Add DISABLE_GLOBAL_RATELIMIT_DETECTION flag, return a 429 not a 500 when /users/@me or /gateway/bot calls 429

Author: germanoeich

CONFIG.md

@@ -65,6 +65,16 @@ Allows you to define custom global request limits for one or multiple bots. The
 Format: Command separated list of **user id** and limit combo, separated by `:` and with no spaces at all. Don't use application ids.
 Example: `392827169497284619:100,227115752396685313:80`
 
+
+##### DISABLE_GLOBAL_RATELIMIT_DETECTION
+Disables the optimistic global rest limit detection. This detection uses the /gateway/bot endpoint, which has a low ratelimit and can cause issues with requests being dropped/delayed as cluster size grows.
+
+You probably want to set BOT_RATELIMIT_OVERRIDES if you set this to true.
+
+Default: false
+
+In the future, this will be the only possible behavior.
+
 ## Unstable env vars
 Collection of env vars that may be removed at any time, mainly used for Discord introducing new behaviour on their edge api versions
 

README.md

@@ -24,23 +24,24 @@ The proxy sits between the client and discord. Instead of pointing to discord.co
 
 Configuration options are
 
-| Variable        | Value                                         | Default                 |
-|-----------------|-----------------------------------------------|-------------------------|
-| LOG_LEVEL       | panic, fatal, error, warn, info, debug, trace | info                    |
-| PORT            | number                                        | 8080                    |
-| METRICS_PORT    | number                                        | 9000                    |
-| ENABLE_METRICS  | boolean                                       | true                    |
-| ENABLE_PPROF    | boolean                                       | false                   |
-| BUFFER_SIZE     | number                                        | 50                      |
-| OUTBOUND_IP     | string                                        | ""                      |
-| BIND_IP         | string                                        | 0.0.0.0                 |
-| REQUEST_TIMEOUT | number (milliseconds)                         | 5000                    |
-| CLUSTER_PORT    | number                                        | 7946                    |
-| CLUSTER_MEMBERS | string list (comma separated)                 | ""                      |
-| CLUSTER_DNS     | string                                        | ""                      |
-| MAX_BEARER_COUNT| number                                        | 1024                    |
-| DISABLE_HTTP_2  | bool                                          | true                    |
-| BOT_RATELIMIT_OVERRIDES | string list (comma separated)          | ""                      |
+| Variable        | Value                                       | Default |
+|-----------------|---------------------------------------------|---------|
+| LOG_LEVEL       | panic, fatal, error, warn, info, debug, trace | info    |
+| PORT            | number                                      | 8080    |
+| METRICS_PORT    | number                                      | 9000    |
+| ENABLE_METRICS  | boolean                                     | true    |
+| ENABLE_PPROF    | boolean                                     | false   |
+| BUFFER_SIZE     | number                                      | 50      |
+| OUTBOUND_IP     | string                                      | ""      |
+| BIND_IP         | string                                      | 0.0.0.0 |
+| REQUEST_TIMEOUT | number (milliseconds)                       | 5000    |
+| CLUSTER_PORT    | number                                      | 7946    |
+| CLUSTER_MEMBERS | string list (comma separated)               | ""      |
+| CLUSTER_DNS     | string                                      | ""      |
+| MAX_BEARER_COUNT| number                                      | 1024    |
+| DISABLE_HTTP_2  | bool                                        | true    |
+| BOT_RATELIMIT_OVERRIDES | string list (comma separated)       | ""      |
+| DISABLE_GLOBAL_RATELIMIT_DETECTION | boolean                  | false   |
 
 Information on each config var can be found [here](https://github.com/germanoeich/nirn-proxy/blob/main/CONFIG.md)
 

lib/discord.go

@@ -22,14 +22,16 @@ var contextTimeout time.Duration
 
 var globalOverrideMap = make(map[string]uint)
 
+var disableRestLimitDetection = false
+
 type BotGatewayResponse struct {
 	SessionStartLimit map[string]int `json:"session_start_limit"`
 }
 
 type BotUserResponse struct {
-	Id string `json:"id"`
+	Id       string `json:"id"`
 	Username string `json:"username"`
-	Discrim string `json:"discriminator"`
+	Discrim  string `json:"discriminator"`
 }
 
 func createTransport(ip string, disableHttp2 bool) http.RoundTripper {
@@ -56,9 +58,9 @@ func createTransport(ip string, disableHttp2 bool) http.RoundTripper {
 		}
 
 		dialer := &net.Dialer{
-			LocalAddr:     addr,
-			Timeout:       30 * time.Second,
-			KeepAlive:     30 * time.Second,
+			LocalAddr: addr,
+			Timeout:   30 * time.Second,
+			KeepAlive: 30 * time.Second,
 		}
 
 		dialContext := func(ctx context.Context, network, addr string) (net.Conn, error) {
@@ -109,15 +111,17 @@ func parseGlobalOverrides(overrides string) {
 	}
 }
 
-func ConfigureDiscordHTTPClient(ip string, timeout time.Duration, disableHttp2 bool, globalOverrides string) {
+func ConfigureDiscordHTTPClient(ip string, timeout time.Duration, disableHttp2 bool, globalOverrides string, disableRestDetection bool) {
 	transport := createTransport(ip, disableHttp2)
 	client = &http.Client{
 		Transport: transport,
-		Timeout: 90 * time.Second,
+		Timeout:   90 * time.Second,
 	}
 
 	contextTimeout = timeout
 
+	disableRestLimitDetection = disableRestDetection
+
 	parseGlobalOverrides(globalOverrides)
 }
 
@@ -137,6 +141,10 @@ func GetBotGlobalLimit(token string, user *BotUserResponse) (uint, error) {
 		return 50, nil
 	}
 
+	if disableRestLimitDetection {
+		return 50, nil
+	}
+
 	bot, err := doDiscordReq(context.Background(), "/api/v9/gateway/bot", "GET", nil, map[string][]string{"Authorization": {token}}, "")
 
 	if err != nil {
@@ -205,7 +213,7 @@ func GetBotUser(token string) (*BotUserResponse, error) {
 }
 
 func doDiscordReq(ctx context.Context, path string, method string, body io.ReadCloser, header http.Header, query string) (*http.Response, error) {
-	discordReq, err := http.NewRequestWithContext(ctx, method, "https://discord.com" + path + "?" + query, body)
+	discordReq, err := http.NewRequestWithContext(ctx, method, "https://discord.com"+path+"?"+query, body)
 	if err != nil {
 		return nil, err
 	}
@@ -231,7 +239,6 @@ func doDiscordReq(ctx context.Context, path string, method string, body io.ReadC
 				status = "429 Shared"
 			}
 		}
-		
 
 		RequestHistogram.With(map[string]string{"route": route, "status": status, "method": method, "clientId": identifier.(string)}).Observe(elapsed)
 	}

lib/queue.go

@@ -75,6 +75,7 @@ func NewRequestQueue(processor func(ctx context.Context, item *QueueItem) (*http
 				botLimit:          limit,
 			}, nil
 		}
+
 		return nil, err
 	}
 

lib/queue_manager.go

@@ -297,10 +297,15 @@ func (m *QueueManager) fulfillRequest(resp *http.ResponseWriter, req *http.Reque
 		}
 
 		if err != nil {
-			(*resp).WriteHeader(500)
-			(*resp).Write([]byte(err.Error()))
-			ErrorCounter.Inc()
-			logEntry.WithFields(logrus.Fields{"function": "getOrCreateQueue", "queueType": queueType}).Error(err)
+			if strings.HasPrefix(err.Error(), "429") {
+				Generate429(resp)
+				logEntry.WithFields(logrus.Fields{"function": "getOrCreateQueue", "queueType": queueType}).Warn(err)
+			} else {
+				(*resp).WriteHeader(500)
+				(*resp).Write([]byte(err.Error()))
+				ErrorCounter.Inc()
+				logEntry.WithFields(logrus.Fields{"function": "getOrCreateQueue", "queueType": queueType}).Error(err)
+			}
 			return
 		}
 

main.go

@@ -74,7 +74,9 @@ func main() {
 
 	globalOverrides := lib.EnvGet("BOT_RATELIMIT_OVERRIDES", "")
 
-	lib.ConfigureDiscordHTTPClient(outboundIp, time.Duration(timeout)*time.Millisecond, disableHttp2, globalOverrides)
+	disableGlobalRatelimitDetection := lib.EnvGetBool("DISABLE_GLOBAL_RATELIMIT_DETECTION", false)
+
+	lib.ConfigureDiscordHTTPClient(outboundIp, time.Duration(timeout)*time.Millisecond, disableHttp2, globalOverrides, disableGlobalRatelimitDetection)
 
 	port := lib.EnvGet("PORT", "8080")
 	bindIp := lib.EnvGet("BIND_IP", "0.0.0.0")