diff options
| author | Lennard Kittner <lennard@kittner.dev> | 2025-03-30 21:08:01 +0200 |
|---|---|---|
| committer | Lennard Kittner <lennard@kittner.dev> | 2025-03-30 21:08:01 +0200 |
| commit | 9bf91697fcacc76761a5d4b8282624397c282553 (patch) | |
| tree | 3075d75ab1ed760b0605f35040ffcadbb8fed3b1 /src | |
| parent | cca7ffe56b28f8425b89aba867a5cd8ee85eccc3 (diff) | |
Update logging via socket
Diffstat (limited to 'src')
| -rw-r--r-- | src/energy.rs | 117 | ||||
| -rw-r--r-- | src/energy/budget.rs | 4 | ||||
| -rw-r--r-- | src/main.rs | 15 | ||||
| -rw-r--r-- | src/scheduler.rs | 10 | ||||
| -rw-r--r-- | src/socket.rs | 83 |
5 files changed, 81 insertions, 148 deletions
diff --git a/src/energy.rs b/src/energy.rs index 1b52c13..02aaf62 100644 --- a/src/energy.rs +++ b/src/energy.rs @@ -22,37 +22,30 @@ pub use trackers::{KernelDriver, PerfEstimator}; pub enum Request { NewTask(Pid, i32), RemoveTask(Pid), - LogEnergyTask(Pid), } #[derive(Clone)] pub struct ProcessInfo { - energy: f64, - tree_energy: f64, - last_update: std::time::Instant, - parent: Pid, - cpu: i32, -} - -pub struct EnergyLog { - pub energy_of_exited_tasks: f64, - pub energy_of_running_tasks: f64, + pub energy: f64, + pub tree_energy: f64, + pub last_update: std::time::Instant, + pub parent: Pid, + pub cpu: i32, } pub struct EnergyService { estimator: Box<dyn Estimator>, budget_policy: Option<Box<dyn BudgetPolicy>>, - // TODO: remove + // contains the same data as the keys of process_info but having this reduces contention and + // avoids unnecessary clone active_processes: BTreeSet<Pid>, - process_info: HashMap<Pid, ProcessInfo>, + process_info: Arc<RwLock<HashMap<Pid, ProcessInfo>>>, shared_budgets: Arc<DashMap<Pid, u64>>, request_receiver: mpsc::Receiver<Request>, update_interval: Duration, shared_cpu_frequency_ranges: Arc<RwLock<Vec<RangeInclusive<FrequencyKHZ>>>>, shared_policy_frequency_ranges: Arc<RwLock<Vec<RangeInclusive<FrequencyKHZ>>>>, shared_cpu_current_frequencies: Arc<RwLock<Vec<FrequencyKHZ>>>, - energy_logging_children_to_root_of_tree: HashMap<Pid, Pid>, - logged_energy: Arc<RwLock<HashMap<Pid, EnergyLog>>>, rapl_offset: f64, old_rapl: f64, bias: f64, @@ -63,27 +56,25 @@ impl EnergyService { pub fn new( estimator: Box<dyn Estimator>, budget_policy: Box<dyn BudgetPolicy>, + process_info: Arc<RwLock<HashMap<Pid, ProcessInfo>>>, shared_budgets: Arc<DashMap<Pid, u64>>, request_receiver: mpsc::Receiver<Request>, update_interval: Duration, shared_cpu_frequency_ranges: Arc<RwLock<Vec<RangeInclusive<FrequencyKHZ>>>>, shared_policy_frequency_ranges: Arc<RwLock<Vec<RangeInclusive<FrequencyKHZ>>>>, shared_cpu_current_frequencies: Arc<RwLock<Vec<FrequencyKHZ>>>, - logged_energy: Arc<RwLock<HashMap<Pid, EnergyLog>>>, ) -> Self { Self { estimator, budget_policy: Some(budget_policy), active_processes: BTreeSet::new(), - process_info: HashMap::new(), + process_info, shared_budgets, request_receiver, update_interval, shared_cpu_frequency_ranges, shared_policy_frequency_ranges, shared_cpu_current_frequencies, - energy_logging_children_to_root_of_tree: HashMap::new(), - logged_energy, rapl_offset: rapl::read_package_energy().unwrap(), old_rapl: 0., bias: 1., @@ -118,17 +109,16 @@ impl EnergyService { fn handle_request(&mut self, request: Request) { match request { Request::NewTask(pid, cpu) => { - if self.process_info.contains_key(&pid) { + if self.process_info.read().unwrap().contains_key(&pid) { return; } self.estimator.start_trace(pid as u64); - self.active_processes.insert(pid); let parent = (|| { let process = procfs::process::Process::new(pid)?; process.stat().map(|stat| stat.ppid) })() .unwrap_or_default(); - self.process_info.insert( + self.process_info.write().unwrap().insert( pid, ProcessInfo { energy: 0., @@ -138,9 +128,10 @@ impl EnergyService { cpu, }, ); + self.active_processes.insert(pid); // Initialize with default budget self.shared_budgets.insert(pid, u64::MAX); - if !self.process_info.contains_key(&parent) && parent != 0 { + if !self.process_info.read().unwrap().contains_key(&parent) && parent != 0 { self.handle_request(Request::NewTask(parent, cpu)); } } @@ -150,46 +141,28 @@ impl EnergyService { } self.estimator.stop_trace(pid as u64); - if let Some(root_pid) = self.energy_logging_children_to_root_of_tree.remove(&pid) { - if let Some(info) = self.process_info.remove(&pid) { - self.logged_energy - .write() - .unwrap() - .get_mut(&root_pid) - .unwrap() - .energy_of_exited_tasks += info.energy; - } - } - self.active_processes.remove(&pid); - self.process_info.remove(&pid); + self.process_info.write().unwrap().remove(&pid); + self.process_info.write().unwrap().remove(&pid); self.shared_budgets.remove(&pid); } - Request::LogEnergyTask(pid) => { - self.energy_logging_children_to_root_of_tree - .insert(pid, pid); - self.logged_energy.write().unwrap().insert( - pid, - EnergyLog { - energy_of_exited_tasks: 0., - energy_of_running_tasks: 0., - }, - ); - } } } fn update_measurements(&mut self) { - let mut logged_energy = self.logged_energy.write().unwrap(); - for (_, value) in logged_energy.iter_mut() { - value.energy_of_running_tasks = 0f64; - } - let old_energy = self.process_info.get(&1).unwrap().tree_energy; - for &pid in &self.active_processes { - if let Some(info) = self.process_info.get_mut(&pid) { - if let Some(energy) = self.estimator.read_consumption(pid as u64, info.cpu) { + let old_energy = self + .process_info + .read() + .unwrap() + .get(&1) + .unwrap() + .tree_energy; + for pid in &self.active_processes { + let mut process_info = self.process_info.write().unwrap(); + if let Some(info) = process_info.get_mut(&pid) { + if let Some(energy) = self.estimator.read_consumption(*pid as u64, info.cpu) { info.energy += energy * self.bias; let mut parent = info.parent; - while let Some(info) = self.process_info.get_mut(&parent) { + while let Some(info) = process_info.get_mut(&parent) { info.tree_energy += energy * self.bias; info.last_update = std::time::Instant::now(); parent = info.parent; @@ -197,33 +170,23 @@ impl EnergyService { } } } - if let Some(init) = self.process_info.get_mut(&1) { + if let Some(init) = self.process_info.write().unwrap().get_mut(&1) { let rapl = rapl::read_package_energy().unwrap() - self.rapl_offset; let rapl_diff = rapl - self.old_rapl; - let est_diff = (init.tree_energy - old_energy); + let est_diff = init.tree_energy - old_energy; if est_diff < 0.1 { self.offset = (self.offset + (rapl_diff - est_diff)) * 0.5; } self.old_rapl = rapl; - init.tree_energy = (init.tree_energy + self.offset); + init.tree_energy = init.tree_energy + self.offset; let offset_bias = (rapl / init.tree_energy).clamp(0.1, 2.); let diff_bias = (rapl_diff / est_diff).clamp(0.1, 2.); let current_bias = (offset_bias + diff_bias) * 0.5; - //dbg!(current_bias, self.bias); self.bias = (self.bias * ((1. / 3.) * current_bias + (2. / 3.))).clamp(0.1, 20.); - //let v = logged_energy.get(&1).unwrap(); println!( "Energy estimation: {:.1} rapl: {:.1}, est diff: {:.1} rapl diff: {:.1}", - init.tree_energy, - rapl, - est_diff, - rapl_diff, - //(rapl_diff - (init.tree_energy - old_energy)) / rapl_diff * 100., + init.tree_energy, rapl, est_diff, rapl_diff, ); - // if init.tree_energy > 1000. || rapl > 1000. { - // self.rapl_offset += rapl; - // init.tree_energy = 0.; - // } } } @@ -251,13 +214,19 @@ impl EnergyService { } pub fn process_energy(&self, pid: Pid) -> Option<f64> { - self.process_info.get(&pid).map(|info| info.energy) + self.process_info + .read() + .unwrap() + .get(&pid) + .map(|info| info.energy) } pub fn all_process_energies(&self) -> HashMap<Pid, f64> { self.process_info + .read() + .unwrap() .iter() - .map(|(&pid, info)| (pid, info.energy)) + .map(|(&key, info)| (key, info.energy)) .collect() } } @@ -286,23 +255,23 @@ pub fn start_energy_service( // shouldn't be a problem because we are privileged // if PackageEnergy::check_paranoid().unwrap_or(3) > 0 {} - let logged_energy = Arc::new(RwLock::new(HashMap::new())); + let process_info = Arc::new(RwLock::new(HashMap::new())); // Create and start the energy service let service = EnergyService::new( estimator, budget_policy, + process_info.clone(), shared_budgets, request_receiver, Duration::from_millis(50), // 50ms update interval shared_cpu_frequency_ranges, shared_policy_frequency_ranges, shared_cpu_current_frequencies, - logged_energy.clone(), ); service.run(); - //socket::start_logging_socket_service("/tmp/pm-sched", logged_energy, Duration::from_secs(1))?; + socket::start_logging_socket_service("/tmp/pm-sched", process_info.clone())?; Ok(request_sender) } diff --git a/src/energy/budget.rs b/src/energy/budget.rs index 7002dd3..6154187 100644 --- a/src/energy/budget.rs +++ b/src/energy/budget.rs @@ -42,8 +42,8 @@ impl BudgetPolicy for SimpleCappingPolicy { } } else { // Under power cap, assign maximum budget to all - for &pid in energy_service.active_processes() { - budgets.insert(pid, u64::MAX); + for pid in energy_service.active_processes() { + budgets.insert(*pid, u64::MAX); } } diff --git a/src/main.rs b/src/main.rs index 6242917..c56a41c 100644 --- a/src/main.rs +++ b/src/main.rs @@ -47,23 +47,11 @@ fn main() -> Result<()> { .required(false) .value_name("power in watts"), ) - .arg( - Arg::new("log_energy") - .long("log_energy") - .help("Provide one or more PIDs to log their process tress energy") - .num_args(1..) - .required(false), - ) .get_matches(); let power_cap = *matches.get_one::<u64>("power_cap").unwrap_or(&u64::MAX); let use_mocking = matches.get_flag("mock"); let benchmark = matches.get_flag("benchmark"); - let pids_to_monitor: Vec<Pid> = matches - .get_many::<String>("log_energy") - .unwrap_or_default() - .map(|v| v.parse::<Pid>().unwrap()) - .collect(); // Initialize and load the scheduler. let mut open_object = MaybeUninit::uninit(); @@ -74,8 +62,7 @@ fn main() -> Result<()> { return Ok(()); } loop { - let mut sched = - Scheduler::init(&mut open_object, use_mocking, power_cap, &pids_to_monitor)?; + let mut sched = Scheduler::init(&mut open_object, use_mocking, power_cap)?; if !sched.run()?.should_restart() { break; } diff --git a/src/scheduler.rs b/src/scheduler.rs index fc4d448..359b9da 100644 --- a/src/scheduler.rs +++ b/src/scheduler.rs @@ -45,7 +45,6 @@ impl<'a> Scheduler<'a> { open_object: &'a mut MaybeUninit<OpenObject>, use_mocking: bool, power_cap: u64, - pids_to_monitor: &[Pid], ) -> Result<Self> { let bpf = BpfScheduler::init( open_object, @@ -76,12 +75,6 @@ impl<'a> Scheduler<'a> { shared_cpu_current_frequencies.clone(), )?; - for pid in pids_to_monitor { - energy_sender - .try_send(EnergyRequest::LogEnergyTask(*pid)) - .unwrap(); - } - let topology = Topology::new().unwrap(); let mut e_core_ids = Vec::new(); let mut p_core_ids = Vec::new(); @@ -115,9 +108,6 @@ impl<'a> Scheduler<'a> { frequency_sender .try_send(FrequencyRequest::GetPossibleCPUFrequencyRange) .unwrap(); - frequency_sender - .try_send(FrequencyRequest::GetCurrentFrequencies) - .unwrap(); Ok(Self { bpf, diff --git a/src/socket.rs b/src/socket.rs index 0503d1f..0b42abe 100644 --- a/src/socket.rs +++ b/src/socket.rs @@ -1,7 +1,7 @@ use std::{ collections::HashMap, fs, - io::{self, Write}, + io::{self, BufRead, BufReader, Write}, os::unix::net::UnixListener, path::Path, sync::{Arc, RwLock}, @@ -9,44 +9,20 @@ use std::{ time::Duration, }; -use crate::{energy::EnergyLog, Pid}; +use dashmap::DashMap; + +use crate::{energy::ProcessInfo, Pid}; pub struct LoggingSocketService { path: String, - logged_energy: Arc<RwLock<HashMap<Pid, EnergyLog>>>, - update_interval: Duration, + process_info: Arc<RwLock<HashMap<Pid, ProcessInfo>>>, } impl LoggingSocketService { - pub fn new( - path: &str, - logged_energy: Arc<RwLock<HashMap<Pid, EnergyLog>>>, - update_interval: Duration, - ) -> Self { + pub fn new(path: &str, process_info: Arc<RwLock<HashMap<Pid, ProcessInfo>>>) -> Self { LoggingSocketService { path: path.to_string(), - logged_energy, - update_interval, - } - } - - fn generate_report(&self) -> String { - loop { - if let Ok(logged_energy) = self.logged_energy.try_read() { - let mut result = logged_energy - .iter() - .map(|(key, value)| { - format!( - "{key}: {}J", - value.energy_of_running_tasks + value.energy_of_exited_tasks - ) - }) - .collect::<Vec<_>>() - .join(", "); - result += "\n"; - break result; - } - thread::sleep(Duration::from_millis(10)); + process_info, } } @@ -55,26 +31,38 @@ impl LoggingSocketService { let listener = UnixListener::bind(self.path.clone()).unwrap(); for stream in listener.incoming() { let mut socket = stream.unwrap(); - loop { - if socket - .write_all(format!("{}", self.generate_report()).as_bytes()) - .is_err() - { - break; - } - thread::sleep(self.update_interval); - } - //TODO: if we ever want to read too - /*let mut reader = BufReader::new(socket.try_clone().unwrap()); + let mut reader = BufReader::new(socket.try_clone().unwrap()); let mut line = String::new(); while let Ok(bytes_read) = reader.read_line(&mut line) { if bytes_read == 0 { break; } - socket.write_all(format!("got: {line}").as_bytes()); - println!("got: {line}"); + if let Ok(pid) = line.trim().parse::<Pid>() { + if let Some(info) = self.process_info.read().unwrap().get(&pid).clone() { + socket + .write_all( + format!( + "pid: {pid} process: {}J process tree: {}J\n", + info.energy, info.tree_energy + ) + .as_bytes(), + ) + .unwrap(); + } else { + socket + .write_all(format!("Unknown pid: {pid}\n").as_bytes()) + .unwrap(); + } + } else { + socket + .write_all( + format!("Failed to get energy for pid: {}\n", line.trim_end()) + .as_bytes(), + ) + .unwrap(); + } line.clear(); - }*/ + } } }); } @@ -82,13 +70,12 @@ impl LoggingSocketService { pub fn start_logging_socket_service( path: &str, - logged_energy: Arc<RwLock<HashMap<Pid, EnergyLog>>>, - update_interval: Duration, + process_info: Arc<RwLock<HashMap<Pid, ProcessInfo>>>, ) -> io::Result<()> { if Path::new(path).exists() { fs::remove_file(path)?; } - let socket = LoggingSocketService::new(path, logged_energy, update_interval); + let socket = LoggingSocketService::new(path, process_info); socket.run(); Ok(()) } |
