github/kata-containers
1
0
Fork 2
mirror of https://github.com/kata-containers/kata-containers.git synced 2026-07-01 22:50:54 +00:00
Code Issues Projects Releases Wiki Activity

runtime-rs: Fix vCPU pinning race with backoff retry

Browse Source 
QEMU can report fewer vCPU threads during early startup, causing partial
affinity setup. Let's retry with exponential backoff until the expected
thread count is visible, then continue with best-effort pinning if the
window is exhausted.

Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
This commit is contained in:
Fabiano Fidêncio
2026-05-22 14:48:42 +02:00
committed by Fabiano Fidêncio
parent f53f427859
commit feeb5d8ecc
1 changed files with 42 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

48
src/runtime-rs/crates/resource/src/cgroups/resource_inner.rs
View File

@@ -7,6 +7,7 @@
use std::collections::{HashMap, HashSet};
use std::process;
use std::str::FromStr;
use std::time::Duration;
use anyhow::{anyhow, Context, Result};
use cgroups_rs::manager::is_systemd_cgroup;
@@ -16,6 +17,7 @@ use kata_types::cpu::CpuSet;
use nix::sched::{sched_setaffinity, CpuSet as NixCpuSet};
use nix::unistd::Pid;
use oci_spec::runtime::{LinuxCpu, LinuxCpuBuilder, LinuxResources, LinuxResourcesBuilder};
use tokio::time::sleep;
use crate::cgroups::utils::get_tgid_from_pid;
use crate::cgroups::CgroupConfig;
@@ -187,12 +189,46 @@ impl CgroupsResourceInner {
let needs_thread_ids = self.overhead_cgroup.is_some() || self.enable_vcpus_pinning;
let thread_ids = if needs_thread_ids {
Some(
hypervisor
.get_thread_ids()
.await
.context("get vCPU thread IDs")?,
)
let mut tids = hypervisor
.get_thread_ids()
.await
.context("get vCPU thread IDs")?;
// QEMU may not have spawned all vCPU threads yet. Retry with
// exponential backoff until we see the expected count.
let expected = hypervisor.hypervisor_config().await.cpu_info.default_vcpus.ceil() as usize;
if expected > 0 && tids.vcpus.len() < expected {
const MAX_ATTEMPTS: u32 = 10;
let mut backoff = Duration::from_millis(50);
for attempt in 2..=MAX_ATTEMPTS {
if tids.vcpus.len() >= expected {
break;
}
info!(
sl!(),
"waiting for all vCPU threads: have {}, want {}, attempt {}",
tids.vcpus.len(),
expected,
attempt
);
sleep(backoff).await;
backoff *= 2;
tids = hypervisor
.get_thread_ids()
.await
.context("get vCPU thread IDs (retry)")?;
}
if tids.vcpus.len() < expected {
warn!(
sl!(),
"not all vCPU threads available after retries: have {}, want {}; pinning available ones",
tids.vcpus.len(),
expected
);
}
}
Some(tids)
} else {
None
};