high-resolution CPU usage

libs/metrics/src/more_process_metrics.rs

@@ -2,14 +2,28 @@
 
 // This module has heavy inspiration from the prometheus crate's `process_collector.rs`.
 
+use once_cell::sync::Lazy;
+use prometheus::Gauge;
+
 use crate::UIntGauge;
 
 pub struct Collector {
     descs: Vec<prometheus::core::Desc>,
     vmlck: crate::UIntGauge,
+    cpu_seconds_highres: Gauge,
 }
 
-const NMETRICS: usize = 1;
+const NMETRICS: usize = 2;
+
+static CLK_TCK_F64: Lazy<f64> = Lazy::new(|| {
+    let long = unsafe { libc::sysconf(libc::_SC_CLK_TCK) };
+    if long == -1 {
+        panic!("sysconf(_SC_CLK_TCK) failed");
+    }
+    let convertible_to_f64: i32 =
+        i32::try_from(long).expect("sysconf(_SC_CLK_TCK) is larger than i32");
+    convertible_to_f64 as f64
+});
 
 impl prometheus::core::Collector for Collector {
     fn desc(&self) -> Vec<&prometheus::core::Desc> {
@@ -27,6 +41,12 @@ impl prometheus::core::Collector for Collector {
                 mfs.extend(self.vmlck.collect())
             }
         }
+        if let Ok(stat) = myself.stat() {
+            let cpu_seconds = stat.utime + stat.stime;
+            self.cpu_seconds_highres
+                .set(cpu_seconds as f64 / *CLK_TCK_F64);
+            mfs.extend(self.cpu_seconds_highres.collect());
+        }
         mfs
     }
 }
@@ -43,7 +63,23 @@ impl Collector {
                 .cloned(),
         );
 
-        Self { descs, vmlck }
+        let cpu_seconds_highres = Gauge::new(
+            "libmetrics_process_cpu_seconds_highres",
+            "Total user and system CPU time spent in seconds.\
+             Sub-second resolution, hence better than `process_cpu_seconds_total`.",
+        )
+        .unwrap();
+        descs.extend(
+            prometheus::core::Collector::desc(&cpu_seconds_highres)
+                .into_iter()
+                .cloned(),
+        );
+
+        Self {
+            descs,
+            vmlck,
+            cpu_seconds_highres,
+        }
     }
 }
 

test_runner/performance/pageserver/test_pageserver_getpage_merge.py

@@ -146,7 +146,7 @@ def test_getpage_merge_smoke(
                 ).value,
                 compute_getpage_count=compute_getpage_count,
                 pageserver_cpu_seconds_total=pageserver_metrics.query_one(
-                    "process_cpu_seconds_total"
+                    "libmetrics_process_cpu_seconds_highres"
                 ).value,
             )