fix(auth): 为登录链路增加有界超时
ci / verify (pull_request) Successful in 11m22s

为 PostgreSQL 连接、就绪检查和本地登录设置分层超时,数据库不可用时返回稳定 503 错误码并记录无凭据的连接池统计。

前端登录在 10 秒后取消请求并兼容调用方 AbortSignal,Nginx 登录精确路由限制上游为 15 秒,同时更新 OpenAPI 和回归测试。
This commit is contained in:
2026-07-21 11:47:09 +08:00
parent 86c374b5c2
commit bfa17a3aba
9 changed files with 322 additions and 19 deletions
+6
View File
@@ -5208,6 +5208,12 @@
"schema": {
"$ref": "#/definitions/httpapi.ErrorEnvelope"
}
},
"503": {
"description": "Service Unavailable",
"schema": {
"$ref": "#/definitions/httpapi.ErrorEnvelope"
}
}
}
}
+4
View File
@@ -6441,6 +6441,10 @@ paths:
description: Internal Server Error
schema:
$ref: '#/definitions/httpapi.ErrorEnvelope'
"503":
description: Service Unavailable
schema:
$ref: '#/definitions/httpapi.ErrorEnvelope'
summary: 本地登录
tags:
- auth
@@ -0,0 +1,111 @@
package httpapi
import (
"bytes"
"context"
"encoding/json"
"io"
"log/slog"
"net/http"
"net/http/httptest"
"net/url"
"os"
"strings"
"testing"
"time"
"github.com/easyai/easyai-ai-gateway/apps/api/internal/store"
)
func TestReadyReturnsPostgresUnavailableWithinTwoSeconds(t *testing.T) {
db := newExhaustedPostgresStore(t)
server := &Server{store: db, logger: slog.New(slog.NewJSONHandler(io.Discard, nil))}
requestContext, cancel := context.WithTimeout(context.Background(), 4*time.Second)
defer cancel()
request := httptest.NewRequest(http.MethodGet, "/readyz", nil).WithContext(requestContext)
recorder := httptest.NewRecorder()
startedAt := time.Now()
server.ready(recorder, request)
elapsed := time.Since(startedAt)
assertUnavailableResponse(t, recorder, "POSTGRES_UNAVAILABLE", "postgres unavailable")
if elapsed > 3*time.Second {
t.Fatalf("readiness timeout took %s, want no more than 3s", elapsed)
}
}
func TestLoginReturnsAuthStoreUnavailableWithinFiveSeconds(t *testing.T) {
db := newExhaustedPostgresStore(t)
var logs bytes.Buffer
server := &Server{store: db, logger: slog.New(slog.NewJSONHandler(&logs, nil))}
requestContext, cancel := context.WithTimeout(context.Background(), 7*time.Second)
defer cancel()
request := httptest.NewRequest(http.MethodPost, "/api/v1/auth/login", strings.NewReader(`{"account":"timeout-test-account","password":"timeout-test-password"}`)).WithContext(requestContext)
recorder := httptest.NewRecorder()
startedAt := time.Now()
server.login(recorder, request)
elapsed := time.Since(startedAt)
assertUnavailableResponse(t, recorder, "AUTH_STORE_UNAVAILABLE", "authentication service temporarily unavailable")
if elapsed > 6*time.Second {
t.Fatalf("login timeout took %s, want no more than 6s", elapsed)
}
logOutput := logs.String()
for _, field := range []string{"postgres_pool_max_connections", "postgres_pool_acquired_connections", "postgres_pool_idle_connections", "postgres_pool_empty_acquire_count", "postgres_pool_canceled_acquire_count"} {
if !strings.Contains(logOutput, field) {
t.Fatalf("login failure log did not include %q: %s", field, logOutput)
}
}
if strings.Contains(logOutput, "timeout-test-account") || strings.Contains(logOutput, "timeout-test-password") {
t.Fatalf("login failure log exposed credentials: %s", logOutput)
}
}
func newExhaustedPostgresStore(t *testing.T) *store.Store {
t.Helper()
databaseURL := strings.TrimSpace(os.Getenv("AI_GATEWAY_TEST_DATABASE_URL"))
if databaseURL == "" {
t.Skip("set AI_GATEWAY_TEST_DATABASE_URL to run PostgreSQL availability timeout tests")
}
parsed, err := url.Parse(databaseURL)
if err != nil {
t.Fatalf("parse test database URL: %v", err)
}
query := parsed.Query()
query.Set("pool_max_conns", "1")
query.Set("pool_min_conns", "0")
parsed.RawQuery = query.Encode()
ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
defer cancel()
db, err := store.Connect(ctx, parsed.String())
if err != nil {
t.Fatalf("connect timeout test store: %v", err)
}
t.Cleanup(db.Close)
connection, err := db.Pool().Acquire(ctx)
if err != nil {
t.Fatalf("exhaust timeout test pool: %v", err)
}
t.Cleanup(connection.Release)
return db
}
func assertUnavailableResponse(t *testing.T, recorder *httptest.ResponseRecorder, expectedCode, expectedMessage string) {
t.Helper()
if recorder.Code != http.StatusServiceUnavailable {
t.Fatalf("status = %d, want 503; body=%s", recorder.Code, recorder.Body.String())
}
var envelope ErrorEnvelope
if err := json.Unmarshal(recorder.Body.Bytes(), &envelope); err != nil {
t.Fatalf("decode unavailable response: %v", err)
}
if envelope.Error.Code != expectedCode {
t.Fatalf("error code = %q, want %q; body=%s", envelope.Error.Code, expectedCode, recorder.Body.String())
}
if envelope.Error.Message != expectedMessage {
t.Fatalf("error message = %q, want %q; body=%s", envelope.Error.Message, expectedMessage, recorder.Body.String())
}
}
+38 -3
View File
@@ -17,6 +17,14 @@ import (
"github.com/easyai/easyai-ai-gateway/apps/api/internal/store"
)
const (
postgresReadinessTimeout = 2 * time.Second
localLoginStoreTimeout = 5 * time.Second
errorCodePostgresDown = "POSTGRES_UNAVAILABLE"
errorCodeAuthStoreDown = "AUTH_STORE_UNAVAILABLE"
authStoreUnavailableError = "authentication service temporarily unavailable"
)
// health godoc
// @Summary 健康检查
// @Description 返回服务进程、运行环境和身份模式,供负载均衡或人工排障使用。
@@ -42,8 +50,11 @@ func (s *Server) health(w http.ResponseWriter, r *http.Request) {
// @Failure 503 {object} ErrorEnvelope
// @Router /readyz [get]
func (s *Server) ready(w http.ResponseWriter, r *http.Request) {
if err := s.store.Ping(r.Context()); err != nil {
writeError(w, http.StatusServiceUnavailable, "postgres unavailable")
ctx, cancel := context.WithTimeout(r.Context(), postgresReadinessTimeout)
defer cancel()
if err := s.store.Ping(ctx); err != nil {
s.logPostgresUnavailable("postgres readiness check failed")
writeError(w, http.StatusServiceUnavailable, "postgres unavailable", errorCodePostgresDown)
return
}
writeJSON(w, http.StatusOK, map[string]any{"ok": true})
@@ -121,6 +132,7 @@ func (s *Server) register(w http.ResponseWriter, r *http.Request) {
// @Failure 401 {object} ErrorEnvelope
// @Failure 403 {object} ErrorEnvelope
// @Failure 500 {object} ErrorEnvelope
// @Failure 503 {object} ErrorEnvelope
// @Router /api/v1/auth/login [post]
func (s *Server) login(w http.ResponseWriter, r *http.Request) {
var input store.LocalLoginInput
@@ -128,12 +140,19 @@ func (s *Server) login(w http.ResponseWriter, r *http.Request) {
writeError(w, http.StatusBadRequest, "invalid json body")
return
}
user, err := s.store.AuthenticateLocalUser(r.Context(), input)
ctx, cancel := context.WithTimeout(r.Context(), localLoginStoreTimeout)
defer cancel()
user, err := s.store.AuthenticateLocalUser(ctx, input)
if err != nil {
if errors.Is(err, store.ErrInvalidCredentials) {
writeError(w, http.StatusUnauthorized, "invalid account or password")
return
}
if store.IsPostgresUnavailable(err) {
s.logPostgresUnavailable("login authentication store unavailable")
writeError(w, http.StatusServiceUnavailable, authStoreUnavailableError, errorCodeAuthStoreDown)
return
}
s.logger.Error("login local user failed", "error", err)
writeError(w, http.StatusInternalServerError, "login failed")
return
@@ -145,6 +164,22 @@ func (s *Server) login(w http.ResponseWriter, r *http.Request) {
s.writeAuthResponse(w, http.StatusOK, user)
}
func (s *Server) logPostgresUnavailable(message string) {
if s.logger == nil || s.store == nil || s.store.Pool() == nil {
return
}
statistics := s.store.Pool().Stat()
s.logger.Error(message,
"error_category", "postgres_unavailable",
"postgres_pool_max_connections", statistics.MaxConns(),
"postgres_pool_total_connections", statistics.TotalConns(),
"postgres_pool_acquired_connections", statistics.AcquiredConns(),
"postgres_pool_idle_connections", statistics.IdleConns(),
"postgres_pool_empty_acquire_count", statistics.EmptyAcquireCount(),
"postgres_pool_canceled_acquire_count", statistics.CanceledAcquireCount(),
)
}
func (s *Server) localIdentityEnabled() bool {
mode := strings.ToLower(strings.TrimSpace(s.cfg.IdentityMode))
return mode == "" || mode == "standalone" || mode == "hybrid"
+43 -1
View File
@@ -7,6 +7,7 @@ import (
"encoding/base64"
"encoding/json"
"errors"
"net"
"strings"
"time"
"unicode"
@@ -22,6 +23,11 @@ type Store struct {
pool *pgxpool.Pool
}
const (
postgresApplicationName = "easyai-ai-gateway"
postgresConnectTimeout = 5 * time.Second
)
func defaultAPIKeyScopes() []string {
return []string{"chat", "embedding", "rerank", "image", "video", "music", "audio", "voice_clone"}
}
@@ -67,7 +73,11 @@ var (
)
func Connect(ctx context.Context, databaseURL string) (*Store, error) {
pool, err := pgxpool.New(ctx, databaseURL)
config, err := postgresPoolConfig(databaseURL)
if err != nil {
return nil, err
}
pool, err := pgxpool.NewWithConfig(ctx, config)
if err != nil {
return nil, err
}
@@ -78,6 +88,38 @@ func Connect(ctx context.Context, databaseURL string) (*Store, error) {
return &Store{pool: pool}, nil
}
func postgresPoolConfig(databaseURL string) (*pgxpool.Config, error) {
config, err := pgxpool.ParseConfig(databaseURL)
if err != nil {
return nil, err
}
config.ConnConfig.ConnectTimeout = postgresConnectTimeout
config.ConnConfig.RuntimeParams["application_name"] = postgresApplicationName
return config, nil
}
func IsPostgresUnavailable(err error) bool {
if err == nil {
return false
}
if errors.Is(err, context.DeadlineExceeded) {
return true
}
var connectError *pgconn.ConnectError
if errors.As(err, &connectError) {
return true
}
var networkError net.Error
if errors.As(err, &networkError) {
return true
}
var postgresError *pgconn.PgError
if errors.As(err, &postgresError) {
return strings.HasPrefix(postgresError.Code, "08") || postgresError.Code == "53300" || strings.HasPrefix(postgresError.Code, "57P0")
}
return pgconn.SafeToRetry(err)
}
func (s *Store) Close() {
s.pool.Close()
}
@@ -0,0 +1,49 @@
package store
import (
"context"
"testing"
"time"
"github.com/jackc/pgx/v5/pgconn"
)
func TestPostgresPoolConfigSetsDiagnosticAndConnectTimeout(t *testing.T) {
config, err := postgresPoolConfig("postgresql://gateway:password@localhost:5432/gateway?sslmode=disable")
if err != nil {
t.Fatalf("parse PostgreSQL pool config: %v", err)
}
if config.ConnConfig.ConnectTimeout != 5*time.Second {
t.Fatalf("connect timeout = %s, want 5s", config.ConnConfig.ConnectTimeout)
}
if applicationName := config.ConnConfig.RuntimeParams["application_name"]; applicationName != "easyai-ai-gateway" {
t.Fatalf("application_name = %q, want easyai-ai-gateway", applicationName)
}
}
func TestPostgresPoolConfigRejectsMalformedURL(t *testing.T) {
if _, err := postgresPoolConfig("://malformed"); err == nil {
t.Fatal("expected malformed PostgreSQL URL to fail")
}
}
func TestIsPostgresUnavailableClassifiesConnectivityFailures(t *testing.T) {
for _, testCase := range []struct {
name string
err error
}{
{name: "deadline", err: context.DeadlineExceeded},
{name: "connection exception", err: &pgconn.PgError{Code: "08006"}},
{name: "too many connections", err: &pgconn.PgError{Code: "53300"}},
{name: "cannot connect now", err: &pgconn.PgError{Code: "57P03"}},
} {
t.Run(testCase.name, func(t *testing.T) {
if !IsPostgresUnavailable(testCase.err) {
t.Fatalf("error %v was not classified as PostgreSQL unavailable", testCase.err)
}
})
}
if IsPostgresUnavailable(&pgconn.PgError{Code: "42601"}) {
t.Fatal("SQL syntax error was incorrectly classified as PostgreSQL unavailable")
}
}