[Breaking] Mark prometheus config as experimental (#473)

Since prometheus has known issues, in prep for the release, we are flagging this config as experimental until we can fix the bugs.
TraceMachina · Dec 12, 2023 · 931e721 · 931e721
1 parent 3329d7c
commit 931e721
Show file tree

Hide file tree

Showing 9 changed files with 12 additions and 12 deletions.
diff --git a/deployment-examples/docker-compose/local-storage-cas.json b/deployment-examples/docker-compose/local-storage-cas.json
@@ -65,7 +65,7 @@
     // Only publish metrics on a private port.
     "listen_address": "0.0.0.0:50061",
     "services": {
-      "prometheus": {
+      "experimental_prometheus": {
         "path": "/metrics"
       }
     }

diff --git a/deployment-examples/terraform/AWS/scripts/scheduler.json b/deployment-examples/terraform/AWS/scripts/scheduler.json
@@ -143,7 +143,7 @@
   }, {
     "listen_address": "0.0.0.0:50061",
     "services": {
-      "prometheus": {
+      "experimental_prometheus": {
         "path": "/metrics"
       },
       // Note: This should be served on a different port, because it has

diff --git a/deployment-examples/terraform/GCP/module/scripts/browser_proxy.json b/deployment-examples/terraform/GCP/module/scripts/browser_proxy.json
@@ -42,7 +42,7 @@
     // an external load balancer.
     "listen_address": "0.0.0.0:50052",
     "services": {
-      "prometheus": {
+      "experimental_prometheus": {
         "path": "/metrics"
       },
       "cas": {

diff --git a/deployment-examples/terraform/GCP/module/scripts/cas.json b/deployment-examples/terraform/GCP/module/scripts/cas.json
@@ -120,7 +120,7 @@
       "http2_keep_alive_interval": 10
     },
     "services": {
-      "prometheus": {
+      "experimental_prometheus": {
         "path": "/metrics"
       },
       "cas": {

diff --git a/deployment-examples/terraform/GCP/module/scripts/scheduler.json b/deployment-examples/terraform/GCP/module/scripts/scheduler.json
@@ -122,7 +122,7 @@
       "http2_keep_alive_interval": 10
     },
     "services": {
-      "prometheus": {
+      "experimental_prometheus": {
         "path": "/metrics"
       },
       "ac": {

diff --git a/deployment-examples/terraform/GCP/module/scripts/worker.json b/deployment-examples/terraform/GCP/module/scripts/worker.json
@@ -96,7 +96,7 @@
       "http2_keep_alive_interval": 10
     },
     "services": {
-      "prometheus": {
+      "experimental_prometheus": {
         "path": "/metrics"
       }
     }

diff --git a/nativelink-config/examples/basic_cas.json b/nativelink-config/examples/basic_cas.json
@@ -128,7 +128,7 @@
     "name": "private_workers_servers",
     "listen_address": "0.0.0.0:50061",
     "services": {
-      "prometheus": {
+      "experimental_prometheus": {
         "path": "/metrics"
       },
       // Note: This should be served on a different port, because it has

diff --git a/nativelink-config/src/cas_server.rs b/nativelink-config/src/cas_server.rs
@@ -198,9 +198,9 @@ pub struct ServicesConfig {
     /// that makes the remote execution/cache requests.
     pub worker_api: Option<WorkerApiConfig>,
 
-    /// Prometheus metrics configuration. Metrics are gathered as a singleton
-    /// but may be served on multiple endpoints.
-    pub prometheus: Option<PrometheusConfig>,
+    /// Experimental - Prometheus metrics configuration. Metrics are gathered
+    /// as a singleton but may be served on multiple endpoints.
+    pub experimental_prometheus: Option<PrometheusConfig>,
 
     /// This is the service for any administrative tasks.
     /// It provides a REST API endpoint for administrative purposes.

diff --git a/src/bin/cas.rs b/src/bin/cas.rs
@@ -350,7 +350,7 @@ async fn inner_main(cfg: CasConfig, server_start_timestamp: u64) -> Result<(), B
             // This is a generic endpoint used to check if the server is up.
             .route_service("/status", axum::routing::get(move || async move { "Ok".to_string() }));
 
-        if let Some(prometheus_cfg) = services.prometheus {
+        if let Some(prometheus_cfg) = services.experimental_prometheus {
             fn error_to_response<E: std::error::Error>(e: E) -> Response<String> {
                 let mut response = Response::new(format!("Error: {e:?}"));
                 *response.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
@@ -683,7 +683,7 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
                     let Some(service) = &v.services else {
                         return true;
                     };
-                    service.prometheus.is_none()
+                    service.experimental_prometheus.is_none()
                 }),
                 default_digest_hash_function: None,
             }