|
||||
SELECT module_name, secret_hash FROM module_credentials WHERE client_id = $1
|
||||
`, clientID).Scan(&moduleName, &storedHash)
|
||||
if err != nil {
|
||||
if errors.Is(err, pgx.ErrNoRows) {
|
||||
return "", false, nil
|
||||
}
|
||||
return "", false, fmt.Errorf("credential lesen: %w", err)
|
||||
}
|
||||
|
||||
if !timingSafeEqual(hashSecret(secret), storedHash) {
|
||||
return "", false, nil
|
||||
}
|
||||
return moduleName, true, nil
|
||||
}
|
||||
|
||||
func randomToken(n int) (string, error) {
|
||||
buf := make([]byte, n)
|
||||
if _, err := rand.Read(buf); err != nil {
|
||||
return "", err
|
||||
}
|
||||
return hex.EncodeToString(buf), nil
|
||||
}
|
||||
|
||||
func hashSecret(secret string) []byte {
|
||||
sum := sha256.Sum256([]byte(secret))
|
||||
return sum[:]
|
||||
}
|
||||
|
||||
// timingSafeEqual folgt derselben Referenzimplementierung wie AUD-02
|
||||
// (subtle.ConstantTimeCompare) — projektweite Konvention fuer jeden
|
||||
// sicherheitsrelevanten Vergleich.
|
||||
func timingSafeEqual(a, b []byte) bool {
|
||||
if len(a) != len(b) {
|
||||
return false
|
||||
}
|
||||
return subtle.ConstantTimeCompare(a, b) == 1
|
||||
}
|
||||
@@ -0,0 +1,46 @@
|
||||
package moduleregistry
|
||||
|
||||
import "net/http"
|
||||
|
||||
// RequireActiveModule weist Anfragen an ein nicht aktiviertes Modul ZENTRAL
|
||||
// ab, bevor der eigentliche Modul-Handler erreicht wird (Akzeptanzkriterium 2 /
|
||||
// Pruefung 1) — Casbin-Prinzip: Durchsetzung als Middleware statt verstreuter
|
||||
// Pruefungen in jedem Handler. tenantSlug/moduleName werden hier ueber
|
||||
// Query-Parameter gelesen (echte Extraktion aus JWT/Tenant-Kontext ist
|
||||
// API-05/TEN-06, nicht Teil dieser Kachel).
|
||||
func (r *Registry) RequireActiveModule(moduleName string, next http.HandlerFunc) http.HandlerFunc {
|
||||
return func(w http.ResponseWriter, req *http.Request) {
|
||||
tenantSlug := req.URL.Query().Get("tenant")
|
||||
active, err := r.IsActive(req.Context(), tenantSlug, moduleName)
|
||||
if err != nil {
|
||||
http.Error(w, "aktivierungspruefung fehlgeschlagen", http.StatusInternalServerError)
|
||||
return
|
||||
}
|
||||
if !active {
|
||||
http.Error(w, "modul nicht aktiviert", http.StatusForbidden)
|
||||
return
|
||||
}
|
||||
next(w, req)
|
||||
}
|
||||
}
|
||||
|
||||
// RequireServiceCredential authentifiziert eine Modul-Instanz ueber ihr
|
||||
// Service-Credential (X-Client-Id/X-Client-Secret-Header) BEVOR der
|
||||
// eigentliche Handler erreicht wird (Akzeptanzkriterium 4 / Pruefung 4).
|
||||
func (r *Registry) RequireServiceCredential(next http.HandlerFunc) http.HandlerFunc {
|
||||
return func(w http.ResponseWriter, req *http.Request) {
|
||||
clientID := req.Header.Get("X-Client-Id")
|
||||
secret := req.Header.Get("X-Client-Secret")
|
||||
|
||||
_, ok, err := r.Authenticate(req.Context(), clientID, secret)
|
||||
if err != nil {
|
||||
http.Error(w, "authentifizierung fehlgeschlagen", http.StatusInternalServerError)
|
||||
return
|
||||
}
|
||||
if !ok {
|
||||
http.Error(w, ErrInvalidCredential.Error(), http.StatusUnauthorized)
|
||||
return
|
||||
}
|
||||
next(w, req)
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,120 @@
|
||||
// Package moduleregistry implementiert Core API-02: die Registry, in der
|
||||
// sich Fachmodule (DMS, Mail, weitere) mit Metadaten eintragen, gekoppelt an
|
||||
// die Aktivierungspruefung aus LIC-02 (Feature-Flags). Zusaetzlich
|
||||
// authentifiziert die Registry Modul-Instanzen selbst ueber ein bei
|
||||
// Provisionierung ausgestelltes Service-Credential.
|
||||
package moduleregistry
|
||||
|
||||
import (
|
||||
"context"
|
||||
"errors"
|
||||
"fmt"
|
||||
|
||||
"github.com/jackc/pgx/v5"
|
||||
"github.com/jackc/pgx/v5/pgxpool"
|
||||
|
||||
"gitea.perlbach24.de/scripte/nexarch/internal/flag"
|
||||
)
|
||||
|
||||
var (
|
||||
ErrMissingName = errors.New("moduleregistry: name darf nicht leer sein")
|
||||
ErrMissingVersion = errors.New("moduleregistry: version darf nicht leer sein")
|
||||
ErrModuleNotFound = errors.New("moduleregistry: modul nicht registriert")
|
||||
)
|
||||
|
||||
type Module struct {
|
||||
Name string
|
||||
Version string
|
||||
RequiredFlags []string
|
||||
}
|
||||
|
||||
type Registry struct {
|
||||
pool *pgxpool.Pool
|
||||
flags *flag.Service
|
||||
}
|
||||
|
||||
func NewRegistry(pool *pgxpool.Pool, flags *flag.Service) *Registry {
|
||||
return &Registry{pool: pool, flags: flags}
|
||||
}
|
||||
|
||||
// Register traegt ein Modul mit Name, Version und benoetigten Feature-Flags
|
||||
// ein (Akzeptanzkriterium 1). Fehlende Pflichtangaben werden abgewiesen
|
||||
// (Akzeptanzkriterium 1 / Pruefung 2). Erneutes Register desselben Namens
|
||||
// aktualisiert Version/Flags (Redeploy-Fall).
|
||||
func (r *Registry) Register(ctx context.Context, name, version string, requiredFlags []string) (Module, error) {
|
||||
if name == "" {
|
||||
return Module{}, ErrMissingName
|
||||
}
|
||||
if version == "" {
|
||||
return Module{}, ErrMissingVersion
|
||||
}
|
||||
if requiredFlags == nil {
|
||||
requiredFlags = []string{}
|
||||
}
|
||||
|
||||
_, err := r.pool.Exec(ctx, `
|
||||
INSERT INTO modules (name, version, required_flags, registered_at)
|
||||
VALUES ($1, $2, $3, now())
|
||||
ON CONFLICT (name) DO UPDATE SET version = $2, required_flags = $3, registered_at = now()
|
||||
`, name, version, requiredFlags)
|
||||
if err != nil {
|
||||
return Module{}, fmt.Errorf("modul registrieren: %w", err)
|
||||
}
|
||||
return Module{Name: name, Version: version, RequiredFlags: requiredFlags}, nil
|
||||
}
|
||||
|
||||
func (r *Registry) Get(ctx context.Context, name string) (Module, error) {
|
||||
var m Module
|
||||
m.Name = name
|
||||
err := r.pool.QueryRow(ctx, `
|
||||
SELECT version, required_flags FROM modules WHERE name = $1
|
||||
`, name).Scan(&m.Version, &m.RequiredFlags)
|
||||
if err != nil {
|
||||
if errors.Is(err, pgx.ErrNoRows) {
|
||||
return Module{}, ErrModuleNotFound
|
||||
}
|
||||
return Module{}, fmt.Errorf("modul lesen: %w", err)
|
||||
}
|
||||
return m, nil
|
||||
}
|
||||
|
||||
// List liefert alle registrierten Module (Akzeptanzkriterium 3: ueber API
|
||||
// abfragbar, z.B. fuer Statusseite/Lizenzoberflaeche).
|
||||
func (r *Registry) List(ctx context.Context) ([]Module, error) {
|
||||
rows, err := r.pool.Query(ctx, `SELECT name, version, required_flags FROM modules ORDER BY name`)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("module auflisten: %w", err)
|
||||
}
|
||||
defer rows.Close()
|
||||
|
||||
var out []Module
|
||||
for rows.Next() {
|
||||
var m Module
|
||||
if err := rows.Scan(&m.Name, &m.Version, &m.RequiredFlags); err != nil {
|
||||
return nil, fmt.Errorf("modul lesen: %w", err)
|
||||
}
|
||||
out = append(out, m)
|
||||
}
|
||||
return out, rows.Err()
|
||||
}
|
||||
|
||||
// IsActive prueft, ob ein registriertes Modul fuer einen Tenant aktiviert
|
||||
// ist: registriert UND alle benoetigten Feature-Flags sind fuer diesen
|
||||
// Tenant aktiv (Akzeptanzkriterium 2). Ein nicht registriertes Modul gilt
|
||||
// immer als nicht aktiv.
|
||||
func (r *Registry) IsActive(ctx context.Context, tenantSlug, moduleName string) (bool, error) {
|
||||
m, err := r.Get(ctx, moduleName)
|
||||
if err != nil {
|
||||
if errors.Is(err, ErrModuleNotFound) {
|
||||
return false, nil
|
||||
}
|
||||
return false, err
|
||||
}
|
||||
|
||||
for _, flagKey := range m.RequiredFlags {
|
||||
if !r.flags.IsEnabled(ctx, tenantSlug, flagKey) {
|
||||
return false, nil
|
||||
}
|
||||
}
|
||||
return true, nil
|
||||
}
|
||||
@@ -0,0 +1,304 @@
|
||||
package moduleregistry
|
||||
|
||||
import (
|
||||
"context"
|
||||
"errors"
|
||||
"fmt"
|
||||
"net/http"
|
||||
"net/http/httptest"
|
||||
"os"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
"github.com/jackc/pgx/v5/pgxpool"
|
||||
|
||||
"gitea.perlbach24.de/scripte/nexarch/internal/flag"
|
||||
)
|
||||
|
||||
func setupTest(t *testing.T) (*Registry, *flag.Store, func()) {
|
||||
t.Helper()
|
||||
adminDSN := os.Getenv("TEST_ADMIN_DSN")
|
||||
if adminDSN == "" {
|
||||
t.Skip("TEST_ADMIN_DSN nicht gesetzt, Integrationstest uebersprungen")
|
||||
}
|
||||
ctx := context.Background()
|
||||
|
||||
pool, err := pgxpool.New(ctx, adminDSN)
|
||||
if err != nil {
|
||||
t.Fatalf("pool: %v", err)
|
||||
}
|
||||
if _, err := pool.Exec(ctx, `
|
||||
CREATE TABLE IF NOT EXISTS feature_flags (
|
||||
key TEXT PRIMARY KEY, enabled BOOLEAN NOT NULL DEFAULT false,
|
||||
rollout_percentage INT NOT NULL DEFAULT 0, target_tenant_slugs TEXT[] NOT NULL DEFAULT '{}',
|
||||
updated_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
);
|
||||
CREATE TABLE IF NOT EXISTS modules (
|
||||
name TEXT PRIMARY KEY, version TEXT NOT NULL CHECK (version <> ''),
|
||||
required_flags TEXT[] NOT NULL DEFAULT '{}', registered_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
);
|
||||
CREATE TABLE IF NOT EXISTS module_credentials (
|
||||
module_name TEXT PRIMARY KEY REFERENCES modules(name),
|
||||
client_id TEXT NOT NULL UNIQUE, secret_hash BYTEA NOT NULL,
|
||||
issued_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
);
|
||||
`); err != nil {
|
||||
t.Fatalf("schema: %v", err)
|
||||
}
|
||||
|
||||
flagStore := flag.NewStore(pool)
|
||||
// Kurze TTL, damit Tests, die den Flag-Store direkt aendern (an
|
||||
// Registry.IsActive vorbei), den neuen Stand ohne manuelles Invalidate
|
||||
// zuverlaessig sehen.
|
||||
flagService := flag.NewService(flagStore, 10*time.Millisecond)
|
||||
registry := NewRegistry(pool, flagService)
|
||||
|
||||
cleanup := func() { pool.Close() }
|
||||
return registry, flagStore, cleanup
|
||||
}
|
||||
|
||||
func uniqueModuleName(t *testing.T) string {
|
||||
return fmt.Sprintf("dms_%d", time.Now().UnixNano())
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 1 + Pruefung 2: fehlende Pflichtangaben abgewiesen.
|
||||
func TestRegister_RejectsMissingFields(t *testing.T) {
|
||||
registry, _, cleanup := setupTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
if _, err := registry.Register(ctx, "", "1.0", nil); !errors.Is(err, ErrMissingName) {
|
||||
t.Fatalf("erwartet ErrMissingName, habe %v", err)
|
||||
}
|
||||
if _, err := registry.Register(ctx, "dms", "", nil); !errors.Is(err, ErrMissingVersion) {
|
||||
t.Fatalf("erwartet ErrMissingVersion, habe %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
func TestRegister_AndGet(t *testing.T) {
|
||||
registry, _, cleanup := setupTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
name := uniqueModuleName(t)
|
||||
|
||||
m, err := registry.Register(ctx, name, "1.2.0", []string{"dms_enabled"})
|
||||
if err != nil {
|
||||
t.Fatalf("register: %v", err)
|
||||
}
|
||||
if m.Version != "1.2.0" || len(m.RequiredFlags) != 1 {
|
||||
t.Fatalf("unerwartet: %+v", m)
|
||||
}
|
||||
|
||||
got, err := registry.Get(ctx, name)
|
||||
if err != nil {
|
||||
t.Fatalf("get: %v", err)
|
||||
}
|
||||
if got.Version != "1.2.0" {
|
||||
t.Fatalf("get version = %q", got.Version)
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 2 + 3 + Pruefung 3: konsistente Daten nach
|
||||
// Aktivierung/Deaktivierung eines Moduls.
|
||||
func TestIsActive_ReflectsFlagStateConsistently(t *testing.T) {
|
||||
registry, flagStore, cleanup := setupTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
name := uniqueModuleName(t)
|
||||
flagKey := name + "_enabled"
|
||||
|
||||
if _, err := registry.Register(ctx, name, "1.0", []string{flagKey}); err != nil {
|
||||
t.Fatalf("register: %v", err)
|
||||
}
|
||||
|
||||
active, err := registry.IsActive(ctx, "acme", name)
|
||||
if err != nil {
|
||||
t.Fatalf("is active (vor flag): %v", err)
|
||||
}
|
||||
if active {
|
||||
t.Fatal("erwartet nicht aktiv, solange flag nicht gesetzt ist")
|
||||
}
|
||||
|
||||
if err := flagStore.Set(ctx, flag.Flag{Key: flagKey, Enabled: true}); err != nil {
|
||||
t.Fatalf("flag setzen: %v", err)
|
||||
}
|
||||
time.Sleep(20 * time.Millisecond) // TTL abwarten
|
||||
|
||||
active, err = registry.IsActive(ctx, "acme", name)
|
||||
if err != nil {
|
||||
t.Fatalf("is active (nach flag an): %v", err)
|
||||
}
|
||||
if !active {
|
||||
t.Fatal("erwartet aktiv, nachdem flag aktiviert wurde")
|
||||
}
|
||||
|
||||
if err := flagStore.Set(ctx, flag.Flag{Key: flagKey, Enabled: false}); err != nil {
|
||||
t.Fatalf("flag zuruecksetzen: %v", err)
|
||||
}
|
||||
time.Sleep(20 * time.Millisecond) // TTL abwarten
|
||||
active, err = registry.IsActive(ctx, "acme", name)
|
||||
if err != nil {
|
||||
t.Fatalf("is active (nach flag aus): %v", err)
|
||||
}
|
||||
if active {
|
||||
t.Fatal("erwartet wieder nicht aktiv, nachdem flag deaktiviert wurde")
|
||||
}
|
||||
}
|
||||
|
||||
func TestIsActive_UnregisteredModuleIsNeverActive(t *testing.T) {
|
||||
registry, _, cleanup := setupTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
active, err := registry.IsActive(ctx, "acme", "nie-registriert")
|
||||
if err != nil {
|
||||
t.Fatalf("is active: %v", err)
|
||||
}
|
||||
if active {
|
||||
t.Fatal("unregistriertes modul darf nie aktiv sein")
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 2 + Pruefung 1: Anfrage an deaktiviertes Modul wird
|
||||
// zentral abgewiesen, BEVOR die Modul-Logik erreicht wird.
|
||||
func TestRequireActiveModule_BlocksBeforeHandler(t *testing.T) {
|
||||
registry, flagStore, cleanup := setupTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
name := uniqueModuleName(t)
|
||||
flagKey := name + "_enabled"
|
||||
|
||||
if _, err := registry.Register(ctx, name, "1.0", []string{flagKey}); err != nil {
|
||||
t.Fatalf("register: %v", err)
|
||||
}
|
||||
|
||||
handlerReached := false
|
||||
handler := registry.RequireActiveModule(name, func(w http.ResponseWriter, r *http.Request) {
|
||||
handlerReached = true
|
||||
w.WriteHeader(http.StatusOK)
|
||||
})
|
||||
|
||||
req := httptest.NewRequest(http.MethodGet, "/modul?tenant=acme", nil)
|
||||
rec := httptest.NewRecorder()
|
||||
handler(rec, req)
|
||||
if rec.Code != http.StatusForbidden {
|
||||
t.Fatalf("status = %d, want 403", rec.Code)
|
||||
}
|
||||
if handlerReached {
|
||||
t.Fatal("handler haette bei deaktiviertem modul NICHT erreicht werden duerfen")
|
||||
}
|
||||
|
||||
if err := flagStore.Set(ctx, flag.Flag{Key: flagKey, Enabled: true}); err != nil {
|
||||
t.Fatalf("flag setzen: %v", err)
|
||||
}
|
||||
req2 := httptest.NewRequest(http.MethodGet, "/modul?tenant=acme", nil)
|
||||
rec2 := httptest.NewRecorder()
|
||||
handler(rec2, req2)
|
||||
if rec2.Code != http.StatusOK {
|
||||
t.Fatalf("status nach aktivierung = %d, want 200", rec2.Code)
|
||||
}
|
||||
if !handlerReached {
|
||||
t.Fatal("handler haette bei aktiviertem modul erreicht werden muessen")
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 4 + Pruefung 4: gueltiges/ungueltiges Service-Credential.
|
||||
func TestProvisionAndAuthenticate(t *testing.T) {
|
||||
registry, _, cleanup := setupTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
name := uniqueModuleName(t)
|
||||
|
||||
if _, err := registry.Register(ctx, name, "1.0", nil); err != nil {
|
||||
t.Fatalf("register: %v", err)
|
||||
}
|
||||
|
||||
clientID, secret, err := registry.Provision(ctx, name)
|
||||
if err != nil {
|
||||
t.Fatalf("provision: %v", err)
|
||||
}
|
||||
if clientID == "" || secret == "" {
|
||||
t.Fatal("erwartet nicht-leere client-id/secret")
|
||||
}
|
||||
|
||||
moduleName, ok, err := registry.Authenticate(ctx, clientID, secret)
|
||||
if err != nil {
|
||||
t.Fatalf("authenticate (korrekt): %v", err)
|
||||
}
|
||||
if !ok || moduleName != name {
|
||||
t.Fatalf("erwartet erfolgreiche authentifizierung fuer %q, habe ok=%v moduleName=%q", name, ok, moduleName)
|
||||
}
|
||||
|
||||
_, ok, err = registry.Authenticate(ctx, clientID, "falsches-secret")
|
||||
if err != nil {
|
||||
t.Fatalf("authenticate (falsch): %v", err)
|
||||
}
|
||||
if ok {
|
||||
t.Fatal("erwartet fehlschlag bei falschem secret")
|
||||
}
|
||||
|
||||
_, ok, err = registry.Authenticate(ctx, "unbekannte-client-id", secret)
|
||||
if err != nil {
|
||||
t.Fatalf("authenticate (unbekannt): %v", err)
|
||||
}
|
||||
if ok {
|
||||
t.Fatal("erwartet fehlschlag bei unbekannter client-id")
|
||||
}
|
||||
}
|
||||
|
||||
func TestProvision_RequiresRegisteredModule(t *testing.T) {
|
||||
registry, _, cleanup := setupTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
if _, _, err := registry.Provision(ctx, "nie-registriert"); !errors.Is(err, ErrModuleNotRegistered) {
|
||||
t.Fatalf("erwartet ErrModuleNotRegistered, habe %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
func TestRequireServiceCredential_RejectsInvalidAcceptsValid(t *testing.T) {
|
||||
registry, _, cleanup := setupTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
name := uniqueModuleName(t)
|
||||
|
||||
if _, err := registry.Register(ctx, name, "1.0", nil); err != nil {
|
||||
t.Fatalf("register: %v", err)
|
||||
}
|
||||
clientID, secret, err := registry.Provision(ctx, name)
|
||||
if err != nil {
|
||||
t.Fatalf("provision: %v", err)
|
||||
}
|
||||
|
||||
handler := registry.RequireServiceCredential(func(w http.ResponseWriter, r *http.Request) {
|
||||
w.WriteHeader(http.StatusOK)
|
||||
})
|
||||
|
||||
// Fehlendes Credential.
|
||||
req := httptest.NewRequest(http.MethodPost, "/service-aufruf", nil)
|
||||
rec := httptest.NewRecorder()
|
||||
handler(rec, req)
|
||||
if rec.Code != http.StatusUnauthorized {
|
||||
t.Fatalf("ohne credential: status = %d, want 401", rec.Code)
|
||||
}
|
||||
|
||||
// Falsches Secret.
|
||||
req2 := httptest.NewRequest(http.MethodPost, "/service-aufruf", nil)
|
||||
req2.Header.Set("X-Client-Id", clientID)
|
||||
req2.Header.Set("X-Client-Secret", "falsch")
|
||||
rec2 := httptest.NewRecorder()
|
||||
handler(rec2, req2)
|
||||
if rec2.Code != http.StatusUnauthorized {
|
||||
t.Fatalf("falsches secret: status = %d, want 401", rec2.Code)
|
||||
}
|
||||
|
||||
// Gueltiges Credential.
|
||||
req3 := httptest.NewRequest(http.MethodPost, "/service-aufruf", nil)
|
||||
req3.Header.Set("X-Client-Id", clientID)
|
||||
req3.Header.Set("X-Client-Secret", secret)
|
||||
rec3 := httptest.NewRecorder()
|
||||
handler(rec3, req3)
|
||||
if rec3.Code != http.StatusOK {
|
||||
t.Fatalf("gueltiges credential: status = %d, want 200", rec3.Code)
|
||||
}
|
||||
}
|
||||
@@ -1,81 +0,0 @@
|
||||
// Package notify implementiert Core CFG-02: den zentralen Benachrichtigungs-
|
||||
// Dispatcher, ueber den beliebige Module Benachrichtigungen ausloesen —
|
||||
// Warteschlange, Wiederholungslogik, Kanal-Abstraktion. Die tatsaechlichen
|
||||
// Kanaele (E-Mail/In-App) sind CFG-03, hier gibt es nur die Sender-
|
||||
// Schnittstelle als Vorbereitung.
|
||||
package notify
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"time"
|
||||
|
||||
"github.com/jackc/pgx/v5/pgxpool"
|
||||
)
|
||||
|
||||
// DefaultMaxAttempts begrenzt Wiederholungsversuche (Akzeptanzkriterium 2) —
|
||||
// nach dieser Anzahl gibt der Dispatcher kontrolliert auf (status=failed)
|
||||
// statt endlos zu wiederholen.
|
||||
const DefaultMaxAttempts = 5
|
||||
|
||||
// DefaultRetryBackoff ist die Basis-Wartezeit zwischen Wiederholungen,
|
||||
// linear mit der Versuchsnummer skaliert.
|
||||
const DefaultRetryBackoff = 200 * time.Millisecond
|
||||
|
||||
type Notification struct {
|
||||
ID string
|
||||
Channel string
|
||||
Recipient string
|
||||
Payload map[string]any
|
||||
Attempts int
|
||||
}
|
||||
|
||||
// Sender ist die schmale Schnittstelle, die ein konkreter Kanal (CFG-03)
|
||||
// implementiert. Der Dispatcher selbst weiss nichts ueber E-Mail/In-App.
|
||||
type Sender interface {
|
||||
Send(ctx context.Context, n Notification) error
|
||||
}
|
||||
|
||||
// Dispatcher ist die EINE Schnittstelle, ueber die Module Benachrichtigungen
|
||||
// ausloesen — kein Modul baut eigenen Versandcode (Akzeptanzkriterium 1).
|
||||
type Dispatcher struct {
|
||||
pool *pgxpool.Pool
|
||||
maxAttempts int
|
||||
retryBackoff time.Duration
|
||||
}
|
||||
|
||||
func NewDispatcher(pool *pgxpool.Pool) *Dispatcher {
|
||||
return &Dispatcher{pool: pool, maxAttempts: DefaultMaxAttempts, retryBackoff: DefaultRetryBackoff}
|
||||
}
|
||||
|
||||
// WithRetryPolicy erlaubt Tests/Betrieb, Versuchsanzahl und Backoff
|
||||
// anzupassen, ohne die Default-Policy im Produktionscode zu veraendern.
|
||||
func (d *Dispatcher) WithRetryPolicy(maxAttempts int, backoff time.Duration) *Dispatcher {
|
||||
return &Dispatcher{pool: d.pool, maxAttempts: maxAttempts, retryBackoff: backoff}
|
||||
}
|
||||
|
||||
// Enqueue reiht eine Benachrichtigung in die Postgres-Warteschlange ein und
|
||||
// kehrt sofort zurueck — die Zeile ueberlebt jeden Neustart des Dispatcher-
|
||||
// Prozesses unveraendert (Akzeptanzkriterium 3), da sie ausschliesslich in
|
||||
// der Datenbank existiert, nicht im Prozessspeicher.
|
||||
func (d *Dispatcher) Enqueue(ctx context.Context, channel, recipient string, payload map[string]any) (string, error) {
|
||||
if payload == nil {
|
||||
payload = map[string]any{}
|
||||
}
|
||||
payloadJSON, err := json.Marshal(payload)
|
||||
if err != nil {
|
||||
return "", fmt.Errorf("payload serialisieren: %w", err)
|
||||
}
|
||||
|
||||
var id string
|
||||
err = d.pool.QueryRow(ctx, `
|
||||
INSERT INTO notification_jobs (channel, recipient, payload, max_attempts)
|
||||
VALUES ($1, $2, $3, $4)
|
||||
RETURNING id
|
||||
`, channel, recipient, payloadJSON, d.maxAttempts).Scan(&id)
|
||||
if err != nil {
|
||||
return "", fmt.Errorf("benachrichtigung einreihen: %w", err)
|
||||
}
|
||||
return id, nil
|
||||
}
|
||||
@@ -1,219 +0,0 @@
|
||||
package notify
|
||||
|
||||
import (
|
||||
"context"
|
||||
"errors"
|
||||
"os"
|
||||
"sync"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
"github.com/jackc/pgx/v5/pgxpool"
|
||||
)
|
||||
|
||||
func setupTest(t *testing.T) (*pgxpool.Pool, func()) {
|
||||
t.Helper()
|
||||
adminDSN := os.Getenv("TEST_ADMIN_DSN")
|
||||
if adminDSN == "" {
|
||||
t.Skip("TEST_ADMIN_DSN nicht gesetzt, Integrationstest uebersprungen")
|
||||
}
|
||||
ctx := context.Background()
|
||||
|
||||
pool, err := pgxpool.New(ctx, adminDSN)
|
||||
if err != nil {
|
||||
t.Fatalf("pool: %v", err)
|
||||
}
|
||||
if _, err := pool.Exec(ctx, `
|
||||
CREATE TABLE IF NOT EXISTS notification_jobs (
|
||||
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
|
||||
channel TEXT NOT NULL,
|
||||
recipient TEXT NOT NULL,
|
||||
payload JSONB NOT NULL DEFAULT '{}'::jsonb,
|
||||
status TEXT NOT NULL DEFAULT 'pending' CHECK (status IN ('pending', 'sent', 'failed')),
|
||||
attempts INT NOT NULL DEFAULT 0,
|
||||
max_attempts INT NOT NULL DEFAULT 5,
|
||||
next_attempt_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
||||
last_error TEXT,
|
||||
created_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
||||
updated_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
)`); err != nil {
|
||||
t.Fatalf("schema: %v", err)
|
||||
}
|
||||
|
||||
cleanup := func() { pool.Close() }
|
||||
return pool, cleanup
|
||||
}
|
||||
|
||||
type fakeSender struct {
|
||||
mu sync.Mutex
|
||||
sentIDs []string
|
||||
failUntil int
|
||||
calls int
|
||||
}
|
||||
|
||||
func (f *fakeSender) Send(ctx context.Context, n Notification) error {
|
||||
f.mu.Lock()
|
||||
defer f.mu.Unlock()
|
||||
f.calls++
|
||||
if f.calls <= f.failUntil {
|
||||
return errors.New("simulierter zustellfehler")
|
||||
}
|
||||
f.sentIDs = append(f.sentIDs, n.ID)
|
||||
return nil
|
||||
}
|
||||
|
||||
func (f *fakeSender) sentCount() int {
|
||||
f.mu.Lock()
|
||||
defer f.mu.Unlock()
|
||||
return len(f.sentIDs)
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 1: Module loesen ueber Enqueue aus, keine eigene
|
||||
// Versandlogik noetig.
|
||||
func TestDispatcher_EnqueueAndProcess(t *testing.T) {
|
||||
pool, cleanup := setupTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
d := NewDispatcher(pool)
|
||||
id, err := d.Enqueue(ctx, "email", "alice@example.com", map[string]any{"subject": "Willkommen"})
|
||||
if err != nil {
|
||||
t.Fatalf("enqueue: %v", err)
|
||||
}
|
||||
if id == "" {
|
||||
t.Fatal("erwartet nicht-leere id")
|
||||
}
|
||||
|
||||
sender := &fakeSender{}
|
||||
sent, failed, err := d.ProcessDue(ctx, sender, 10)
|
||||
if err != nil {
|
||||
t.Fatalf("process: %v", err)
|
||||
}
|
||||
if sent != 1 || failed != 0 {
|
||||
t.Fatalf("erwartet sent=1 failed=0, habe sent=%d failed=%d", sent, failed)
|
||||
}
|
||||
if sender.sentCount() != 1 {
|
||||
t.Fatalf("erwartet 1 zustellung, habe %d", sender.sentCount())
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 2 + Pruefung 2: Wiederholungslogik greift bei
|
||||
// simuliertem Fehler und bricht nach definierter Anzahl kontrolliert ab.
|
||||
func TestProcessDue_RetriesThenGivesUpAfterMaxAttempts(t *testing.T) {
|
||||
pool, cleanup := setupTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
d := NewDispatcher(pool).WithRetryPolicy(3, time.Millisecond)
|
||||
id, err := d.Enqueue(ctx, "email", "bob@example.com", nil)
|
||||
if err != nil {
|
||||
t.Fatalf("enqueue: %v", err)
|
||||
}
|
||||
|
||||
sender := &fakeSender{failUntil: 100} // schlaegt bei jedem versuch fehl
|
||||
|
||||
for i := 0; i < 3; i++ {
|
||||
time.Sleep(5 * time.Millisecond) // next_attempt_at abwarten
|
||||
if _, _, err := d.ProcessDue(ctx, sender, 10); err != nil {
|
||||
t.Fatalf("process %d: %v", i, err)
|
||||
}
|
||||
}
|
||||
|
||||
var status string
|
||||
var attempts int
|
||||
if err := pool.QueryRow(ctx, `SELECT status, attempts FROM notification_jobs WHERE id = $1`, id).Scan(&status, &attempts); err != nil {
|
||||
t.Fatalf("status lesen: %v", err)
|
||||
}
|
||||
if status != "failed" {
|
||||
t.Fatalf("erwartet status failed nach max_attempts, habe %q", status)
|
||||
}
|
||||
if attempts != 3 {
|
||||
t.Fatalf("erwartet 3 versuche, habe %d", attempts)
|
||||
}
|
||||
|
||||
// Weiteres ProcessDue darf den bereits aufgegebenen job nicht mehr anfassen.
|
||||
sent, failed, err := d.ProcessDue(ctx, sender, 10)
|
||||
if err != nil {
|
||||
t.Fatalf("process nach abbruch: %v", err)
|
||||
}
|
||||
if sent != 0 || failed != 0 {
|
||||
t.Fatalf("erwartet keine weitere verarbeitung, habe sent=%d failed=%d", sent, failed)
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 3 + Pruefung 1: Neustart des Dienstes waehrend offener
|
||||
// Zustellung verliert keine Nachricht — simuliert durch eine komplett neue
|
||||
// Dispatcher/Pool-Instanz nach dem Enqueue, bevor irgendetwas verarbeitet wurde.
|
||||
func TestQueue_SurvivesRestartWithoutMessageLoss(t *testing.T) {
|
||||
pool, cleanup := setupTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
firstInstance := NewDispatcher(pool)
|
||||
id, err := firstInstance.Enqueue(ctx, "email", "carol@example.com", nil)
|
||||
if err != nil {
|
||||
t.Fatalf("enqueue: %v", err)
|
||||
}
|
||||
|
||||
// "Neustart": eine voellig neue Dispatcher-Instanz (repraesentiert einen
|
||||
// neuen Prozess) verbindet sich neu und verarbeitet die Warteschlange —
|
||||
// die Nachricht existiert ausschliesslich in Postgres, nicht im
|
||||
// Prozessspeicher der ersten Instanz.
|
||||
restartedInstance := NewDispatcher(pool)
|
||||
sender := &fakeSender{}
|
||||
sent, failed, err := restartedInstance.ProcessDue(ctx, sender, 10)
|
||||
if err != nil {
|
||||
t.Fatalf("process nach neustart: %v", err)
|
||||
}
|
||||
if sent != 1 || failed != 0 {
|
||||
t.Fatalf("erwartet sent=1 nach neustart, habe sent=%d failed=%d", sent, failed)
|
||||
}
|
||||
if len(sender.sentIDs) != 1 || sender.sentIDs[0] != id {
|
||||
t.Fatalf("erwartet zustellung der urspruenglichen nachricht %q, habe %v", id, sender.sentIDs)
|
||||
}
|
||||
}
|
||||
|
||||
// Akzeptanzkriterium 3 + Pruefung 3: zwei gleichzeitig ausloesende Module,
|
||||
// beide Nachrichten werden korrekt (und nicht doppelt) zugestellt.
|
||||
func TestProcessDue_ConcurrentDispatchBothDelivered(t *testing.T) {
|
||||
pool, cleanup := setupTest(t)
|
||||
defer cleanup()
|
||||
ctx := context.Background()
|
||||
|
||||
d := NewDispatcher(pool)
|
||||
idA, err := d.Enqueue(ctx, "email", "modul-a@example.com", nil)
|
||||
if err != nil {
|
||||
t.Fatalf("enqueue a: %v", err)
|
||||
}
|
||||
idB, err := d.Enqueue(ctx, "email", "modul-b@example.com", nil)
|
||||
if err != nil {
|
||||
t.Fatalf("enqueue b: %v", err)
|
||||
}
|
||||
|
||||
sender := &fakeSender{}
|
||||
var wg sync.WaitGroup
|
||||
for i := 0; i < 2; i++ {
|
||||
wg.Add(1)
|
||||
go func() {
|
||||
defer wg.Done()
|
||||
if _, _, err := d.ProcessDue(ctx, sender, 10); err != nil {
|
||||
t.Errorf("process: %v", err)
|
||||
}
|
||||
}()
|
||||
}
|
||||
wg.Wait()
|
||||
|
||||
if sender.sentCount() != 2 {
|
||||
t.Fatalf("erwartet genau 2 zustellungen, habe %d: %v", sender.sentCount(), sender.sentIDs)
|
||||
}
|
||||
seen := map[string]bool{}
|
||||
for _, id := range sender.sentIDs {
|
||||
if seen[id] {
|
||||
t.Fatalf("nachricht %q wurde doppelt zugestellt", id)
|
||||
}
|
||||
seen[id] = true
|
||||
}
|
||||
if !seen[idA] || !seen[idB] {
|
||||
t.Fatalf("erwartet beide nachrichten zugestellt, habe %v", sender.sentIDs)
|
||||
}
|
||||
}
|
||||
@@ -1,102 +0,0 @@
|
||||
package notify
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"time"
|
||||
)
|
||||
|
||||
// ProcessDue holt bis zu limit faellige Benachrichtigungen und versucht sie
|
||||
// ueber sender zuzustellen. FOR UPDATE SKIP LOCKED serialisiert konkurrierende
|
||||
// Aufrufe (Akzeptanzkriterium 3 / Pruefung 3: zwei gleichzeitig ausloesende
|
||||
// Module duerfen sich nicht gegenseitig blockieren oder Nachrichten doppelt
|
||||
// zustellen) — dieselbe Konvention wie internal/tenant.Lifecycle.ProcessDueDeletions.
|
||||
func (d *Dispatcher) ProcessDue(ctx context.Context, sender Sender, limit int) (sent, failed int, err error) {
|
||||
tx, err := d.pool.Begin(ctx)
|
||||
if err != nil {
|
||||
return 0, 0, fmt.Errorf("transaktion starten: %w", err)
|
||||
}
|
||||
defer func() { _ = tx.Rollback(ctx) }()
|
||||
|
||||
rows, err := tx.Query(ctx, `
|
||||
SELECT id, channel, recipient, payload, attempts, max_attempts
|
||||
FROM notification_jobs
|
||||
WHERE status = 'pending' AND next_attempt_at <= now()
|
||||
ORDER BY created_at
|
||||
FOR UPDATE SKIP LOCKED
|
||||
LIMIT $1
|
||||
`, limit)
|
||||
if err != nil {
|
||||
return 0, 0, fmt.Errorf("faellige benachrichtigungen abfragen: %w", err)
|
||||
}
|
||||
|
||||
type due struct {
|
||||
id, channel, recipient string
|
||||
payload []byte
|
||||
attempts, maxAttempts int
|
||||
}
|
||||
var candidates []due
|
||||
for rows.Next() {
|
||||
var c due
|
||||
if err := rows.Scan(&c.id, &c.channel, &c.recipient, &c.payload, &c.attempts, &c.maxAttempts); err != nil {
|
||||
rows.Close()
|
||||
return 0, 0, fmt.Errorf("faellige benachrichtigung lesen: %w", err)
|
||||
}
|
||||
candidates = append(candidates, c)
|
||||
}
|
||||
rows.Close()
|
||||
if err := rows.Err(); err != nil {
|
||||
return 0, 0, err
|
||||
}
|
||||
|
||||
for _, c := range candidates {
|
||||
var payload map[string]any
|
||||
if err := json.Unmarshal(c.payload, &payload); err != nil {
|
||||
payload = map[string]any{}
|
||||
}
|
||||
|
||||
sendErr := sender.Send(ctx, Notification{
|
||||
ID: c.id, Channel: c.channel, Recipient: c.recipient, Payload: payload, Attempts: c.attempts,
|
||||
})
|
||||
|
||||
if sendErr == nil {
|
||||
if _, err := tx.Exec(ctx, `
|
||||
UPDATE notification_jobs SET status = 'sent', updated_at = now() WHERE id = $1
|
||||
`, c.id); err != nil {
|
||||
return sent, failed, fmt.Errorf("erfolg speichern: %w", err)
|
||||
}
|
||||
sent++
|
||||
continue
|
||||
}
|
||||
|
||||
newAttempts := c.attempts + 1
|
||||
if newAttempts >= c.maxAttempts {
|
||||
// Akzeptanzkriterium 2: kontrollierter Abbruch nach definierter
|
||||
// Anzahl Versuche, kein endloses Wiederholen.
|
||||
if _, err := tx.Exec(ctx, `
|
||||
UPDATE notification_jobs
|
||||
SET status = 'failed', attempts = $2, last_error = $3, updated_at = now()
|
||||
WHERE id = $1
|
||||
`, c.id, newAttempts, sendErr.Error()); err != nil {
|
||||
return sent, failed, fmt.Errorf("fehlschlag speichern: %w", err)
|
||||
}
|
||||
failed++
|
||||
continue
|
||||
}
|
||||
|
||||
nextAttempt := time.Now().Add(time.Duration(newAttempts) * d.retryBackoff)
|
||||
if _, err := tx.Exec(ctx, `
|
||||
UPDATE notification_jobs
|
||||
SET attempts = $2, next_attempt_at = $3, last_error = $4, updated_at = now()
|
||||
WHERE id = $1
|
||||
`, c.id, newAttempts, nextAttempt, sendErr.Error()); err != nil {
|
||||
return sent, failed, fmt.Errorf("wiederholung planen: %w", err)
|
||||
}
|
||||
}
|
||||
|
||||
if err := tx.Commit(ctx); err != nil {
|
||||
return 0, 0, fmt.Errorf("transaktion committen: %w", err)
|
||||
}
|
||||
return sent, failed, nil
|
||||
}
|
||||
@@ -1,2 +0,0 @@
|
||||
DROP TABLE IF EXISTS config_value_history;
|
||||
DROP TABLE IF EXISTS config_values;
|
||||
@@ -1,23 +0,0 @@
|
||||
-- Zentraler Konfigurationsdienst (CFG-01, siehe core-kanban/tickets/CFG-01.md).
|
||||
-- scope = 'global' fuer globale Defaults, sonst der Tenant-Slug. config_values
|
||||
-- haelt den AKTUELLEN Stand je (key, scope); config_value_history haelt JEDE
|
||||
-- Aenderung fest (Akzeptanzkriterium 2: versioniert nachvollziehbar).
|
||||
CREATE TABLE config_values (
|
||||
key TEXT NOT NULL,
|
||||
scope TEXT NOT NULL CHECK (scope <> ''),
|
||||
value TEXT NOT NULL,
|
||||
version INT NOT NULL,
|
||||
updated_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
||||
PRIMARY KEY (key, scope)
|
||||
);
|
||||
|
||||
CREATE TABLE config_value_history (
|
||||
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
|
||||
key TEXT NOT NULL,
|
||||
scope TEXT NOT NULL,
|
||||
value TEXT NOT NULL,
|
||||
version INT NOT NULL,
|
||||
changed_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
);
|
||||
|
||||
CREATE INDEX config_value_history_key_scope_idx ON config_value_history (key, scope, version);
|
||||
@@ -0,0 +1 @@
|
||||
DROP TABLE IF EXISTS feature_flags;
|
||||
@@ -0,0 +1,10 @@
|
||||
-- Feature-Flags zentral je Mandant/Zielgruppe (LIC-02, siehe core-kanban/tickets/LIC-02.md).
|
||||
-- Lebt in der Registry-DB, nicht pro Tenant-Datenbank — Flags sind eine
|
||||
-- Core-weite Konfiguration, keine Mandanten-Geschaeftsdaten.
|
||||
CREATE TABLE feature_flags (
|
||||
key TEXT PRIMARY KEY,
|
||||
enabled BOOLEAN NOT NULL DEFAULT false,
|
||||
rollout_percentage INT NOT NULL DEFAULT 0 CHECK (rollout_percentage BETWEEN 0 AND 100),
|
||||
target_tenant_slugs TEXT[] NOT NULL DEFAULT '{}',
|
||||
updated_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
);
|
||||
@@ -0,0 +1,2 @@
|
||||
DROP TABLE IF EXISTS module_credentials;
|
||||
DROP TABLE IF EXISTS modules;
|
||||
@@ -0,0 +1,18 @@
|
||||
-- Modul-Registry & Aktivierungspruefung (API-02, siehe core-kanban/tickets/API-02.md).
|
||||
CREATE TABLE modules (
|
||||
name TEXT PRIMARY KEY,
|
||||
version TEXT NOT NULL CHECK (version <> ''),
|
||||
required_flags TEXT[] NOT NULL DEFAULT '{}',
|
||||
registered_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
);
|
||||
|
||||
-- Service-Credential je Modul-Instanz, bei Provisionierung ausgestellt
|
||||
-- (Akzeptanzkriterium 4). secret_hash enthaelt NIEMALS das Secret im
|
||||
-- Klartext, nur dessen SHA-256-Hash (Timing-safe-Vergleich beim Login,
|
||||
-- Referenzmuster siehe AUD-02).
|
||||
CREATE TABLE module_credentials (
|
||||
module_name TEXT PRIMARY KEY REFERENCES modules(name),
|
||||
client_id TEXT NOT NULL UNIQUE,
|
||||
secret_hash BYTEA NOT NULL,
|
||||
issued_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
);
|
||||
@@ -1 +0,0 @@
|
||||
DROP TABLE IF EXISTS notification_jobs;
|
||||
@@ -1,20 +0,0 @@
|
||||
-- Benachrichtigungs-Dispatcher-Warteschlange (CFG-02, siehe
|
||||
-- core-kanban/tickets/CFG-02.md). Postgres-basiert statt Redis/AMQP
|
||||
-- (Projekt-Konvention, siehe nexarch-state.json techstack.job_queue) —
|
||||
-- Zeilen ueberleben einen Neustart des Dispatcher-Prozesses unveraendert
|
||||
-- (Akzeptanzkriterium 3).
|
||||
CREATE TABLE notification_jobs (
|
||||
id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
|
||||
channel TEXT NOT NULL,
|
||||
recipient TEXT NOT NULL,
|
||||
payload JSONB NOT NULL DEFAULT '{}'::jsonb,
|
||||
status TEXT NOT NULL DEFAULT 'pending' CHECK (status IN ('pending', 'sent', 'failed')),
|
||||
attempts INT NOT NULL DEFAULT 0,
|
||||
max_attempts INT NOT NULL DEFAULT 5,
|
||||
next_attempt_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
||||
last_error TEXT,
|
||||
created_at TIMESTAMPTZ NOT NULL DEFAULT now(),
|
||||
updated_at TIMESTAMPTZ NOT NULL DEFAULT now()
|
||||
);
|
||||
|
||||
CREATE INDEX notification_jobs_due_idx ON notification_jobs (status, next_attempt_at);
|
||||
@@ -1,26 +1,11 @@
|
||||
#!/usr/bin/env bash
|
||||
# Setzt die nexarch-Testumgebung zurueck: loescht die geteilte
|
||||
# Registry-Tabelle "tenants" in der postgres-Wartungsdatenbank sowie alle
|
||||
# tenant_*-Datenbanken. Noetig, weil verschiedene Feature-Branches
|
||||
# unterschiedliche Registry-Schemata erwarten, aber dieselbe physische
|
||||
# Postgres-Instanz auf dem Testhost teilen (siehe [[project-nexarch-test-infra]]).
|
||||
#
|
||||
# Aufruf: NEXARCH_TEST_DB_PASSWORD=... ./scripts/reset-test-env.sh
|
||||
set -euo pipefail
|
||||
|
||||
PASS="${NEXARCH_TEST_DB_PASSWORD:?Setze NEXARCH_TEST_DB_PASSWORD vor dem Aufruf}"
|
||||
ROLE="nexarch_test"
|
||||
|
||||
export PGPASSWORD="$PASS"
|
||||
|
||||
psql -h localhost -U "$ROLE" -d postgres -v ON_ERROR_STOP=1 -c "DROP TABLE IF EXISTS tenants CASCADE;"
|
||||
psql -h localhost -U "$ROLE" -d postgres -v ON_ERROR_STOP=1 -c "DROP TABLE IF EXISTS audit_events CASCADE;"
|
||||
psql -h localhost -U "$ROLE" -d postgres -v ON_ERROR_STOP=1 -c "DROP TABLE IF EXISTS config_value_history CASCADE;"
|
||||
psql -h localhost -U "$ROLE" -d postgres -v ON_ERROR_STOP=1 -c "DROP TABLE IF EXISTS config_values CASCADE;"
|
||||
|
||||
dbs=$(psql -h localhost -U "$ROLE" -d postgres -tAc "SELECT datname FROM pg_database WHERE datname LIKE 'tenant\_%' ESCAPE '\'")
|
||||
for db in $dbs; do
|
||||
psql -h localhost -U "$ROLE" -d postgres -v ON_ERROR_STOP=1 -c "DROP DATABASE IF EXISTS \"${db}\";"
|
||||
done
|
||||
|
||||
echo "Testumgebung zurueckgesetzt: registry-tabelle + $(echo "$dbs" | grep -c . || true) tenant-datenbank(en) entfernt."
|
||||
|
||||
@@ -1,24 +1,12 @@
|
||||
#!/usr/bin/env bash
|
||||
# Ein-Kommando-Pruefung fuer den aktuellen Code-Stand auf dem Testhost:
|
||||
# Registry+Tenant-DBs zuruecksetzen, dann build/vet/test in einem Rutsch.
|
||||
# -p 1 ist Pflicht, da mehrere Pakete dieselbe physische Registry-Tabelle auf
|
||||
# dem Testhost teilen (siehe [[project-nexarch-test-infra]]).
|
||||
#
|
||||
# Aufruf: NEXARCH_TEST_DB_PASSWORD=... ./scripts/run-checks.sh
|
||||
set -euo pipefail
|
||||
|
||||
PASS="${NEXARCH_TEST_DB_PASSWORD:?Setze NEXARCH_TEST_DB_PASSWORD vor dem Aufruf}"
|
||||
cd "$(dirname "$0")/.."
|
||||
|
||||
NEXARCH_TEST_DB_PASSWORD="$PASS" bash scripts/reset-test-env.sh
|
||||
|
||||
export TEST_ADMIN_DSN="postgresql://nexarch_test:${PASS}@localhost:5432/postgres?sslmode=disable"
|
||||
|
||||
echo "== go build =="
|
||||
go build ./...
|
||||
|
||||
echo "== go vet =="
|
||||
go vet ./...
|
||||
|
||||
echo "== go test (-p 1) =="
|
||||
go test ./... -p 1 -count=1
|
||||
|
||||
Reference in New Issue
Block a user