perf(postgres): 回收突发连接并等待验收副作用

失败 Run 的未提交任务已取消,但退款 outbox 尚未完成时下一轮会继承 API 连接高水位和 canceled acquire 增量,污染容量验收。

新增 AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS,生产配置 30 秒回收突发空闲连接;验收中止后等待旧任务、退款、回调和数据库连接全部收敛后才允许新负载启动。

验证:Go 全量测试、gofmt、bash -n、ShellCheck、kubectl kustomize 和 git diff --check 通过。
This commit is contained in:
2026-07-31 03:46:41 +08:00
parent 75a56b21bc
commit 36546bd5c4
11 changed files with 80 additions and 18 deletions
+1
View File
@@ -40,6 +40,7 @@ func main() {
db, err := store.ConnectWithPoolOptions(ctx, cfg.DatabaseURL, store.PostgresPoolOptions{
MaxConns: cfg.DatabaseMaxConns,
MinIdleConns: cfg.DatabaseMinIdleConns,
MaxConnIdleTime: time.Duration(cfg.DatabaseMaxConnIdleSeconds) * time.Second,
IdleInTransactionTimeout: time.Duration(cfg.DatabaseIdleInTransactionTimeoutSeconds) * time.Second,
LockTimeout: time.Duration(cfg.DatabaseLockTimeoutSeconds) * time.Second,
})
+5
View File
@@ -60,6 +60,7 @@ type Config struct {
ProcessRole string
DatabaseMaxConns int
DatabaseMinIdleConns int
DatabaseMaxConnIdleSeconds int
DatabaseIdleInTransactionTimeoutSeconds int
DatabaseLockTimeoutSeconds int
MediaRequestConcurrency int
@@ -124,6 +125,7 @@ func Load() Config {
ProcessRole: strings.ToLower(strings.TrimSpace(env("AI_GATEWAY_PROCESS_ROLE", "all"))),
DatabaseMaxConns: envInt("AI_GATEWAY_DATABASE_MAX_CONNS", 0),
DatabaseMinIdleConns: envOptionalIntValidated("AI_GATEWAY_DATABASE_MIN_IDLE_CONNS", 0),
DatabaseMaxConnIdleSeconds: envOptionalIntValidated("AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS", 0),
DatabaseIdleInTransactionTimeoutSeconds: envOptionalIntValidated(
"AI_GATEWAY_DATABASE_IDLE_IN_TRANSACTION_TIMEOUT_SECONDS",
@@ -153,6 +155,9 @@ func (c Config) Validate() error {
(c.DatabaseMaxConns > 0 && c.DatabaseMinIdleConns > c.DatabaseMaxConns) {
return errors.New("AI_GATEWAY_DATABASE_MIN_IDLE_CONNS must be between 0 and AI_GATEWAY_DATABASE_MAX_CONNS")
}
if c.DatabaseMaxConnIdleSeconds < 0 || c.DatabaseMaxConnIdleSeconds > 3600 {
return errors.New("AI_GATEWAY_DATABASE_MAX_CONN_IDLE_SECONDS must be between 0 and 3600")
}
if c.DatabaseIdleInTransactionTimeoutSeconds < 0 || c.DatabaseIdleInTransactionTimeoutSeconds > 3600 {
return errors.New("AI_GATEWAY_DATABASE_IDLE_IN_TRANSACTION_TIMEOUT_SECONDS must be between 0 and 3600")
}
+8 -1
View File
@@ -145,11 +145,13 @@ func TestProcessRolePrecedenceAndCompatibility(t *testing.T) {
func TestValidateProcessRoleAndDatabasePool(t *testing.T) {
cfg := Load()
if cfg.DatabaseMinIdleConns != 0 ||
cfg.DatabaseMaxConnIdleSeconds != 0 ||
cfg.DatabaseIdleInTransactionTimeoutSeconds != 60 ||
cfg.DatabaseLockTimeoutSeconds != 30 {
t.Fatalf(
"database pool minimum/transaction timeouts = %d/%d/%d, want 0/60/30",
"database pool minimum/max-idle/transaction timeouts = %d/%d/%d/%d, want 0/0/60/30",
cfg.DatabaseMinIdleConns,
cfg.DatabaseMaxConnIdleSeconds,
cfg.DatabaseIdleInTransactionTimeoutSeconds,
cfg.DatabaseLockTimeoutSeconds,
)
@@ -174,6 +176,11 @@ func TestValidateProcessRoleAndDatabasePool(t *testing.T) {
t.Fatalf("Validate() error = %v, want invalid non-integer database min idle conns", err)
}
t.Setenv("AI_GATEWAY_DATABASE_MIN_IDLE_CONNS", "4")
cfg.DatabaseMaxConnIdleSeconds = 3601
if err := cfg.Validate(); err == nil || !strings.Contains(err.Error(), "DATABASE_MAX_CONN_IDLE_SECONDS") {
t.Fatalf("Validate() error = %v, want invalid database max connection idle seconds", err)
}
cfg.DatabaseMaxConnIdleSeconds = 30
cfg.DatabaseIdleInTransactionTimeoutSeconds = 3601
if err := cfg.Validate(); err == nil || !strings.Contains(err.Error(), "IDLE_IN_TRANSACTION") {
t.Fatalf("Validate() error = %v, want invalid idle transaction timeout", err)
+4
View File
@@ -32,6 +32,7 @@ const (
type PostgresPoolOptions struct {
MaxConns int
MinIdleConns int
MaxConnIdleTime time.Duration
IdleInTransactionTimeout time.Duration
LockTimeout time.Duration
}
@@ -132,6 +133,9 @@ func postgresPoolConfigWithOptions(databaseURL string, options PostgresPoolOptio
config.MaxConns = int32(options.MaxConns)
}
config.MinIdleConns = int32(min(max(options.MinIdleConns, 0), int(config.MaxConns)))
if options.MaxConnIdleTime > 0 {
config.MaxConnIdleTime = options.MaxConnIdleTime
}
return config, nil
}
@@ -61,6 +61,7 @@ func TestPostgresPoolConfigSetsBoundedTransactionTimeouts(t *testing.T) {
PostgresPoolOptions{
MaxConns: 32,
MinIdleConns: 4,
MaxConnIdleTime: 30 * time.Second,
IdleInTransactionTimeout: 60 * time.Second,
LockTimeout: 30 * time.Second,
},
@@ -77,6 +78,9 @@ func TestPostgresPoolConfigSetsBoundedTransactionTimeouts(t *testing.T) {
if config.MaxConns != 32 || config.MinIdleConns != 4 {
t.Fatalf("pool bounds max=%d minIdle=%d, want 32/4", config.MaxConns, config.MinIdleConns)
}
if config.MaxConnIdleTime != 30*time.Second {
t.Fatalf("pool max idle time=%s, want 30s", config.MaxConnIdleTime)
}
}
func TestIsPostgresUnavailableClassifiesConnectivityFailures(t *testing.T) {