eoc/eoc-rs/crates/eoc-meter/src/lib.rs

@@ -463,6 +463,10 @@ mod rail_liveness {
         pub(crate) opened: bool,
         /// Ticks run between samples.
         pub(crate) ticks: u32,
+        /// Each tick's wall time, in order. Recorded, never decided on: a test
+        /// reads it to tell a tick that its wall cap cut short, on a host that
+        /// gave the thread almost no CPU, from one that did not burn at all.
+        pub(crate) tick_walls: Vec<Duration>,
         /// Every sample call, in order.
         pub(crate) calls: Vec<SampleCall>,
         /// The rule, folded over every sample that was taken.
@@ -493,9 +497,9 @@ mod rail_liveness {
         S: FnMut() -> Option<Sample>,
     {
         let Some(mut sampler) = open() else {
-            return ProbeRun { opened: false, ticks: 0, calls: Vec::new(), watch: RailWatch::default() };
+            return ProbeRun { opened: false, ticks: 0, tick_walls: Vec::new(), calls: Vec::new(), watch: RailWatch::default() };
         };
-        let (mut calls, mut ticks) = (Vec::new(), 0u32);
+        let (mut calls, mut ticks, mut tick_walls) = (Vec::new(), 0u32, Vec::new());
         let watch = watch(
             || {
                 let (cpu_before, started) = (thread_cpu(), wall());
@@ -512,12 +516,14 @@ mod rail_liveness {
             },
             || {
                 ticks += 1;
+                let at = wall();
                 tick();
+                tick_walls.push(wall().saturating_sub(at));
             },
             &wall,
             budget,
         );
-        ProbeRun { opened: true, ticks, calls, watch }
+        ProbeRun { opened: true, ticks, tick_walls, calls, watch }
     }
 
     /// The status a probe verdict gives this host. No verdict — no
@@ -3142,7 +3148,26 @@ mod macos_ioreport {
             );
             assert!(ticks.len() >= 2, "the probe made only {} sample calls", run.calls.len());
             assert_eq!(run.ticks as usize, ticks.len(), "every tick runs between two sample calls");
+            assert_eq!(run.tick_walls.len(), ticks.len(), "every tick's wall time is recorded");
             for (i, t) in ticks.iter().enumerate() {
+                // A tick may burn less only where its wall cap cut it short: a
+                // host loaded past a hundred runnable threads gave this thread
+                // under 2% of a core for 500 ms (9.6 ms of CPU, load average
+                // 101). A tick that stopped short of both its CPU and its wall
+                // time waited or skipped, and fails.
+                // A tick cut short still spun whenever it ran, so it burned
+                // some CPU. A tick that slept through its 500 ms burned next to
+                // none, and fails either way.
+                if run.tick_walls[i] >= crate::clocks::PROBE_TICK_WALL_CAP {
+                    eprintln!("ioreport probe tick {i}: {t:?} of CPU, cut short by its {:?} wall cap", run.tick_walls[i]);
+                    assert!(
+                        *t >= TICK / 25,
+                        "tick {i} ran its whole {:?} wall cap and burned only {t:?} of this thread's CPU: it waited, \
+                         not spun",
+                        run.tick_walls[i]
+                    );
+                    continue;
+                }
                 assert!(
                     *t >= TICK / 2,
                     "tick {i} burned only {t:?} of this thread's CPU (every tick: {ticks:?}): the probe must burn \

jouleclaw/jouleclaw-rs/crates/jouleclaw-energy/tests/process_cpu.rs

@@ -102,7 +102,11 @@ fn the_process_counter_meters_this_process_and_nothing_else() {
 
     // 5. A child process's work is not attributed: a child burns CPU while this
     //    process sleeps, and is killed and reaped inside the window. The child's
-    //    own counter, read independently, shows the work happened.
+    //    own counter, read independently, shows the work happened. The child
+    //    runs until its own counter has read 0.2 J (at least 300 ms, at most
+    //    5 s), not for a fixed 300 ms: on a heavily loaded host a fixed 300 ms
+    //    got the child too little CPU (0.076 J at a load average of 170), and
+    //    the test failed on the host's load, not on the meter.
     let mut child = Command::new("/bin/sh")
         .args(["-c", "while :; do :; done"])
         .stdout(Stdio::null())
@@ -111,16 +115,24 @@ fn the_process_counter_meters_this_process_and_nothing_else() {
         .expect("spawn a busy child");
     let pid = child.id() as libc::pid_t;
     let child_before = energy_nj_of(pid);
+    let started = Instant::now();
     let (child_burned_nj, parent) = counter.measure(|| {
-        std::thread::sleep(Duration::from_millis(300));
-        let burned = energy_nj_of(pid) - child_before;
+        let mut burned = 0;
+        while started.elapsed() < Duration::from_secs(5) {
+            std::thread::sleep(Duration::from_millis(50));
+            burned = energy_nj_of(pid) - child_before;
+            if burned >= 200_000_000 && started.elapsed() >= Duration::from_millis(300) {
+                break;
+            }
+        }
         child.kill().expect("kill the child");
         child.wait().expect("reap the child");
         burned
     });
     println!(
-        "process-cpu: child burned {:.3} J in 300 ms; this process's window = {parent:?} µJ",
-        child_burned_nj as f64 / 1e9
+        "process-cpu: child burned {:.3} J in {:?}; this process's window = {parent:?} µJ",
+        child_burned_nj as f64 / 1e9,
+        started.elapsed()
     );
     assert!(child_burned_nj >= 200_000_000, "the child did not burn: {child_burned_nj} nJ");
     match parent {