Files
cgroups-rs/src/cpuset.rs
2020-08-27 23:06:44 +08:00

536 lines
20 KiB
Rust

//! This module contains the implementation of the `cpuset` cgroup subsystem.
//!
//! See the Kernel's documentation for more information about this subsystem, found at:
//! [Documentation/cgroup-v1/cpusets.txt](https://www.kernel.org/doc/Documentation/cgroup-v1/cpusets.txt)
use log::*;
use std::fs::File;
use std::io::{Read, Write};
use std::path::PathBuf;
use crate::error::*;
use crate::error::ErrorKind::*;
use crate::{
ControllIdentifier, ControllerInternal, Controllers, CpuResources, Resources, Subsystem,
};
/// A controller that allows controlling the `cpuset` subsystem of a Cgroup.
///
/// In essence, this controller is responsible for restricting the tasks in the control group to a
/// set of CPUs and/or memory nodes.
#[derive(Debug, Clone)]
pub struct CpuSetController {
base: PathBuf,
path: PathBuf,
v2: bool,
}
/// The current state of the `cpuset` controller for this control group.
pub struct CpuSet {
/// If true, no other control groups can share the CPUs listed in the `cpus` field.
pub cpu_exclusive: bool,
/// The list of CPUs the tasks of the control group can run on.
///
/// This is a vector of `(start, end)` tuples, where each tuple is a range of CPUs where the
/// control group is allowed to run on. Both sides of the range are inclusive.
pub cpus: Vec<(u64, u64)>,
/// The list of CPUs that the tasks can effectively run on. This removes the list of CPUs that
/// the parent (and all of its parents) cannot run on from the `cpus` field of this control
/// group.
pub effective_cpus: Vec<(u64, u64)>,
/// The list of memory nodes that the tasks can effectively use. This removes the list of nodes that
/// the parent (and all of its parents) cannot use from the `mems` field of this control
/// group.
pub effective_mems: Vec<(u64, u64)>,
/// If true, no other control groups can share the memory nodes listed in the `mems` field.
pub mem_exclusive: bool,
/// If true, the control group is 'hardwalled'. Kernel memory allocations (except for a few
/// minor exceptions) are made from the memory nodes designated in the `mems` field.
pub mem_hardwall: bool,
/// If true, whenever `mems` is changed via `set_mems()`, the memory stored on the previous
/// nodes are migrated to the new nodes selected by the new `mems`.
pub memory_migrate: bool,
/// Running average of the memory pressured faced by the tasks in the control group.
pub memory_pressure: u64,
/// This field is only at the root control group and controls whether the kernel will compute
/// the memory pressure for control groups or not.
pub memory_pressure_enabled: Option<bool>,
/// If true, filesystem buffers are spread across evenly between the nodes specified in `mems`.
pub memory_spread_page: bool,
/// If true, kernel slab caches for file I/O are spread across evenly between the nodes
/// specified in `mems`.
pub memory_spread_slab: bool,
/// The list of memory nodes the tasks of the control group can use.
///
/// The format is the same as the `cpus`, `effective_cpus` and `effective_mems` fields.
pub mems: Vec<(u64, u64)>,
/// If true, the kernel will attempt to rebalance the load between the CPUs specified in the
/// `cpus` field of this control group.
pub sched_load_balance: bool,
/// Represents how much work the kernel should do to rebalance this cpuset.
///
/// | `sched_load_balance` | Effect |
/// | -------------------- | ------ |
/// | -1 | Use the system default value |
/// | 0 | Only balance loads periodically |
/// | 1 | Immediately balance the load across tasks on the same core |
/// | 2 | Immediately balance the load across cores in the same CPU package |
/// | 4 | Immediately balance the load across CPUs on the same node |
/// | 5 | Immediately balance the load between CPUs even if the system is NUMA |
/// | 6 | Immediately balance the load between all CPUs |
pub sched_relax_domain_level: u64,
}
impl ControllerInternal for CpuSetController {
fn control_type(&self) -> Controllers {
Controllers::CpuSet
}
fn get_path(&self) -> &PathBuf {
&self.path
}
fn get_path_mut(&mut self) -> &mut PathBuf {
&mut self.path
}
fn get_base(&self) -> &PathBuf {
&self.base
}
fn is_v2(&self) -> bool {
self.v2
}
fn apply(&self, res: &Resources) -> Result<()> {
// get the resources that apply to this controller
let res: &CpuResources = &res.cpu;
if res.update_values {
if res.cpus.is_some(){
let _ = self.set_cpus(res.cpus.as_ref().unwrap().as_str());
}
let _ = self.set_mems(&res.mems);
}
Ok(())
}
fn post_create(&self){
if self.is_v2(){
return
}
let current = self.get_path();
let parent = match current.parent() {
Some(p) => p,
None => return,
};
if current != self.get_base() {
match copy_from_parent(current.to_str().unwrap(), parent.to_str().unwrap()) {
Ok(_)=>(),
Err(err) => error!("error create_dir {:?}", err),
}
}
}
}
/// copy_from_parent copy the cpuset.cpus and cpuset.mems from the parent
/// directory to the current directory if the file's contents are 0
fn copy_from_parent(current: &str, parent: &str) -> Result<()> {
let cpus_str: &str = "cpuset.cpus";
let mems_str: &str = "cpuset.mems";
let current_cpus_path = ::std::path::Path::new(current).join(cpus_str);
let current_mems_path = ::std::path::Path::new(current).join(mems_str);
let parent_cpus_path = ::std::path::Path::new(parent).join(cpus_str);
let parent_mems_path = ::std::path::Path::new(parent).join(mems_str);
let current_cpus = match ::std::fs::read_to_string(current_cpus_path.to_str().unwrap()) {
Ok(cpus) => String::from(cpus.trim()),
Err(e) => return Err(Error::with_cause(ReadFailed, e)),
};
let current_mems = match ::std::fs::read_to_string(current_mems_path.to_str().unwrap()) {
Ok(mems) => String::from(mems.trim()),
Err(e) => return Err(Error::with_cause(ReadFailed, e)),
};
let parent_cpus = match ::std::fs::read_to_string(parent_cpus_path.to_str().unwrap()) {
Ok(cpus) => cpus,
Err(e) => return Err(Error::with_cause(ReadFailed, e)),
};
let parent_mems = match ::std::fs::read_to_string(parent_mems_path.to_str().unwrap()) {
Ok(mems) => mems,
Err(e) => return Err(Error::with_cause(ReadFailed, e)),
};
if current_cpus == "" {
match ::std::fs::write(current_cpus_path.to_str().unwrap(), parent_cpus.as_bytes()) {
Ok(_) => (),
Err(e) => return Err(Error::with_cause(WriteFailed, e)),
}
}
if current_mems == "" {
match ::std::fs::write(current_mems_path.to_str().unwrap(), parent_mems.as_bytes()) {
Ok(_) => (),
Err(e) => return Err(Error::with_cause(WriteFailed, e)),
}
}
Ok(())
}
impl ControllIdentifier for CpuSetController {
fn controller_type() -> Controllers {
Controllers::CpuSet
}
}
impl<'a> From<&'a Subsystem> for &'a CpuSetController {
fn from(sub: &'a Subsystem) -> &'a CpuSetController {
unsafe {
match sub {
Subsystem::CpuSet(c) => c,
_ => {
assert_eq!(1, 0);
::std::mem::uninitialized()
}
}
}
}
}
fn read_string_from(mut file: File) -> Result<String> {
let mut string = String::new();
match file.read_to_string(&mut string) {
Ok(_) => Ok(string.trim().to_string()),
Err(e) => Err(Error::with_cause(ReadFailed, e)),
}
}
fn read_u64_from(mut file: File) -> Result<u64> {
let mut string = String::new();
match file.read_to_string(&mut string) {
Ok(_) => string.trim().parse().map_err(|e| Error::with_cause(ParseError, e)),
Err(e) => Err(Error::with_cause(ReadFailed, e)),
}
}
/// Parse a string like "1,2,4-5,8" into a list of (start, end) tuples.
fn parse_range(s: String) -> Result<Vec<(u64, u64)>> {
let mut fin = Vec::new();
if s == "".to_string() {
return Ok(fin);
}
// first split by commas
let comma_split = s.split(",");
for sp in comma_split {
if sp.contains("-") {
// this is a true range
let dash_split = sp.split("-").collect::<Vec<_>>();
if dash_split.len() != 2 {
return Err(Error::new(ParseError));
}
let first = dash_split[0].parse::<u64>();
let second = dash_split[1].parse::<u64>();
if first.is_err() || second.is_err() {
return Err(Error::new(ParseError));
}
fin.push((first.unwrap(), second.unwrap()));
} else {
// this is just a single number
let num = sp.parse::<u64>();
if num.is_err() {
return Err(Error::new(ParseError));
}
fin.push((num.clone().unwrap(), num.clone().unwrap()));
}
}
Ok(fin)
}
impl CpuSetController {
/// Contructs a new `CpuSetController` with `oroot` serving as the root of the control group.
pub fn new(oroot: PathBuf, v2: bool) -> Self {
let mut root = oroot;
if !v2{
root.push(Self::controller_type().to_string());
}
Self {
base: root.clone(),
path: root,
v2: v2,
}
}
/// Returns the statistics gathered by the kernel for this control group. See the struct for
/// more information on what information this entails.
pub fn cpuset(&self) -> CpuSet {
CpuSet {
cpu_exclusive: {
self.open_path("cpuset.cpu_exclusive", false)
.and_then(|file| read_u64_from(file))
.map(|x| x == 1)
.unwrap_or(false)
},
cpus: {
self.open_path("cpuset.cpus", false)
.and_then(read_string_from)
.and_then(parse_range)
.unwrap_or(Vec::new())
},
effective_cpus: {
self.open_path("cpuset.effective_cpus", false)
.and_then(read_string_from)
.and_then(parse_range)
.unwrap_or(Vec::new())
},
effective_mems: {
self.open_path("cpuset.effective_mems", false)
.and_then(read_string_from)
.and_then(parse_range)
.unwrap_or(Vec::new())
},
mem_exclusive: {
self.open_path("cpuset.mem_exclusive", false)
.and_then(read_u64_from)
.map(|x| x == 1)
.unwrap_or(false)
},
mem_hardwall: {
self.open_path("cpuset.mem_hardwall", false)
.and_then(read_u64_from)
.map(|x| x == 1)
.unwrap_or(false)
},
memory_migrate: {
self.open_path("cpuset.memory_migrate", false)
.and_then(read_u64_from)
.map(|x| x == 1)
.unwrap_or(false)
},
memory_pressure: {
self.open_path("cpuset.memory_pressure", false)
.and_then(read_u64_from)
.unwrap_or(0)
},
memory_pressure_enabled: {
self.open_path("cpuset.memory_pressure_enabled", false)
.and_then(read_u64_from)
.map(|x| x == 1)
.ok()
},
memory_spread_page: {
self.open_path("cpuset.memory_spread_page", false)
.and_then(read_u64_from)
.map(|x| x == 1)
.unwrap_or(false)
},
memory_spread_slab: {
self.open_path("cpuset.memory_spread_slab", false)
.and_then(read_u64_from)
.map(|x| x == 1)
.unwrap_or(false)
},
mems: {
self.open_path("cpuset.mems", false)
.and_then(read_string_from)
.and_then(parse_range)
.unwrap_or(Vec::new())
},
sched_load_balance: {
self.open_path("cpuset.sched_load_balance", false)
.and_then(read_u64_from)
.map(|x| x == 1)
.unwrap_or(false)
},
sched_relax_domain_level: {
self.open_path("cpuset.sched_relax_domain_level", false)
.and_then(read_u64_from)
.unwrap_or(0)
},
}
}
/// Control whether the CPUs selected via `set_cpus()` should be exclusive to this control
/// group or not.
pub fn set_cpu_exclusive(&self, b: bool) -> Result<()> {
self.open_path("cpuset.cpu_exclusive", true)
.and_then(|mut file| {
if b {
file.write_all(b"1").map_err(|e| Error::with_cause(WriteFailed, e))
} else {
file.write_all(b"0").map_err(|e| Error::with_cause(WriteFailed, e))
}
})
}
/// Control whether the memory nodes selected via `set_memss()` should be exclusive to this control
/// group or not.
pub fn set_mem_exclusive(&self, b: bool) -> Result<()> {
self.open_path("cpuset.mem_exclusive", true)
.and_then(|mut file| {
if b {
file.write_all(b"1").map_err(|e| Error::with_cause(WriteFailed, e))
} else {
file.write_all(b"0").map_err(|e| Error::with_cause(WriteFailed, e))
}
})
}
/// Set the CPUs that the tasks in this control group can run on.
///
/// Syntax is a comma separated list of CPUs, with an additional extension that ranges can
/// be represented via dashes.
pub fn set_cpus(&self, cpus: &str) -> Result<()> {
self.open_path("cpuset.cpus", true).and_then(|mut file| {
file.write_all(cpus.as_ref())
.map_err(|e| Error::with_cause(WriteFailed, e))
})
}
/// Set the memory nodes that the tasks in this control group can use.
///
/// Syntax is the same as with `set_cpus()`.
pub fn set_mems(&self, mems: &str) -> Result<()> {
self.open_path("cpuset.mems", true).and_then(|mut file| {
file.write_all(mems.as_ref())
.map_err(|e| Error::with_cause(WriteFailed, e))
})
}
/// Controls whether the control group should be "hardwalled", i.e., whether kernel allocations
/// should exclusively use the memory nodes set via `set_mems()`.
///
/// Note that some kernel allocations, most notably those that are made in interrupt handlers
/// may disregard this.
pub fn set_hardwall(&self, b: bool) -> Result<()> {
self.open_path("cpuset.mem_hardwall", true)
.and_then(|mut file| {
if b {
file.write_all(b"1").map_err(|e| Error::with_cause(WriteFailed, e))
} else {
file.write_all(b"0").map_err(|e| Error::with_cause(WriteFailed, e))
}
})
}
/// Controls whether the kernel should attempt to rebalance the load between the CPUs specified in the
/// `cpus` field of this control group.
pub fn set_load_balancing(&self, b: bool) -> Result<()> {
self.open_path("cpuset.sched_load_balance", true)
.and_then(|mut file| {
if b {
file.write_all(b"1").map_err(|e| Error::with_cause(WriteFailed, e))
} else {
file.write_all(b"0").map_err(|e| Error::with_cause(WriteFailed, e))
}
})
}
/// Contorl how much effort the kernel should invest in rebalacing the control group.
///
/// See @CpuSet 's similar field for more information.
pub fn set_rebalance_relax_domain_level(&self, i: i64) -> Result<()> {
self.open_path("cpuset.sched_relax_domain_level", true)
.and_then(|mut file| {
file.write_all(i.to_string().as_ref())
.map_err(|e| Error::with_cause(WriteFailed, e))
})
}
/// Control whether when using `set_mems()` the existing memory used by the tasks should be
/// migrated over to the now-selected nodes.
pub fn set_memory_migration(&self, b: bool) -> Result<()> {
self.open_path("cpuset.memory_migrate", true)
.and_then(|mut file| {
if b {
file.write_all(b"1").map_err(|e| Error::with_cause(WriteFailed, e))
} else {
file.write_all(b"0").map_err(|e| Error::with_cause(WriteFailed, e))
}
})
}
/// Control whether filesystem buffers should be evenly split across the nodes selected via
/// `set_mems()`.
pub fn set_memory_spread_page(&self, b: bool) -> Result<()> {
self.open_path("cpuset.memory_spread_page", true)
.and_then(|mut file| {
if b {
file.write_all(b"1").map_err(|e| Error::with_cause(WriteFailed, e))
} else {
file.write_all(b"0").map_err(|e| Error::with_cause(WriteFailed, e))
}
})
}
/// Control whether the kernel's slab cache for file I/O should be evenly split across the
/// nodes selected via `set_mems()`.
pub fn set_memory_spread_slab(&self, b: bool) -> Result<()> {
self.open_path("cpuset.memory_spread_slab", true)
.and_then(|mut file| {
if b {
file.write_all(b"1").map_err(|e| Error::with_cause(WriteFailed, e))
} else {
file.write_all(b"0").map_err(|e| Error::with_cause(WriteFailed, e))
}
})
}
/// Control whether the kernel should collect information to calculate memory pressure for
/// control groups.
///
/// Note: This will fail with `InvalidOperation` if the current congrol group is not the root
/// control group.
pub fn set_enable_memory_pressure(&self, b: bool) -> Result<()> {
if !self.path_exists("cpuset.memory_pressure_enabled") {
return Err(Error::new(InvalidOperation));
}
self.open_path("cpuset.memory_pressure_enabled", true)
.and_then(|mut file| {
if b {
file.write_all(b"1").map_err(|e| Error::with_cause(WriteFailed, e))
} else {
file.write_all(b"0").map_err(|e| Error::with_cause(WriteFailed, e))
}
})
}
}
#[cfg(test)]
mod tests {
use crate::cpuset;
#[test]
fn test_parse_range() {
let test_cases = vec![
"1,2,4-6,9".to_string(),
"".to_string(),
"1".to_string(),
"1-111".to_string(),
"1,2,3,4".to_string(),
"1-5,6-7,8-9".to_string(),
];
let expecteds = vec![
vec![(1, 1), (2, 2), (4, 6), (9, 9)],
vec![],
vec![(1, 1)],
vec![(1, 111)],
vec![(1, 1), (2, 2), (3, 3), (4, 4)],
vec![(1, 5), (6, 7), (8, 9)],
];
for (i, case) in test_cases.into_iter().enumerate() {
let range = cpuset::parse_range(case.clone());
println!("{:?} => {:?}", case, range);
assert!(range.is_ok());
assert_eq!(range.unwrap(), expecteds[i]);
}
}
}