feat: add OpenAI image generation controls

This commit is contained in:
2ue
2026-05-05 03:26:54 +08:00
parent 4de28fec8c
commit 6faa344916
85 changed files with 6086 additions and 568 deletions

View File

@@ -575,6 +575,24 @@ type ConcurrencyConfig struct {
PingInterval int `mapstructure:"ping_interval"`
}
type ImageConcurrencyConfig struct {
// Enabled: 是否启用图片生成独立并发限制,默认关闭以保持现有行为
Enabled bool `mapstructure:"enabled"`
// MaxConcurrentRequests: 当前进程允许同时处理的图片生成请求数0表示不限制
MaxConcurrentRequests int `mapstructure:"max_concurrent_requests"`
// OverflowMode: 图片并发达到上限后的处理方式reject/wait
OverflowMode string `mapstructure:"overflow_mode"`
// WaitTimeoutSeconds: overflow_mode=wait 时等待图片并发槽位的超时时间(秒)
WaitTimeoutSeconds int `mapstructure:"wait_timeout_seconds"`
// MaxWaitingRequests: overflow_mode=wait 时当前进程允许排队等待的图片请求数
MaxWaitingRequests int `mapstructure:"max_waiting_requests"`
}
const (
ImageConcurrencyOverflowModeReject = "reject"
ImageConcurrencyOverflowModeWait = "wait"
)
// GatewayConfig API网关相关配置
type GatewayConfig struct {
// 等待上游响应头的超时时间0表示无超时
@@ -604,6 +622,8 @@ type GatewayConfig struct {
OpenAIPassthroughAllowTimeoutHeaders bool `mapstructure:"openai_passthrough_allow_timeout_headers"`
// OpenAIWS: OpenAI Responses WebSocket 配置(默认开启,可按需回滚到 HTTP
OpenAIWS GatewayOpenAIWSConfig `mapstructure:"openai_ws"`
// ImageConcurrency: 图片生成独立并发限制配置(默认关闭)
ImageConcurrency ImageConcurrencyConfig `mapstructure:"image_concurrency"`
// HTTP 上游连接池配置(性能优化:支持高并发场景调优)
// MaxIdleConns: 所有主机的最大空闲连接总数
@@ -635,6 +655,10 @@ type GatewayConfig struct {
StreamDataIntervalTimeout int `mapstructure:"stream_data_interval_timeout"`
// StreamKeepaliveInterval: 流式 keepalive 间隔0表示禁用
StreamKeepaliveInterval int `mapstructure:"stream_keepalive_interval"`
// ImageStreamDataIntervalTimeout: 图片流数据间隔超时0表示禁用
ImageStreamDataIntervalTimeout int `mapstructure:"image_stream_data_interval_timeout"`
// ImageStreamKeepaliveInterval: 图片流式 keepalive 间隔0表示禁用
ImageStreamKeepaliveInterval int `mapstructure:"image_stream_keepalive_interval"`
// MaxLineSize: 上游 SSE 单行最大字节数0使用默认值
MaxLineSize int `mapstructure:"max_line_size"`
@@ -1672,6 +1696,11 @@ func setDefaults() {
viper.SetDefault("gateway.openai_ws.scheduler_score_weights.queue", 0.7)
viper.SetDefault("gateway.openai_ws.scheduler_score_weights.error_rate", 0.8)
viper.SetDefault("gateway.openai_ws.scheduler_score_weights.ttft", 0.5)
viper.SetDefault("gateway.image_concurrency.enabled", false)
viper.SetDefault("gateway.image_concurrency.max_concurrent_requests", 0)
viper.SetDefault("gateway.image_concurrency.overflow_mode", ImageConcurrencyOverflowModeReject)
viper.SetDefault("gateway.image_concurrency.wait_timeout_seconds", 30)
viper.SetDefault("gateway.image_concurrency.max_waiting_requests", 100)
viper.SetDefault("gateway.antigravity_fallback_cooldown_minutes", 1)
viper.SetDefault("gateway.antigravity_extra_retries", 10)
viper.SetDefault("gateway.max_body_size", int64(256*1024*1024))
@@ -1689,6 +1718,8 @@ func setDefaults() {
viper.SetDefault("gateway.concurrency_slot_ttl_minutes", 30) // 并发槽位过期时间(支持超长请求)
viper.SetDefault("gateway.stream_data_interval_timeout", 180)
viper.SetDefault("gateway.stream_keepalive_interval", 10)
viper.SetDefault("gateway.image_stream_data_interval_timeout", 900)
viper.SetDefault("gateway.image_stream_keepalive_interval", 10)
viper.SetDefault("gateway.max_line_size", 500*1024*1024)
viper.SetDefault("gateway.scheduling.sticky_session_max_waiting", 3)
viper.SetDefault("gateway.scheduling.sticky_session_wait_timeout", 120*time.Second)
@@ -2239,6 +2270,21 @@ func (c *Config) Validate() error {
ConnectionPoolIsolationProxy, ConnectionPoolIsolationAccount, ConnectionPoolIsolationAccountProxy)
}
}
if c.Gateway.ImageConcurrency.MaxConcurrentRequests < 0 {
return fmt.Errorf("gateway.image_concurrency.max_concurrent_requests must be non-negative")
}
switch strings.TrimSpace(c.Gateway.ImageConcurrency.OverflowMode) {
case "", ImageConcurrencyOverflowModeReject, ImageConcurrencyOverflowModeWait:
default:
return fmt.Errorf("gateway.image_concurrency.overflow_mode must be one of: %s/%s",
ImageConcurrencyOverflowModeReject, ImageConcurrencyOverflowModeWait)
}
if c.Gateway.ImageConcurrency.WaitTimeoutSeconds < 0 {
return fmt.Errorf("gateway.image_concurrency.wait_timeout_seconds must be non-negative")
}
if c.Gateway.ImageConcurrency.MaxWaitingRequests < 0 {
return fmt.Errorf("gateway.image_concurrency.max_waiting_requests must be non-negative")
}
if c.Gateway.MaxIdleConns <= 0 {
return fmt.Errorf("gateway.max_idle_conns must be positive")
}
@@ -2277,6 +2323,20 @@ func (c *Config) Validate() error {
(c.Gateway.StreamKeepaliveInterval < 5 || c.Gateway.StreamKeepaliveInterval > 30) {
return fmt.Errorf("gateway.stream_keepalive_interval must be 0 or between 5-30 seconds")
}
if c.Gateway.ImageStreamDataIntervalTimeout < 0 {
return fmt.Errorf("gateway.image_stream_data_interval_timeout must be non-negative")
}
if c.Gateway.ImageStreamDataIntervalTimeout != 0 &&
(c.Gateway.ImageStreamDataIntervalTimeout < 60 || c.Gateway.ImageStreamDataIntervalTimeout > 1800) {
return fmt.Errorf("gateway.image_stream_data_interval_timeout must be 0 or between 60-1800 seconds")
}
if c.Gateway.ImageStreamKeepaliveInterval < 0 {
return fmt.Errorf("gateway.image_stream_keepalive_interval must be non-negative")
}
if c.Gateway.ImageStreamKeepaliveInterval != 0 &&
(c.Gateway.ImageStreamKeepaliveInterval < 5 || c.Gateway.ImageStreamKeepaliveInterval > 60) {
return fmt.Errorf("gateway.image_stream_keepalive_interval must be 0 or between 5-60 seconds")
}
// 兼容旧键 sticky_previous_response_ttl_seconds
if c.Gateway.OpenAIWS.StickyResponseIDTTLSeconds <= 0 && c.Gateway.OpenAIWS.StickyPreviousResponseTTLSeconds > 0 {
c.Gateway.OpenAIWS.StickyResponseIDTTLSeconds = c.Gateway.OpenAIWS.StickyPreviousResponseTTLSeconds

View File

@@ -1282,6 +1282,46 @@ func TestValidateConfigErrors(t *testing.T) {
mutate: func(c *Config) { c.Gateway.StreamDataIntervalTimeout = -1 },
wantErr: "gateway.stream_data_interval_timeout must be non-negative",
},
{
name: "gateway image stream keepalive range",
mutate: func(c *Config) { c.Gateway.ImageStreamKeepaliveInterval = 4 },
wantErr: "gateway.image_stream_keepalive_interval",
},
{
name: "gateway image stream keepalive negative",
mutate: func(c *Config) { c.Gateway.ImageStreamKeepaliveInterval = -1 },
wantErr: "gateway.image_stream_keepalive_interval must be non-negative",
},
{
name: "gateway image stream data interval range",
mutate: func(c *Config) { c.Gateway.ImageStreamDataIntervalTimeout = 30 },
wantErr: "gateway.image_stream_data_interval_timeout",
},
{
name: "gateway image stream data interval negative",
mutate: func(c *Config) { c.Gateway.ImageStreamDataIntervalTimeout = -1 },
wantErr: "gateway.image_stream_data_interval_timeout must be non-negative",
},
{
name: "gateway image concurrency max negative",
mutate: func(c *Config) { c.Gateway.ImageConcurrency.MaxConcurrentRequests = -1 },
wantErr: "gateway.image_concurrency.max_concurrent_requests must be non-negative",
},
{
name: "gateway image concurrency overflow mode invalid",
mutate: func(c *Config) { c.Gateway.ImageConcurrency.OverflowMode = "queue" },
wantErr: "gateway.image_concurrency.overflow_mode",
},
{
name: "gateway image concurrency wait timeout negative",
mutate: func(c *Config) { c.Gateway.ImageConcurrency.WaitTimeoutSeconds = -1 },
wantErr: "gateway.image_concurrency.wait_timeout_seconds must be non-negative",
},
{
name: "gateway image concurrency max waiting negative",
mutate: func(c *Config) { c.Gateway.ImageConcurrency.MaxWaitingRequests = -1 },
wantErr: "gateway.image_concurrency.max_waiting_requests must be non-negative",
},
{
name: "gateway max line size",
mutate: func(c *Config) { c.Gateway.MaxLineSize = 1024 },
@@ -1754,3 +1794,41 @@ func TestLoad_DefaultGatewayUsageRecordConfig(t *testing.T) {
t.Fatalf("auto_scale_cooldown_seconds = %d, want 10", cfg.Gateway.UsageRecord.AutoScaleCooldownSeconds)
}
}
func TestLoad_DefaultGatewayImageStreamConfig(t *testing.T) {
resetViperWithJWTSecret(t)
cfg, err := Load()
if err != nil {
t.Fatalf("Load() error: %v", err)
}
if cfg.Gateway.StreamDataIntervalTimeout != 180 {
t.Fatalf("stream_data_interval_timeout = %d, want 180", cfg.Gateway.StreamDataIntervalTimeout)
}
if cfg.Gateway.StreamKeepaliveInterval != 10 {
t.Fatalf("stream_keepalive_interval = %d, want 10", cfg.Gateway.StreamKeepaliveInterval)
}
if cfg.Gateway.ImageStreamDataIntervalTimeout != 900 {
t.Fatalf("image_stream_data_interval_timeout = %d, want 900", cfg.Gateway.ImageStreamDataIntervalTimeout)
}
if cfg.Gateway.ImageStreamKeepaliveInterval != 10 {
t.Fatalf("image_stream_keepalive_interval = %d, want 10", cfg.Gateway.ImageStreamKeepaliveInterval)
}
if cfg.Gateway.ImageConcurrency.Enabled {
t.Fatalf("image_concurrency.enabled = true, want false")
}
if cfg.Gateway.ImageConcurrency.MaxConcurrentRequests != 0 {
t.Fatalf("image_concurrency.max_concurrent_requests = %d, want 0", cfg.Gateway.ImageConcurrency.MaxConcurrentRequests)
}
if cfg.Gateway.ImageConcurrency.OverflowMode != ImageConcurrencyOverflowModeReject {
t.Fatalf("image_concurrency.overflow_mode = %q, want %q", cfg.Gateway.ImageConcurrency.OverflowMode, ImageConcurrencyOverflowModeReject)
}
if cfg.Gateway.ImageConcurrency.WaitTimeoutSeconds != 30 {
t.Fatalf("image_concurrency.wait_timeout_seconds = %d, want 30", cfg.Gateway.ImageConcurrency.WaitTimeoutSeconds)
}
if cfg.Gateway.ImageConcurrency.MaxWaitingRequests != 100 {
t.Fatalf("image_concurrency.max_waiting_requests = %d, want 100", cfg.Gateway.ImageConcurrency.MaxWaitingRequests)
}
if cfg.Gateway.ImageStreamDataIntervalTimeout <= cfg.Gateway.StreamDataIntervalTimeout {
t.Fatalf("image stream timeout = %d, want greater than ordinary stream timeout %d", cfg.Gateway.ImageStreamDataIntervalTimeout, cfg.Gateway.StreamDataIntervalTimeout)
}
}