fix(database): 数据库连接 - 消费 libpq keepalive 参数并默认限制闲置事务 (#737)

pgx 不识别 DSN 中 libpq 的 keepalives* 参数,会作为服务端运行参数发送导致连接被拒;改为在建连前消费并应用到 TCP 拨号器。
未显式配置时为每条连接默认设置 idle_in_transaction_session_timeout=5min,避免卡住的事务长期持锁;DSN 显式配置时以 DSN 为准。
建连钩子先复制共享的 RuntimeParams,避免多条连接之间累积修改。

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Catlaxy AI
2026-10-11 22:04:27 +08:00
committed by GitHub
parent 28f2362e3e
commit 6dedc5973c
5 changed files with 334 additions and 2 deletions

View File

@@ -7,6 +7,7 @@ require (
github.com/gabriel-vasile/mimetype v1.4.13
github.com/gin-gonic/gin v1.12.0
github.com/google/uuid v1.5.0
github.com/jackc/pgx/v5 v5.10.0
github.com/qiniu/go-sdk/v7 v7.27.0
github.com/redis/go-redis/v9 v9.22.0
github.com/tencentyun/cos-go-sdk-v5 v0.7.75
@@ -44,7 +45,6 @@ require (
github.com/google/go-querystring v1.1.0 // indirect
github.com/jackc/pgpassfile v1.0.0 // indirect
github.com/jackc/pgservicefile v0.0.0-20240606120523-5a60cdf6a761 // indirect
github.com/jackc/pgx/v5 v5.10.0 // indirect
github.com/jackc/puddle/v2 v2.2.2 // indirect
github.com/jinzhu/inflection v1.0.0 // indirect
github.com/jinzhu/now v1.1.5 // indirect

View File

@@ -8,6 +8,7 @@ import (
"yingce/backend/internal/logging"
"github.com/jackc/pgx/v5/stdlib"
"gorm.io/driver/postgres"
"gorm.io/driver/sqlite"
"gorm.io/gorm"
@@ -39,7 +40,10 @@ func Open(config Config) (*gorm.DB, error) {
if dsn == "" {
return nil, errors.New("PostgreSQL 模式必须配置 DATABASE_URL")
}
return gorm.Open(postgres.Open(dsn), &gorm.Config{Logger: logging.GormLogger()})
return gorm.Open(postgres.New(postgres.Config{
DSN: dsn,
OptionOpenDB: []stdlib.OptionOpenDB{stdlib.OptionBeforeConnect(configurePostgresConnection)},
}), &gorm.Config{Logger: logging.GormLogger()})
default:
return nil, fmt.Errorf("不支持的数据库驱动:%s", driver)
}

View File

@@ -0,0 +1,86 @@
package database
import (
"context"
"fmt"
"maps"
"net"
"strconv"
"strings"
"time"
"github.com/jackc/pgx/v5"
)
// defaultIdleInTransactionSessionTimeout 限制事务内等待下一条命令的时间,
// 避免卡住的 idle in transaction 会话长期持有锁。DSN 显式配置时以 DSN 为准。
const defaultIdleInTransactionSessionTimeout = "5min"
var libpqKeepaliveParams = []string{"keepalives", "keepalives_idle", "keepalives_interval", "keepalives_count"}
// configurePostgresConnection 在每次建立物理连接前调整 pgx 连接配置。
func configurePostgresConnection(_ context.Context, config *pgx.ConnConfig) error {
// stdlib 传入的是浅拷贝,RuntimeParams 与连接器共享,修改前必须复制,否则会跨连接累积。
config.RuntimeParams = maps.Clone(config.RuntimeParams)
if config.RuntimeParams == nil {
config.RuntimeParams = make(map[string]string)
}
if err := applyLibpqKeepalives(config); err != nil {
return err
}
if _, exists := config.RuntimeParams["idle_in_transaction_session_timeout"]; !exists {
// PostgreSQL 按顺序处理 options,DSN 原有 options 放在后面,同名设置以 DSN 为准。
options := "-c idle_in_transaction_session_timeout=" + defaultIdleInTransactionSessionTimeout
if existing := strings.TrimSpace(config.RuntimeParams["options"]); existing != "" {
options += " " + existing
}
config.RuntimeParams["options"] = options
}
return nil
}
// applyLibpqKeepalives 消费 libpq 的 keepalives* 参数并应用到 TCP 拨号器。
// pgx 不识别这些参数,会把它们当作服务端运行参数发送,导致 PostgreSQL 拒绝连接。
// 未配置任何 keepalives* 参数时保持 pgx 默认拨号器(Go 默认 keepalive)不变。
func applyLibpqKeepalives(config *pgx.ConnConfig) error {
values := make(map[string]int)
for _, key := range libpqKeepaliveParams {
raw, ok := config.RuntimeParams[key]
if !ok {
continue
}
delete(config.RuntimeParams, key)
value, err := strconv.Atoi(strings.TrimSpace(raw))
if err != nil || value < 0 || (key == "keepalives" && value > 1) {
return fmt.Errorf("invalid PostgreSQL connection parameter %s=%q", key, raw)
}
values[key] = value
}
if len(values) == 0 {
return nil
}
dialer := &net.Dialer{Timeout: config.ConnectTimeout}
if enabled, ok := values["keepalives"]; ok && enabled == 0 {
dialer.KeepAlive = -1
} else {
// 未配置的项使用 Go 默认值(0);libpq 的 0 表示操作系统默认值,对应 Go 的 -1。
setting := func(key string) int {
value, ok := values[key]
if !ok {
return 0
}
if value == 0 {
return -1
}
return value
}
dialer.KeepAliveConfig = net.KeepAliveConfig{
Enable: true,
Idle: time.Duration(setting("keepalives_idle")) * time.Second,
Interval: time.Duration(setting("keepalives_interval")) * time.Second,
Count: setting("keepalives_count"),
}
}
config.DialFunc = dialer.DialContext
return nil
}

View File

@@ -0,0 +1,81 @@
package database
import (
"net"
"os"
"strconv"
"sync"
"syscall"
"testing"
)
// clientSocketKeepalive 在当前进程中按端口找到客户端 socket,读取其 keepalive 选项。
func clientSocketKeepalive(port int) ([4]int, bool) {
entries, err := os.ReadDir("/proc/self/fd")
if err != nil {
return [4]int{}, false
}
for _, entry := range entries {
fd, err := strconv.Atoi(entry.Name())
if err != nil {
continue
}
addr, err := syscall.Getsockname(fd)
inet, ok := addr.(*syscall.SockaddrInet4)
if err != nil || !ok || inet.Port != port {
continue
}
var values [4]int
for i, option := range [][2]int{
{syscall.SOL_SOCKET, syscall.SO_KEEPALIVE},
{syscall.IPPROTO_TCP, syscall.TCP_KEEPIDLE},
{syscall.IPPROTO_TCP, syscall.TCP_KEEPINTVL},
{syscall.IPPROTO_TCP, syscall.TCP_KEEPCNT},
} {
if values[i], err = syscall.GetsockoptInt(fd, option[0], option[1]); err != nil {
return [4]int{}, false
}
}
return values, true
}
return [4]int{}, false
}
func TestPostgresOpenAppliesLibpqKeepaliveToSocket(t *testing.T) {
for _, tc := range []struct {
name, params string
want [4]int
}{
{"explicit", "keepalives_idle=90 keepalives_interval=20 keepalives_count=6", [4]int{1, 90, 20, 6}},
{"partial keeps Go defaults", "keepalives_idle=45", [4]int{1, 45, 15, 9}},
{"no params keeps Go defaults", "", [4]int{1, 15, 15, 9}},
{"disabled", "keepalives=0", [4]int{0, -1, -1, -1}},
} {
t.Run(tc.name, func(t *testing.T) {
server := newFakePostgresServer(t)
var mu sync.Mutex
var observed [][4]int
server.onStartup = func(conn net.Conn) {
// 已收到启动消息,说明客户端拨号已完成且 socket 选项已设置。
values, ok := clientSocketKeepalive(conn.RemoteAddr().(*net.TCPAddr).Port)
if !ok {
values = [4]int{-2, -2, -2, -2}
}
mu.Lock()
observed = append(observed, values)
mu.Unlock()
}
openFakePostgres(t, server, tc.params, 0)
mu.Lock()
defer mu.Unlock()
for _, values := range observed {
if values[0] == -2 {
t.Fatal("client socket not found")
}
if values[0] != tc.want[0] || (tc.want[0] == 1 && values != tc.want) {
t.Fatalf("keepalive (enabled, idle, interval, count) = %v, want %v", values, tc.want)
}
}
})
}
}

View File

@@ -0,0 +1,161 @@
package database
import (
"bytes"
"encoding/binary"
"fmt"
"io"
"net"
"strings"
"sync"
"testing"
"time"
)
// fakePostgresServer 只实现启动握手和简单查询,用于检查客户端发送的启动参数。
type fakePostgresServer struct {
listener net.Listener
onStartup func(conn net.Conn)
mu sync.Mutex
startups []map[string]string
}
func newFakePostgresServer(t *testing.T) *fakePostgresServer {
t.Helper()
listener, err := net.Listen("tcp", "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
server := &fakePostgresServer{listener: listener}
t.Cleanup(func() { _ = listener.Close() })
go func() {
for {
conn, err := listener.Accept()
if err != nil {
return
}
go server.serve(conn)
}
}()
return server
}
func (s *fakePostgresServer) port() int {
return s.listener.Addr().(*net.TCPAddr).Port
}
func (s *fakePostgresServer) recorded() []map[string]string {
s.mu.Lock()
defer s.mu.Unlock()
return append([]map[string]string(nil), s.startups...)
}
func (s *fakePostgresServer) serve(conn net.Conn) {
defer conn.Close()
_ = conn.SetDeadline(time.Now().Add(5 * time.Second))
var size uint32
if err := binary.Read(conn, binary.BigEndian, &size); err != nil || size < 8 {
return
}
body := make([]byte, size-4)
if _, err := io.ReadFull(conn, body); err != nil {
return
}
params := map[string]string{}
fields := bytes.Split(body[4:], []byte{0})
for i := 0; i+1 < len(fields) && len(fields[i]) > 0; i += 2 {
params[string(fields[i])] = string(fields[i+1])
}
s.mu.Lock()
s.startups = append(s.startups, params)
s.mu.Unlock()
if s.onStartup != nil {
s.onStartup(conn)
}
// AuthenticationOk + ReadyForQuery
_, _ = conn.Write([]byte{'R', 0, 0, 0, 8, 0, 0, 0, 0, 'Z', 0, 0, 0, 5, 'I'})
header := make([]byte, 5)
for {
if _, err := io.ReadFull(conn, header); err != nil {
return
}
payload := make([]byte, binary.BigEndian.Uint32(header[1:])-4)
if _, err := io.ReadFull(conn, payload); err != nil {
return
}
switch header[0] {
case 'Q': // EmptyQueryResponse + ReadyForQuery
_, _ = conn.Write([]byte{'I', 0, 0, 0, 4, 'Z', 0, 0, 0, 5, 'I'})
case 'X':
return
}
}
}
func openFakePostgres(t *testing.T, server *fakePostgresServer, params string, pings int) []map[string]string {
t.Helper()
dsn := fmt.Sprintf("host=127.0.0.1 port=%d user=test dbname=test sslmode=disable %s", server.port(), params)
db, err := Open(Config{Driver: "postgres", DSN: dsn})
if err != nil {
t.Fatalf("open: %v", err)
}
sqlDB, err := db.DB()
if err != nil {
t.Fatal(err)
}
defer sqlDB.Close()
// 不保留空闲连接,每次 Ping 都会建立新的物理连接。
sqlDB.SetMaxIdleConns(0)
for range pings {
if err := sqlDB.Ping(); err != nil {
t.Fatalf("ping: %v", err)
}
}
startups := server.recorded()
if len(startups) < pings+1 {
t.Fatalf("startup count = %d, want at least %d", len(startups), pings+1)
}
return startups
}
func TestPostgresOpenConsumesLibpqKeepaliveParams(t *testing.T) {
server := newFakePostgresServer(t)
startups := openFakePostgres(t, server, "keepalives=1 keepalives_idle=90 keepalives_interval=20 keepalives_count=6", 0)
for _, params := range startups {
for key := range params {
if strings.HasPrefix(key, "keepalives") {
t.Fatalf("libpq parameter %s sent to server: %v", key, params)
}
}
}
}
func TestPostgresOpenRejectsInvalidKeepaliveParam(t *testing.T) {
server := newFakePostgresServer(t)
dsn := fmt.Sprintf("host=127.0.0.1 port=%d user=test dbname=test sslmode=disable keepalives_idle=abc", server.port())
if _, err := Open(Config{Driver: "postgres", DSN: dsn}); err == nil || !strings.Contains(err.Error(), "keepalives_idle") {
t.Fatalf("expected invalid keepalives_idle error, got %v", err)
}
}
func TestPostgresOpenSetsIdleInTransactionTimeoutPerConnection(t *testing.T) {
server := newFakePostgresServer(t)
startups := openFakePostgres(t, server, "options='-c statement_timeout=4s'", 3)
want := "-c idle_in_transaction_session_timeout=5min -c statement_timeout=4s"
for i, params := range startups {
// 每条连接都应得到相同的 options,不能因共享 RuntimeParams 而跨连接累积。
if params["options"] != want {
t.Fatalf("connection %d options = %q, want %q", i, params["options"], want)
}
}
}
func TestPostgresOpenKeepsExplicitIdleInTransactionTimeout(t *testing.T) {
server := newFakePostgresServer(t)
startups := openFakePostgres(t, server, "idle_in_transaction_session_timeout=0 options='-c statement_timeout=4s'", 1)
for i, params := range startups {
if params["idle_in_transaction_session_timeout"] != "0" || params["options"] != "-c statement_timeout=4s" {
t.Fatalf("connection %d overrode explicit timeout: %v", i, params)
}
}
}