Add option to configure prometheus port (#3187)

* add prometheus port

* fix doc

* add port for trtllm and llamacpp

* Fixing format after rebase.

---------

Co-authored-by: Nicolas Patry <patry.nicolas@protonmail.com>

backends/llamacpp/src/main.rs

@@ -119,6 +119,9 @@ struct Args {
     #[clap(default_value = "3000", long, short, env)]
     port: u16,
 
+    #[clap(default_value = "9000", long, short, env)]
+    prometheus_port: u16,
+
     /// Enable JSON output format.
     #[clap(long, env)]
     json_output: bool,
@@ -317,6 +320,7 @@ async fn main() -> Result<(), RouterError> {
         args.max_client_batch_size,
         args.usage_stats,
         args.payload_limit,
+        args.prometheus_port,
     )
     .await?;
     Ok(())

backends/trtllm/src/main.rs

@@ -37,6 +37,8 @@ struct Args {
     hostname: String,
     #[clap(default_value = "3000", long, short, env)]
     port: u16,
+    #[clap(default_value = "9000", long, short, env)]
+    prometheus_port: u16,
     #[clap(long, env, required = true)]
     tokenizer_name: String,
     #[clap(long, env)]
@@ -227,6 +229,7 @@ async fn main() -> Result<(), TensorRtLlmBackendError> {
         max_batch_total_tokens,
         hostname,
         port,
+        prometheus_port,
         tokenizer_name,
         tokenizer_config_path,
         revision,
@@ -322,6 +325,7 @@ async fn main() -> Result<(), TensorRtLlmBackendError> {
                 max_client_batch_size,
                 usage_stats,
                 payload_limit,
+                prometheus_port,
             )
             .await?;
             Ok(())

backends/v2/src/main.rs

@@ -36,6 +36,8 @@ struct Args {
     hostname: String,
     #[clap(default_value = "3000", long, short, env)]
     port: u16,
+    #[clap(default_value = "9000", long, short, env)]
+    prometheus_port: u16,
     #[clap(default_value = "/tmp/text-generation-server-0", long, env)]
     master_shard_uds_path: String,
     #[clap(default_value = "bigscience/bloom", long, env)]
@@ -99,6 +101,7 @@ async fn main() -> Result<(), RouterError> {
         max_batch_size,
         hostname,
         port,
+        prometheus_port,
         master_shard_uds_path,
         tokenizer_name,
         tokenizer_config_path,
@@ -198,6 +201,7 @@ async fn main() -> Result<(), RouterError> {
         max_client_batch_size,
         usage_stats,
         payload_limit,
+        prometheus_port,
     )
     .await?;
     Ok(())

backends/v3/src/main.rs

@@ -36,6 +36,8 @@ struct Args {
     hostname: String,
     #[clap(default_value = "3000", long, short, env)]
     port: u16,
+    #[clap(default_value = "9000", long, short, env)]
+    prometheus_port: u16,
     #[clap(default_value = "/tmp/text-generation-server-0", long, env)]
     master_shard_uds_path: String,
     #[clap(default_value = "bigscience/bloom", long, env)]
@@ -99,6 +101,7 @@ async fn main() -> Result<(), RouterError> {
         max_batch_size,
         hostname,
         port,
+        prometheus_port,
         master_shard_uds_path,
         tokenizer_name,
         tokenizer_config_path,
@@ -214,6 +217,7 @@ async fn main() -> Result<(), RouterError> {
         max_client_batch_size,
         usage_stats,
         payload_limit,
+        prometheus_port,
     )
     .await?;
     Ok(())

docs/source/reference/launcher.md

@@ -251,6 +251,15 @@ Options:
           [env: PORT=]
           [default: 3000]
 
+```
+## PROMETHEUS_PORT
+```shell
+  -p, --prometheus-port <PROMETHEUS_PORT>
+          The Prometheus port to listen on
+          
+          [env: PROMETHEUS_PORT=]
+          [default: 9000]
+
 ```
 ## SHARD_UDS_PATH
 ```shell

launcher/src/main.rs

@@ -773,6 +773,10 @@ struct Args {
     #[clap(default_value = "3000", long, short, env)]
     port: u16,
 
+    /// The Prometheus port to listen on.
+    #[clap(default_value = "9000", long, short, env)]
+    prometheus_port: u16,
+
     /// The name of the socket for gRPC communication between the webserver
     /// and the shards.
     #[clap(default_value = "/tmp/text-generation-server", long, env)]
@@ -1848,6 +1852,8 @@ fn spawn_webserver(
         args.hostname.to_string(),
         "--port".to_string(),
         args.port.to_string(),
+        "--prometheus-port".to_string(),
+        args.prometheus_port.to_string(),
         "--master-shard-uds-path".to_string(),
         format!("{}-0", args.shard_uds_path),
         "--tokenizer-name".to_string(),

router/src/server.rs

@@ -1522,6 +1522,7 @@ pub async fn run(
     max_client_batch_size: usize,
     usage_stats_level: usage_stats::UsageStatsLevel,
     payload_limit: usize,
+    prometheus_port: u16,
 ) -> Result<(), WebServerError> {
     // CORS allowed origins
     // map to go inside the option and then map to parse from String to HeaderValue
@@ -1825,6 +1826,7 @@ pub async fn run(
         compat_return_full_text,
         allow_origin,
         payload_limit,
+        prometheus_port,
     )
     .await;
 
@@ -1886,6 +1888,7 @@ async fn start(
     compat_return_full_text: bool,
     allow_origin: Option<AllowOrigin>,
     payload_limit: usize,
+    prometheus_port: u16,
 ) -> Result<(), WebServerError> {
     // Determine the server port based on the feature and environment variable.
     let port = if cfg!(feature = "google") {
@@ -1959,8 +1962,12 @@ async fn start(
     // let skipped_matcher = Matcher::Full(String::from("tgi_request_skipped_tokens"));
     // let skipped_buckets: Vec<f64> = (0..shard_info.speculate + 1).map(|x| x as f64).collect();
 
+    let mut p_addr = addr;
+    p_addr.set_port(prometheus_port);
+
     // Prometheus handler
     let builder = PrometheusBuilder::new()
+        .with_http_listener(p_addr)
         .set_buckets_for_metric(duration_matcher, &duration_buckets)
         .unwrap()
         .set_buckets_for_metric(input_length_matcher, &input_length_buckets)