runtime-rs: check the update memory size

check the update memory size greater than default max memory size

Fixes:#6875
Signed-off-by: Zhongtao Hu <zhongtaohu.tim@linux.alibaba.com>
This commit is contained in:
Zhongtao Hu 2023-06-21 18:10:38 +08:00
parent 6039417104
commit 9a37e77f2a
15 changed files with 201 additions and 122 deletions

View File

@ -14,7 +14,7 @@ edition = "2018"
bitmask-enum = "2.1.0"
anyhow = "1.0"
base64 = "0.13.0"
byte-unit = "3.1.4"
byte-unit = "5.0.3"
glob = "0.3.0"
lazy_static = "1.4.0"
num_cpus = "1.13.1"
@ -26,6 +26,7 @@ serde_json = "1.0.73"
thiserror = "1.0"
toml = "0.5.8"
serde-enum-str = "0.4"
sysinfo = "0.29.11"
oci = { path = "../oci" }
safe-path = { path = "../safe-path" }

View File

@ -705,10 +705,10 @@ impl Annotation {
}
// Hypervisor Memory related annotations
KATA_ANNO_CFG_HYPERVISOR_DEFAULT_MEMORY => {
match byte_unit::Byte::from_str(value) {
match byte_unit::Byte::parse_str(value,true) {
Ok(mem_bytes) => {
let memory_size = mem_bytes
.get_adjusted_unit(byte_unit::ByteUnit::MiB)
.get_adjusted_unit(byte_unit::Unit::MiB)
.get_value()
as u32;
info!(sl!(), "get mem {} from annotations: {}", memory_size, value);

View File

@ -20,7 +20,7 @@ pub enum CapabilityBits {
/// hypervisor supports hybrid-vsock
HybridVsockSupport,
/// hypervisor supports memory hotplug probe interface
GuestMemoryHotplugProbe,
GuestMemoryProbe,
}
/// Capabilities describe a virtcontainers hypervisor capabilities through a bit mask.
@ -81,7 +81,7 @@ impl Capabilities {
/// is_mem_hotplug_probe_supported tells if the hypervisor supports hotplug probe interface
pub fn is_mem_hotplug_probe_supported(&self) -> bool {
self.flags.and(CapabilityBits::GuestMemoryHotplugProbe) != 0
self.flags.and(CapabilityBits::GuestMemoryProbe) != 0
}
}
@ -130,7 +130,7 @@ mod tests {
cap.set(CapabilityBits::HybridVsockSupport);
assert!(cap.is_hybrid_vsock_supported());
// test append capabilities
cap.add(CapabilityBits::GuestMemoryHotplugProbe);
cap.add(CapabilityBits::GuestMemoryProbe);
assert!(cap.is_mem_hotplug_probe_supported());
assert!(cap.is_fs_sharing_supported());
}

View File

@ -22,18 +22,18 @@
//! part and common part. But the Kata 2.0 has adopted a policy to build a superset for all
//! hypervisors, so let's contain it...
use super::{default, ConfigOps, ConfigPlugin, TomlConfig};
use crate::annotations::KATA_ANNO_CFG_HYPERVISOR_PREFIX;
use crate::{eother, resolve_path, sl, validate_path};
use byte_unit::{Byte, Unit};
use lazy_static::lazy_static;
use regex::RegexSet;
use serde_enum_str::{Deserialize_enum_str, Serialize_enum_str};
use std::collections::HashMap;
use std::io::{self, Result};
use std::path::Path;
use std::sync::{Arc, Mutex};
use lazy_static::lazy_static;
use regex::RegexSet;
use serde_enum_str::{Deserialize_enum_str, Serialize_enum_str};
use super::{default, ConfigOps, ConfigPlugin, TomlConfig};
use crate::annotations::KATA_ANNO_CFG_HYPERVISOR_PREFIX;
use crate::{eother, resolve_path, sl, validate_path};
use sysinfo::{System, SystemExt};
mod dragonball;
pub use self::dragonball::{DragonballConfig, HYPERVISOR_NAME_DRAGONBALL};
@ -586,6 +586,13 @@ pub struct MemoryInfo {
#[serde(default)]
pub default_memory: u32,
/// Default maximum memory in MiB per SB / VM
/// unspecified or == 0 --> will be set to the actual amount of physical RAM
/// > 0 <= amount of physical RAM --> will be set to the specified number
/// > amount of physical RAM --> will be set to the actual amount of physical RAM
#[serde(default)]
pub default_maxmemory: u32,
/// Default memory slots per SB/VM.
///
/// This is will determine the times that memory will be hotadded to sandbox/VM.
@ -653,13 +660,6 @@ pub struct MemoryInfo {
/// If swap_in_bytes and memory_limit_in_bytes is not set, the size should be default_memory.
#[serde(default)]
pub enable_guest_swap: bool,
/// Default maximum memory in MiB per SB / VM
/// unspecified or == 0 --> will be set to the actual amount of physical RAM
/// > 0 <= amount of physical RAM --> will be set to the specified number
/// > amount of physical RAM --> will be set to the actual amount of physical RAM
#[serde(default)]
pub default_maxmemory: u32,
}
impl MemoryInfo {
@ -669,6 +669,12 @@ impl MemoryInfo {
self.file_mem_backend,
"Memory backend file {} is invalid: {}"
)?;
if self.default_maxmemory == 0 {
let s = System::new_all();
self.default_maxmemory = Byte::from_u64(s.total_memory())
.get_adjusted_unit(Unit::MiB)
.get_value() as u32;
}
Ok(())
}

View File

@ -372,12 +372,6 @@ version = "3.13.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "a3e2c3daef883ecc1b5d58c15adae93470a91d425f3532ba1695849656af3fc1"
[[package]]
name = "byte-unit"
version = "3.1.4"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "415301c9de11005d4b92193c0eb7ac7adc37e5a49e0ac9bed0a42343512744b8"
[[package]]
name = "byte-unit"
version = "4.0.19"
@ -388,6 +382,17 @@ dependencies = [
"utf8-width",
]
[[package]]
name = "byte-unit"
version = "5.0.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "bc40af92e0f7f964b7ab1ebc81315cce78fc484802d534143321c956f58d7be3"
dependencies = [
"rust_decimal",
"serde",
"utf8-width",
]
[[package]]
name = "byteorder"
version = "1.4.3"
@ -601,6 +606,30 @@ dependencies = [
"crossbeam-utils",
]
[[package]]
name = "crossbeam-deque"
version = "0.8.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "ce6fd6f855243022dcecf8702fef0c297d4338e226845fe067f6341ad9fa0cef"
dependencies = [
"cfg-if 1.0.0",
"crossbeam-epoch",
"crossbeam-utils",
]
[[package]]
name = "crossbeam-epoch"
version = "0.9.15"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "ae211234986c545741a7dc064309f67ee1e5ad243d0e48335adc0484d960bcc7"
dependencies = [
"autocfg",
"cfg-if 1.0.0",
"crossbeam-utils",
"memoffset 0.9.0",
"scopeguard",
]
[[package]]
name = "crossbeam-utils"
version = "0.8.16"
@ -1657,7 +1686,7 @@ dependencies = [
"anyhow",
"base64 0.13.1",
"bitmask-enum",
"byte-unit 3.1.4",
"byte-unit 5.0.3",
"glob",
"lazy_static",
"num_cpus",
@ -1669,6 +1698,7 @@ dependencies = [
"serde_json",
"slog",
"slog-scope",
"sysinfo",
"thiserror",
"toml 0.5.11",
]
@ -1843,6 +1873,15 @@ dependencies = [
"autocfg",
]
[[package]]
name = "memoffset"
version = "0.9.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "5a634b1c61a95585bd15607c6ab0c4e5b226e695ff2800ba0cdccddf208c406c"
dependencies = [
"autocfg",
]
[[package]]
name = "mime"
version = "0.3.17"
@ -2023,6 +2062,15 @@ dependencies = [
"static_assertions",
]
[[package]]
name = "ntapi"
version = "0.4.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "e8a3895c6391c39d7fe7ebc444a87eb2991b2a0bc718fdabd071eec617fc68e4"
dependencies = [
"winapi",
]
[[package]]
name = "nu-ansi-term"
version = "0.46.0"
@ -2809,6 +2857,26 @@ dependencies = [
"rand_core 0.5.1",
]
[[package]]
name = "rayon"
version = "1.8.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "9c27db03db7734835b3f53954b534c91069375ce6ccaa2e065441e07d9b6cdb1"
dependencies = [
"either",
"rayon-core",
]
[[package]]
name = "rayon-core"
version = "1.12.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "5ce3fb6ad83f861aac485e76e1985cd109d9a3713802152be56c3b1f0e0658ed"
dependencies = [
"crossbeam-deque",
"crossbeam-utils",
]
[[package]]
name = "rdrand"
version = "0.4.0"
@ -3024,6 +3092,16 @@ dependencies = [
"ordered-multimap",
]
[[package]]
name = "rust_decimal"
version = "1.33.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "06676aec5ccb8fc1da723cc8c0f9a46549f21ebb8753d3915c6c41db1e7f1dc4"
dependencies = [
"arrayvec",
"num-traits",
]
[[package]]
name = "rustc-demangle"
version = "0.1.23"
@ -3559,6 +3637,21 @@ dependencies = [
"unicode-ident",
]
[[package]]
name = "sysinfo"
version = "0.29.11"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "cd727fc423c2060f6c92d9534cef765c65a6ed3f428a03d7def74a8c4348e666"
dependencies = [
"cfg-if 1.0.0",
"core-foundation-sys",
"libc",
"ntapi",
"once_cell",
"rayon",
"winapi",
]
[[package]]
name = "take_mut"
version = "0.2.2"

View File

@ -756,7 +756,7 @@ impl CloudHypervisorInner {
todo!()
}
pub(crate) fn resize_memory(&self, _req_mem_mb: u32) -> Result<(u32, MemoryConfig)> {
pub(crate) fn resize_memory(&self, _new_mem_mb: u32) -> Result<(u32, MemoryConfig)> {
todo!()
}
}

View File

@ -178,9 +178,9 @@ impl Hypervisor for CloudHypervisor {
inner.guest_memory_block_size_mb()
}
async fn resize_memory(&self, req_mem_mb: u32) -> Result<(u32, MemoryConfig)> {
async fn resize_memory(&self, new_mem_mb: u32) -> Result<(u32, MemoryConfig)> {
let inner = self.inner.read().await;
inner.resize_memory(req_mem_mb)
inner.resize_memory(new_mem_mb)
}
}

View File

@ -32,7 +32,8 @@ use std::{collections::HashSet, fs::create_dir_all};
const DRAGONBALL_KERNEL: &str = "vmlinux";
const DRAGONBALL_ROOT_FS: &str = "rootfs";
const BALLOON_DEVICE_ID: &str = "balloon0";
const MEM_DEVICE_ID: &str = "memmr0";
#[derive(Debug)]
pub struct DragonballInner {
/// sandbox id
@ -76,6 +77,9 @@ pub struct DragonballInner {
/// the hotplug memory size
pub(crate) mem_hotplug_size_mb: u32,
/// the balloon size
pub(crate) balloon_size: u32,
}
impl DragonballInner {
@ -85,7 +89,8 @@ impl DragonballInner {
CapabilityBits::BlockDeviceSupport
| CapabilityBits::BlockDeviceHotplugSupport
| CapabilityBits::FsSharingSupport
| CapabilityBits::HybridVsockSupport,
| CapabilityBits::HybridVsockSupport
| CapabilityBits::GuestMemoryProbe,
);
DragonballInner {
id: "".to_string(),
@ -102,6 +107,7 @@ impl DragonballInner {
capabilities,
guest_memory_block_size_mb: 0,
mem_hotplug_size_mb: 0,
balloon_size: 0,
}
}
@ -349,13 +355,30 @@ impl DragonballInner {
Ok((old_vcpus, new_vcpus))
}
pub(crate) fn resize_memory(&mut self, req_mem_mb: u32) -> Result<(u32, MemoryConfig)> {
pub(crate) fn resize_memory(&mut self, new_mem_mb: u32) -> Result<(u32, MemoryConfig)> {
// check the invalid request memory
if new_mem_mb > self.hypervisor_config().memory_info.default_maxmemory {
warn!(
sl!(),
"memory size unchanged, the request memory size {} is greater than the max memory size {}",
new_mem_mb, self.hypervisor_config().memory_info.default_maxmemory
);
return Ok((
0,
MemoryConfig {
..Default::default()
},
));
}
let had_mem_mb = self.config.memory_info.default_memory + self.mem_hotplug_size_mb;
match req_mem_mb.cmp(&had_mem_mb) {
match new_mem_mb.cmp(&had_mem_mb) {
Ordering::Greater => {
// clean virtio-ballon device before hotplug memory, resize to 0
if self.balloon_size > 0 {
let balloon_config = BalloonDeviceConfigInfo {
balloon_id: "balloon0".to_owned(),
balloon_id: BALLOON_DEVICE_ID.to_owned(),
size_mib: 0,
use_shared_irq: None,
use_generic_irq: None,
@ -365,17 +388,18 @@ impl DragonballInner {
self.vmm_instance
.insert_balloon_device(balloon_config)
.context("failed to insert balloon device")?;
}
// update the hotplug size
self.mem_hotplug_size_mb = req_mem_mb - self.config.memory_info.default_memory;
self.mem_hotplug_size_mb = new_mem_mb - self.config.memory_info.default_memory;
self.balloon_size = 0;
// insert a new memory device
let add_mem_mb = req_mem_mb - had_mem_mb;
let add_mem_mb = new_mem_mb - had_mem_mb;
self.vmm_instance.insert_mem_device(MemDeviceConfigInfo {
mem_id: format!("mem{}", self.mem_hotplug_size_mb),
mem_id: MEM_DEVICE_ID.to_owned(),
size_mib: add_mem_mb as u64,
capacity_mib: add_mem_mb as u64,
multi_region: false,
capacity_mib: self.hypervisor_config().memory_info.default_maxmemory as u64,
multi_region: true,
host_numa_node_id: None,
guest_numa_node_id: None,
use_shared_irq: None,
@ -384,15 +408,16 @@ impl DragonballInner {
}
Ordering::Less => {
// we only use one balloon device here, and resize it to release memory
// the operation we do here is inserting a new balloon0 device or resizing it
// the operation we do here is inserting a new BALLOON_DEVICE_ID device or resizing it
let balloon_config = BalloonDeviceConfigInfo {
balloon_id: "balloon0".to_owned(),
size_mib: (had_mem_mb - req_mem_mb) as u64,
balloon_id: BALLOON_DEVICE_ID.to_owned(),
size_mib: (had_mem_mb - new_mem_mb) as u64,
use_shared_irq: None,
use_generic_irq: None,
f_deflate_on_oom: false,
f_reporting: false,
};
self.balloon_size = had_mem_mb - new_mem_mb;
self.vmm_instance
.insert_balloon_device(balloon_config)
.context("failed to insert balloon device")?;
@ -407,7 +432,7 @@ impl DragonballInner {
};
Ok((
req_mem_mb,
new_mem_mb,
MemoryConfig {
..Default::default()
},
@ -476,6 +501,7 @@ impl Persist for DragonballInner {
capabilities: Capabilities::new(),
guest_memory_block_size_mb: 0,
mem_hotplug_size_mb: 0,
balloon_size: 0,
})
}
}

View File

@ -194,9 +194,9 @@ impl Hypervisor for Dragonball {
inner.guest_memory_block_size_mb()
}
async fn resize_memory(&self, req_mem_mb: u32) -> Result<(u32, MemoryConfig)> {
async fn resize_memory(&self, new_mem_mb: u32) -> Result<(u32, MemoryConfig)> {
let mut inner = self.inner.write().await;
inner.resize_memory(req_mem_mb)
inner.resize_memory(new_mem_mb)
}
}

View File

@ -91,7 +91,7 @@ pub trait Hypervisor: std::fmt::Debug + Send + Sync {
async fn save_vm(&self) -> Result<()>;
async fn resume_vm(&self) -> Result<()>;
async fn resize_vcpu(&self, old_vcpus: u32, new_vcpus: u32) -> Result<(u32, u32)>; // returns (old_vcpus, new_vcpus)
async fn resize_memory(&self, req_mem_mb: u32) -> Result<(u32, MemoryConfig)>;
async fn resize_memory(&self, new_mem_mb: u32) -> Result<(u32, MemoryConfig)>;
// device manager
async fn add_device(&self, device: DeviceType) -> Result<DeviceType>;

View File

@ -153,7 +153,7 @@ impl QemuInner {
todo!()
}
pub(crate) fn resize_memory(&self, _req_mem_mb: u32) -> Result<(u32, MemoryConfig)> {
pub(crate) fn resize_memory(&self, _new_mem_mb: u32) -> Result<(u32, MemoryConfig)> {
todo!()
}
}

View File

@ -173,8 +173,8 @@ impl Hypervisor for Qemu {
inner.guest_memory_block_size_mb()
}
async fn resize_memory(&self, req_mem_mb: u32) -> Result<(u32, MemoryConfig)> {
async fn resize_memory(&self, new_mem_mb: u32) -> Result<(u32, MemoryConfig)> {
let inner = self.inner.read().await;
inner.resize_memory(req_mem_mb)
inner.resize_memory(new_mem_mb)
}
}

View File

@ -9,7 +9,6 @@ use std::sync::Arc;
use anyhow::{Context, Ok, Result};
use hypervisor::Hypervisor;
use kata_types::config::TomlConfig;
use oci::LinuxResources;
use tokio::sync::RwLock;
@ -21,31 +20,17 @@ pub const MIB_TO_BYTES_SHIFT: i32 = 20;
#[derive(Default, Debug, Clone)]
pub struct MemResource {
/// Current memory
pub(crate) current_mem: Arc<RwLock<u32>>,
/// Default memory
pub(crate) orig_toml_default_mem: u32,
/// MemResource of each container
pub(crate) container_mem_resources: Arc<RwLock<HashMap<String, LinuxResources>>>,
/// Use guest swap
pub(crate) use_guest_swap: bool,
}
impl MemResource {
pub fn new(config: Arc<TomlConfig>, init_size_manager: InitialSizeManager) -> Result<Self> {
let hypervisor_name = config.runtime.hypervisor_name.clone();
let hypervisor_config = config
.hypervisor
.get(&hypervisor_name)
.context("failed to get hypervisor")?;
pub fn new(init_size_manager: InitialSizeManager) -> Result<Self> {
Ok(Self {
current_mem: Arc::new(RwLock::new(hypervisor_config.memory_info.default_memory)),
container_mem_resources: Arc::new(RwLock::new(HashMap::new())),
use_guest_swap: hypervisor_config.memory_info.enable_guest_swap,
orig_toml_default_mem: init_size_manager.get_orig_toml_default_mem(),
})
}
@ -61,36 +46,23 @@ impl MemResource {
.await
.context("update container memory resources")?;
// the unit here is MB
let (mut mem_sb_mb, need_pod_swap, swap_sb_mb) = self
.total_mems(self.use_guest_swap)
let mut mem_sb_mb = self
.total_mems()
.await
.context("failed to calculate total memory requirement for containers")?;
mem_sb_mb += self.orig_toml_default_mem;
if need_pod_swap {
mem_sb_mb += swap_sb_mb;
}
info!(sl!(), "calculate mem_sb_mb {}", mem_sb_mb);
let curr_mem = self
.do_update_mem_resource(mem_sb_mb, swap_sb_mb, hypervisor)
let _curr_mem = self
.do_update_mem_resource(mem_sb_mb, hypervisor)
.await
.context("failed to update_mem_resource")?;
self.update_current_mem(curr_mem).await;
Ok(())
}
async fn update_current_mem(&self, new_mem: u32) {
let mut current_mem = self.current_mem.write().await;
*current_mem = new_mem;
}
async fn total_mems(&self, use_guest_swap: bool) -> Result<(u32, bool, u32)> {
// sb stands for sandbox
async fn total_mems(&self) -> Result<u32> {
let mut mem_sandbox = 0;
let mut need_pod_swap = false;
let mut swap_sandbox = 0;
let resources = self.container_mem_resources.read().await;
for (_, r) in resources.iter() {
@ -100,36 +72,17 @@ impl MemResource {
if let Some(memory) = &r.memory {
// set current_limit to 0 if memory limit is not set to container
let current_limit = memory.limit.map_or(0, |limit| {
let _current_limit = memory.limit.map_or(0, |limit| {
mem_sandbox += limit as u64;
info!(sl!(), "memory sb: {}, memory limit: {}", mem_sandbox, limit);
limit
});
if let Some(swappiness) = memory.swappiness {
if swappiness > 0 && use_guest_swap {
if let Some(swap) = memory.swap {
if swap > current_limit {
swap_sandbox = swap.saturating_sub(current_limit);
}
}
// if current_limit is 0, the container will have access to the entire memory available on the host system
// so we add swap for this
else if current_limit == 0 {
need_pod_swap = true;
} else {
swap_sandbox += current_limit;
}
}
}
// TODO support memory guest swap
// https://github.com/kata-containers/kata-containers/issues/7293
}
}
Ok((
(mem_sandbox >> MIB_TO_BYTES_SHIFT) as u32,
need_pod_swap,
(swap_sandbox >> MIB_TO_BYTES_SHIFT) as u32,
))
Ok((mem_sandbox >> MIB_TO_BYTES_SHIFT) as u32)
}
// update container_cpu_resources field
@ -156,10 +109,10 @@ impl MemResource {
async fn do_update_mem_resource(
&self,
new_mem: u32,
_swap_sz_mb: u32,
hypervisor: &dyn Hypervisor,
) -> Result<u32> {
info!(sl!(), "requesting vmm to update memory to {:?}", new_mem);
let (new_memory, _mem_config) = hypervisor
.resize_memory(new_mem)
.await

View File

@ -66,7 +66,7 @@ impl ResourceManagerInner {
let cgroups_resource = CgroupsResource::new(sid, &toml_config)?;
let cpu_resource = CpuResource::new(toml_config.clone())?;
let mem_resource = MemResource::new(toml_config.clone(), init_size_manager)?;
let mem_resource = MemResource::new(init_size_manager)?;
Ok(Self {
sid: sid.to_string(),
toml_config,

View File

@ -222,7 +222,7 @@ impl VirtSandbox {
// set memory hotplug probe
if guest_details.support_mem_hotplug_probe {
self.hypervisor
.set_capabilities(CapabilityBits::GuestMemoryHotplugProbe)
.set_capabilities(CapabilityBits::GuestMemoryProbe)
.await;
}
info!(