fix(metrics): serve the process collector's memory and CPU metrics

Cargo.toml

@@ -95,7 +95,7 @@ spawned-rt = "0.5.0"
 tokio = "1.0"
 tokio-util = "0.7"
 
-prometheus = "0.14"
+prometheus = { version = "0.14", features = ["process"] }
 
 clap = { version = "4.3", features = ["derive", "env"] }
 

crates/common/metrics/src/gather.rs

@@ -27,3 +27,21 @@ pub fn gather_default_metrics() -> Result<String, GatherError> {
 
     Ok(res)
 }
+
+/// The process collector comes from the `prometheus` crate's `process`
+/// feature, which registers it with the default registry on Linux only.
+/// Dashboards read the node's memory from it, so dropping the feature would
+/// leave them blank without any build failing.
+#[cfg(all(test, target_os = "linux"))]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn the_default_registry_serves_the_process_metrics() {
+        let metrics = gather_default_metrics().unwrap();
+        assert!(
+            metrics.contains("process_resident_memory_bytes"),
+            "{metrics}"
+        );
+    }
+}

docs/metrics.md

@@ -13,6 +13,13 @@ The Lean Chain's metrics follow [the leanMetrics specification](https://github.c
 | `lean_node_info` | Gauge | Node information (always 1) | On node start | name, version | ✅ |
 | `lean_node_start_time_seconds` | Gauge | Start timestamp | On node start | | ✅ |
 
+On Linux the endpoint also serves the `prometheus` crate's process collector,
+read from `/proc/self` at each scrape: `process_resident_memory_bytes`,
+`process_virtual_memory_bytes`, `process_cpu_seconds_total`,
+`process_open_fds`, `process_max_fds`, `process_threads` and
+`process_start_time_seconds`. They are not part of the leanMetrics spec, and
+other platforms do not export them.
+
 
 ## PQ Signature Metrics