jason/socktop
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

perf(agent,non-linux): add per-process CPU scaling heuristic to reduce overreporting

Browse Source
This commit is contained in:
jasonwitty 2025-08-25 23:47:31 -07:00
parent beddba0072
commit e2dc5e8ac9
1 changed files with 21 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
socktop_agent/src/metrics.rs
View File

@ -444,11 +444,13 @@ pub async fn collect_processes_all(state: &AppState) -> ProcessesPayload {
// Sleep briefly to allow cpu deltas to accumulate; 200-250ms is typical; we keep 200ms to lower agent overhead. // Sleep briefly to allow cpu deltas to accumulate; 200-250ms is typical; we keep 200ms to lower agent overhead.
sleep(Duration::from_millis(delay_ms.min(500))).await; sleep(Duration::from_millis(delay_ms.min(500))).await;
// Second refresh: only CPU counters (lighter than full everything) to reduce overhead. // Second refresh: only CPU counters (lighter than full everything) to reduce overhead.
let (total_count, procs) = { let (total_count, mut procs) = {
let mut sys = state.sys.lock().await; let mut sys = state.sys.lock().await;
// Build a lightweight refresh kind: only CPU times. // Build a lightweight refresh kind: only CPU times.
let cpu_only = ProcessRefreshKind::nothing().with_cpu(); let cpu_only = ProcessRefreshKind::nothing().with_cpu();
sys.refresh_processes_specifics(ProcessesToUpdate::All, false, cpu_only); sys.refresh_processes_specifics(ProcessesToUpdate::All, false, cpu_only);
// Refresh global CPU usage once for scaling heuristic
sys.refresh_cpu_usage();
let total_count = sys.processes().len(); let total_count = sys.processes().len();
let norm = normalize_cpu_enabled(); let norm = normalize_cpu_enabled();
let cores = if norm { let cores = if norm {
@ -456,7 +458,7 @@ pub async fn collect_processes_all(state: &AppState) -> ProcessesPayload {
} else { } else {
1.0 1.0
}; };
let list: Vec<ProcessInfo> = sys let mut list: Vec<ProcessInfo> = sys
.processes() .processes()
.values() .values()
.map(|p| { .map(|p| {
@ -474,6 +476,23 @@ pub async fn collect_processes_all(state: &AppState) -> ProcessesPayload {
} }
}) })
.collect(); .collect();
// Automatic scaling (enabled by default): if sum of per-process CPU exceeds global
// CPU by >25%, scale all process CPU values proportionally so the sum matches global.
if std::env::var("SOCKTOP_AGENT_SCALE_PROC_CPU")
.map(|v| v != "0")
.unwrap_or(true)
{
let sum: f32 = list.iter().map(|p| p.cpu_usage).sum();
let global = sys.global_cpu_usage();
if sum > 0.0 && global > 0.0 {
let scale = global / sum;
if scale < 0.75 { // only scale if we're at least 25% over
for p in &mut list {
p.cpu_usage = (p.cpu_usage * scale).clamp(0.0, 100.0);
}
}
}
}
(total_count, list) (total_count, list)
}; };
let payload = ProcessesPayload { let payload = ProcessesPayload {