feat: improve metrics

Author: keroro520

Diff for: Cargo.lock

@@ -19,6 +19,7 @@ members = [
     "reqpool",
     "reqactor",
     "ballot",
+    "metrics",
 ]
 
 # Always optimize; building and running the guest takes much longer without optimization.
@@ -45,6 +46,7 @@ raiko-redis-derive = { path = "./redis-derive" }
 raiko-reqpool = { path = "./reqpool" }
 raiko-reqactor = { path = "./reqactor" }
 raiko-ballot = { path = "./ballot" }
+raiko-metrics = { path = "./metrics" }
 
 # reth
 reth-primitives = { git = "https://github.com/taikoxyz/taiko-reth.git", branch = "v1.0.0-rc.2-taiko", default-features = false, features = [

Diff for: Cargo.toml

@@ -18,6 +18,8 @@ services:
       - "3000:3000"
     volumes:
       - 'grafana_storage:/var/lib/grafana'
+    depends_on:
+      - prometheus
 volumes:
   grafana_storage: {}
   prometheus_storage: {}

Diff for: docker/docker-compose.metrics.yml

@@ -18,6 +18,7 @@ raiko-tasks = { workspace = true }
 raiko-reqpool = { workspace = true }
 raiko-reqactor = { workspace = true }
 raiko-ballot = { workspace = true }
+raiko-metrics = { workspace = true }
 
 # alloy
 alloy-rlp = { workspace = true }
@@ -72,6 +73,7 @@ url = { workspace = true }
 cfg-if = { workspace = true }
 cap = { workspace = true }
 dotenv = { workspace = true }
+chrono = { workspace = true }
 
 
 # reth

Diff for: host/Cargo.toml

@@ -15,7 +15,6 @@ use crate::interfaces::HostResult;
 
 pub mod cache;
 pub mod interfaces;
-pub mod metrics;
 pub mod server;
 
 #[derive(Default, Clone, Serialize, Deserialize, Debug, Parser)]

Diff for: host/src/lib.rs

@@ -13,6 +13,7 @@ use tower_http::{
 };
 
 pub mod admin;
+pub mod metrics;
 pub mod v1;
 pub mod v2;
 pub mod v3;
@@ -39,12 +40,14 @@ pub fn create_router(concurrency_limit: usize, jwt_secret: Option<&str>) -> Rout
     let v2_api = v2::create_router();
     let v3_api = v3::create_router();
     let admin_api = admin::create_router();
+    let metrics_api = metrics::create_router();
     let router = Router::new()
         .nest("/v1", v1_api)
         .nest("/v2", v2_api)
         .nest("/v3", v3_api.clone())
         .merge(v3_api)
         .nest("/admin", admin_api)
+        .nest("/metrics", metrics_api)
         .layer(middleware)
         .layer(DefaultBodyLimit::max(MAX_BODY_SIZE))
         .layer(trace)

Diff for: host/src/metrics.rs

@@ -11,7 +11,6 @@ use crate::interfaces::HostError;
 use raiko_reqactor::Actor;
 
 pub mod health;
-pub mod metrics;
 pub mod proof;
 
 #[derive(OpenApi)]
@@ -43,7 +42,6 @@ pub mod proof;
     tags(
         (name = "Proving", description = "Routes that handle proving requests"),
         (name = "Health", description = "Routes that report the server health status"),
-        (name = "Metrics", description = "Routes that give detailed insight into the server")
     )
 )]
 /// The root API struct which is generated from the `OpenApi` derive macro.
@@ -105,31 +103,24 @@ pub struct GuestOutputDoc {
 
 #[must_use]
 pub fn create_docs() -> utoipa::openapi::OpenApi {
-    [
-        health::create_docs(),
-        metrics::create_docs(),
-        proof::create_docs(),
-    ]
-    .into_iter()
-    .fold(Docs::openapi(), |mut doc, sub_doc| {
-        doc.merge(sub_doc);
-        doc
-    })
+    [health::create_docs(), proof::create_docs()]
+        .into_iter()
+        .fold(Docs::openapi(), |mut doc, sub_doc| {
+            doc.merge(sub_doc);
+            doc
+        })
 }
 
 pub fn create_router(concurrency_limit: usize) -> Router<Actor> {
     let docs = create_docs();
 
     Router::new()
-        // Only add the concurrency limit to the proof route. We want to still be able to call
-        // healthchecks and metrics to have insight into the system.
         .nest(
             "/proof",
             proof::create_router()
                 .layer(ServiceBuilder::new().concurrency_limit(concurrency_limit)),
         )
         .nest("/health", health::create_router())
-        .nest("/metrics", metrics::create_router())
         .merge(SwaggerUi::new("/swagger-ui").url("/api-docs/openapi.json", docs.clone()))
         .merge(Scalar::with_url("/scalar", docs))
 }

Diff for: host/src/server/api/v1/metrics.rs renamed to host/src/server/api/metrics.rs

@@ -44,7 +44,6 @@ pub mod proof;
     tags(
         (name = "Proving", description = "Routes that handle proving requests"),
         (name = "Health", description = "Routes that report the server health status"),
-        (name = "Metrics", description = "Routes that give detailed insight into the server")
     )
 )]
 /// The root API struct which is generated from the `OpenApi` derive macro.
@@ -119,29 +118,22 @@ impl IntoResponse for PruneStatus {
 
 #[must_use]
 pub fn create_docs() -> utoipa::openapi::OpenApi {
-    [
-        v1::health::create_docs(),
-        v1::metrics::create_docs(),
-        proof::create_docs(),
-    ]
-    .into_iter()
-    .fold(Docs::openapi(), |mut doc, sub_doc| {
-        doc.merge(sub_doc);
-        doc
-    })
+    [v1::health::create_docs(), proof::create_docs()]
+        .into_iter()
+        .fold(Docs::openapi(), |mut doc, sub_doc| {
+            doc.merge(sub_doc);
+            doc
+        })
 }
 
 pub fn create_router() -> Router<Actor> {
     let docs = create_docs();
 
     Router::new()
-        // Only add the concurrency limit to the proof route. We want to still be able to call
-        // healthchecks and metrics to have insight into the system.
         .nest("/proof", proof::create_router())
         // TODO: Separate task or try to get it into /proof somehow? Probably separate
         .nest("/aggregate", proof::create_router())
         .nest("/health", v1::health::create_router())
-        .nest("/metrics", v1::metrics::create_router())
         .merge(SwaggerUi::new("/swagger-ui").url("/api-docs/openapi.json", docs.clone()))
         .merge(Scalar::with_url("/scalar", docs))
 }