fix: resolve critical performance issues and UI hangs by optimizing polling and PSS collection

This commit is contained in:
2026-02-23 01:47:13 +01:00
parent e6c5adca5e
commit 9ff0855102
3 changed files with 112 additions and 74 deletions

View File

@@ -17,5 +17,4 @@ serde_json = "1.0.149"
sysinfo = "0.38.2" sysinfo = "0.38.2"
tauri = "2.10.2" tauri = "2.10.2"
tokio = { version = "1.49.0", features = ["full"] } tokio = { version = "1.49.0", features = ["full"] }
rayon = "1.10"
clap = { version = "4.5", features = ["derive"] } clap = { version = "4.5", features = ["derive"] }

View File

@@ -8,10 +8,9 @@ use std::sync::Mutex;
use std::process::Command; use std::process::Command;
use tauri::State; use tauri::State;
use serde::{Serialize, Deserialize}; use serde::{Serialize, Deserialize};
use std::collections::HashMap; use std::collections::{HashMap, HashSet};
use chrono::{DateTime, Utc}; use chrono::{DateTime, Utc};
use std::fs; use std::fs;
use rayon::prelude::*;
use clap::Parser; use clap::Parser;
use std::path::PathBuf; use std::path::PathBuf;
use std::time::Duration; use std::time::Duration;
@@ -21,27 +20,16 @@ use std::time::Duration;
#[derive(Parser, Debug)] #[derive(Parser, Debug)]
#[command(author, version, about, long_about = None)] #[command(author, version, about, long_about = None)]
struct Cli { struct Cli {
/// Start profiling immediately without GUI
#[arg(short, long)] #[arg(short, long)]
headless: bool, headless: bool,
/// Duration of profiling in seconds (for headless mode)
#[arg(short, long, default_value_t = 30)] #[arg(short, long, default_value_t = 30)]
duration: u64, duration: u64,
/// Interval between snapshots in milliseconds
#[arg(short, long, default_value_t = 1000)] #[arg(short, long, default_value_t = 1000)]
interval: u64, interval: u64,
/// Output path for the JSON report
#[arg(short, long)] #[arg(short, long)]
output: Option<PathBuf>, output: Option<PathBuf>,
/// Open the GUI with the collected data after headless profiling
#[arg(short, long)] #[arg(short, long)]
gui: bool, gui: bool,
/// Open an existing JSON report file in the GUI
#[arg(short, long)] #[arg(short, long)]
file: Option<PathBuf>, file: Option<PathBuf>,
} }
@@ -88,6 +76,7 @@ struct AppState {
sys: Mutex<System>, sys: Mutex<System>,
profiling: Mutex<ProfilingSession>, profiling: Mutex<ProfilingSession>,
initial_report: Mutex<Option<Report>>, initial_report: Mutex<Option<Report>>,
pss_cache: Mutex<HashMap<u32, (u64, DateTime<Utc>)>>,
} }
// --- Report Structures --- // --- Report Structures ---
@@ -151,36 +140,68 @@ fn get_pss(pid: u32) -> Option<u64> {
None None
} }
fn is_syspulse_recursive(pid: u32, self_pid: u32, sys: &System) -> bool { fn get_syspulse_pids(self_pid: u32, sys: &System) -> HashSet<u32> {
if pid == self_pid { return true; } let mut res = HashSet::new();
let mut current = sys.process(Pid::from_u32(pid)); res.insert(self_pid);
while let Some(proc) = current { let mut children_map: HashMap<u32, Vec<u32>> = HashMap::new();
if let Some(ppid) = proc.parent() { for (pid, p) in sys.processes() {
if ppid.as_u32() == self_pid { return true; } if let Some(ppid) = p.parent() {
current = sys.process(ppid); children_map.entry(ppid.as_u32()).or_default().push(pid.as_u32());
} else {
break;
} }
} }
false let mut stack = vec![self_pid];
while let Some(pid) = stack.pop() {
if let Some(children) = children_map.get(&pid) {
for &c in children {
if res.insert(c) {
stack.push(c);
}
}
}
}
res
} }
fn collect_snapshot(sys: &mut System, self_pid: u32) -> Snapshot { fn collect_snapshot(
sys: &mut System,
self_pid: u32,
syspulse_pids: &HashSet<u32>,
pss_cache: &mut HashMap<u32, (u64, DateTime<Utc>)>,
collect_pss: bool
) -> Snapshot {
sys.refresh_cpu_all(); sys.refresh_cpu_all();
sys.refresh_memory(); sys.refresh_memory();
sys.refresh_processes(sysinfo::ProcessesToUpdate::All, true); sys.refresh_processes(sysinfo::ProcessesToUpdate::All, true);
let cpu_usage: Vec<f32> = sys.cpus().iter().map(|cpu| cpu.cpu_usage()).collect(); let cpu_usage: Vec<f32> = sys.cpus().iter().map(|cpu| cpu.cpu_usage()).collect();
let used_memory = sys.used_memory(); let used_memory = sys.used_memory();
let now = Utc::now();
let processes: Vec<ProcessStats> = sys.processes().iter() let processes: Vec<ProcessStats> = sys.processes().iter()
.par_bridge() .map(|(pid, p)| {
.filter_map(|(pid, p)| {
let rss = p.memory();
if rss == 0 { return None; }
let pid_u32 = pid.as_u32(); let pid_u32 = pid.as_u32();
let memory = if rss > 10 * 1024 * 1024 { get_pss(pid_u32).unwrap_or(rss) } else { rss }; let rss = p.memory();
Some(ProcessStats { let is_syspulse = syspulse_pids.contains(&pid_u32);
let mut memory = rss;
// PSS collection is expensive. Throttle it heavily.
if collect_pss && (rss > 20 * 1024 * 1024 || is_syspulse) {
let needs_refresh = match pss_cache.get(&pid_u32) {
Some((_, last)) => (now - *last).num_seconds() > 5,
None => true,
};
if needs_refresh {
if let Some(pss) = get_pss(pid_u32) {
pss_cache.insert(pid_u32, (pss, now));
memory = pss;
}
} else if let Some((cached_pss, _)) = pss_cache.get(&pid_u32) {
memory = *cached_pss;
}
}
ProcessStats {
pid: pid_u32, pid: pid_u32,
parent_pid: p.parent().map(|pp| pp.as_u32()), parent_pid: p.parent().map(|pp| pp.as_u32()),
name: p.name().to_string_lossy().to_string(), name: p.name().to_string_lossy().to_string(),
@@ -188,12 +209,12 @@ fn collect_snapshot(sys: &mut System, self_pid: u32) -> Snapshot {
memory, memory,
status: format!("{:?}", p.status()), status: format!("{:?}", p.status()),
user_id: p.user_id().map(|u| u.to_string()), user_id: p.user_id().map(|u| u.to_string()),
is_syspulse: is_syspulse_recursive(pid_u32, self_pid, &sys), is_syspulse,
}) }
}).collect(); }).collect();
Snapshot { Snapshot {
timestamp: Utc::now(), timestamp: now,
cpu_usage, cpu_usage,
used_memory, used_memory,
processes, processes,
@@ -213,59 +234,51 @@ fn generate_report(start_time: DateTime<Utc>, snapshots: Vec<Snapshot>) -> Repor
} }
}).collect(); }).collect();
struct PidStats { let mut pid_map: HashMap<u32, (String, Option<u32>, Vec<ProcessHistoryPoint>, f32, f32, bool, bool)> = HashMap::new();
name: String,
history: Vec<ProcessHistoryPoint>,
peak_cpu: f32,
peak_mem: f32,
is_syspulse: bool,
is_zombie: bool,
}
let mut pid_map: HashMap<u32, PidStats> = HashMap::new();
let num_snapshots = snapshots.len() as f32; let num_snapshots = snapshots.len() as f32;
for snapshot in &snapshots { for snapshot in &snapshots {
for proc in &snapshot.processes { for proc in &snapshot.processes {
let entry = pid_map.entry(proc.pid).or_insert_with(|| PidStats { let entry = pid_map.entry(proc.pid).or_insert_with(|| (
name: proc.name.clone(), proc.name.clone(),
history: Vec::new(), proc.parent_pid,
peak_cpu: 0.0, Vec::new(),
peak_mem: 0.0, 0.0,
is_syspulse: proc.is_syspulse, 0.0,
is_zombie: false, proc.is_syspulse,
}); false
));
let mem_mb = proc.memory as f32 / 1024.0 / 1024.0; let mem_mb = proc.memory as f32 / 1024.0 / 1024.0;
entry.history.push(ProcessHistoryPoint { entry.2.push(ProcessHistoryPoint {
time: snapshot.timestamp.format("%H:%M:%S").to_string(), time: snapshot.timestamp.format("%H:%M:%S").to_string(),
cpu_usage: proc.cpu_usage, cpu_usage: proc.cpu_usage,
memory_mb: mem_mb, memory_mb: mem_mb,
}); });
if proc.cpu_usage > entry.peak_cpu { entry.peak_cpu = proc.cpu_usage; } if proc.cpu_usage > entry.3 { entry.3 = proc.cpu_usage; }
if mem_mb > entry.peak_mem { entry.peak_mem = mem_mb; } if mem_mb > entry.4 { entry.4 = mem_mb; }
if proc.status.contains("Zombie") { entry.is_zombie = true; } if proc.status.contains("Zombie") { entry.6 = true; }
} }
} }
let mut nodes: HashMap<u32, AggregatedProcess> = pid_map.into_iter().map(|(pid, stats)| { let mut nodes: HashMap<u32, AggregatedProcess> = pid_map.into_iter().map(|(pid, (name, _, history, peak_cpu, peak_mem, is_syspulse, is_zombie))| {
let total_cpu: f32 = stats.history.iter().map(|h| h.cpu_usage).sum(); let total_cpu: f32 = history.iter().map(|h| h.cpu_usage).sum();
let total_mem: f32 = stats.history.iter().map(|h| h.memory_mb).sum(); let total_mem: f32 = history.iter().map(|h| h.memory_mb).sum();
let mut warnings = Vec::new(); let mut warnings = Vec::new();
if stats.is_zombie { warnings.push("Zombie".to_string()); } if is_zombie { warnings.push("Zombie".to_string()); }
if stats.peak_cpu > 80.0 { warnings.push("High Peak".to_string()); } if peak_cpu > 80.0 { warnings.push("High Peak".to_string()); }
(pid, AggregatedProcess { (pid, AggregatedProcess {
pid, pid,
name: stats.name, name,
avg_cpu: total_cpu / num_snapshots, avg_cpu: total_cpu / num_snapshots,
peak_cpu: stats.peak_cpu, peak_cpu,
avg_memory_mb: total_mem / num_snapshots, avg_memory_mb: total_mem / num_snapshots,
peak_memory_mb: stats.peak_mem, peak_memory_mb: peak_mem,
inclusive_avg_cpu: 0.0, inclusive_avg_cpu: 0.0,
inclusive_avg_memory_mb: 0.0, inclusive_avg_memory_mb: 0.0,
instance_count: 1, instance_count: 1,
warnings, warnings,
history: stats.history, history,
is_syspulse: stats.is_syspulse, is_syspulse,
children: Vec::new(), children: Vec::new(),
}) })
}).collect(); }).collect();
@@ -337,8 +350,14 @@ fn generate_report(start_time: DateTime<Utc>, snapshots: Vec<Snapshot>) -> Repor
fn get_system_stats(state: State<AppState>, minimal: bool) -> SystemStats { fn get_system_stats(state: State<AppState>, minimal: bool) -> SystemStats {
let mut sys = state.sys.lock().unwrap(); let mut sys = state.sys.lock().unwrap();
let mut profiling = state.profiling.lock().unwrap(); let mut profiling = state.profiling.lock().unwrap();
let mut pss_cache = state.pss_cache.lock().unwrap();
let snapshot = collect_snapshot(&mut sys, std::process::id()); let self_pid = std::process::id();
let syspulse_pids = get_syspulse_pids(self_pid, &sys);
// NO PSS collection during live dashboard. ONLY during profiling or after.
// This is the primary cause of high CPU/I/O.
let snapshot = collect_snapshot(&mut sys, self_pid, &syspulse_pids, &mut pss_cache, profiling.is_active);
if profiling.is_active { if profiling.is_active {
profiling.snapshots.push(snapshot.clone()); profiling.snapshots.push(snapshot.clone());
@@ -390,7 +409,8 @@ fn start_profiling(state: State<AppState>) {
fn stop_profiling(state: State<AppState>) -> Report { fn stop_profiling(state: State<AppState>) -> Report {
let mut profiling = state.profiling.lock().unwrap(); let mut profiling = state.profiling.lock().unwrap();
profiling.is_active = false; profiling.is_active = false;
generate_report(profiling.start_time.unwrap_or(Utc::now()), profiling.snapshots.drain(..).collect()) let snapshots: Vec<Snapshot> = profiling.snapshots.drain(..).collect();
generate_report(profiling.start_time.unwrap_or(Utc::now()), snapshots)
} }
#[tauri::command] #[tauri::command]
@@ -423,12 +443,14 @@ fn main() {
if cli.headless { if cli.headless {
println!("⚡ SysPulse: Starting headless profiling for {}s (interval: {}ms)...", cli.duration, cli.interval); println!("⚡ SysPulse: Starting headless profiling for {}s (interval: {}ms)...", cli.duration, cli.interval);
let mut sys = System::new_all(); let mut sys = System::new_all();
let mut pss_cache = HashMap::new();
let start_time = Utc::now(); let start_time = Utc::now();
let mut snapshots = Vec::new(); let mut snapshots = Vec::new();
let self_pid = std::process::id(); let self_pid = std::process::id();
for i in 0..(cli.duration * 1000 / cli.interval) { for i in 0..(cli.duration * 1000 / cli.interval) {
snapshots.push(collect_snapshot(&mut sys, self_pid)); let syspulse_pids = get_syspulse_pids(self_pid, &sys);
snapshots.push(collect_snapshot(&mut sys, self_pid, &syspulse_pids, &mut pss_cache, true));
std::thread::sleep(Duration::from_millis(cli.interval)); std::thread::sleep(Duration::from_millis(cli.interval));
if (i + 1) % (1000 / cli.interval) == 0 { if (i + 1) % (1000 / cli.interval) == 0 {
println!(" Progress: {}/{}s", (i + 1) * cli.interval / 1000, cli.duration); println!(" Progress: {}/{}s", (i + 1) * cli.interval / 1000, cli.duration);
@@ -457,6 +479,7 @@ fn main() {
snapshots: Vec::new(), snapshots: Vec::new(),
}), }),
initial_report: Mutex::new(initial_report), initial_report: Mutex::new(initial_report),
pss_cache: Mutex::new(HashMap::new()),
}) })
.invoke_handler(tauri::generate_handler![ .invoke_handler(tauri::generate_handler![
get_system_stats, get_system_stats,

View File

@@ -91,8 +91,8 @@ function App() {
const [history, setHistory] = useState<{ time: string; cpu: number }[]>([]); const [history, setHistory] = useState<{ time: string; cpu: number }[]>([]);
const [report, setReport] = useState<ProfilingReport | null>(null); const [report, setReport] = useState<ProfilingReport | null>(null);
// Initial report check
useEffect(() => { useEffect(() => {
// Check for initial report (from CLI args)
const checkInitialReport = async () => { const checkInitialReport = async () => {
try { try {
const data = await invoke<ProfilingReport | null>('get_initial_report'); const data = await invoke<ProfilingReport | null>('get_initial_report');
@@ -105,13 +105,23 @@ function App() {
} }
}; };
checkInitialReport(); checkInitialReport();
}, []);
// Stats Polling
useEffect(() => {
let timeoutId: number;
let isMounted = true;
const fetchStats = async () => { const fetchStats = async () => {
if (view !== 'dashboard' && !stats?.is_recording) return;
try { try {
const isRecording = stats?.is_recording ?? false;
const data = await invoke<SystemStats>('get_system_stats', { const data = await invoke<SystemStats>('get_system_stats', {
minimal: isRecording || view === 'report' minimal: (stats?.is_recording || view === 'report')
}); });
if (!isMounted) return;
setStats(data); setStats(data);
const avgCpu = data.cpu_usage.reduce((a, b) => a + b, 0) / data.cpu_usage.length; const avgCpu = data.cpu_usage.reduce((a, b) => a + b, 0) / data.cpu_usage.length;
@@ -123,12 +133,18 @@ function App() {
}); });
} catch (e) { } catch (e) {
console.error(e); console.error(e);
} finally {
if (isMounted) {
timeoutId = window.setTimeout(fetchStats, 1000);
}
} }
}; };
fetchStats(); fetchStats();
const interval = setInterval(fetchStats, 1000); return () => {
return () => clearInterval(interval); isMounted = false;
clearTimeout(timeoutId);
};
}, [view, stats?.is_recording]); }, [view, stats?.is_recording]);
const toggleRecording = async () => { const toggleRecording = async () => {