github/kata-containers
1
0
Fork 1
mirror of https://github.com/kata-containers/kata-containers.git synced 2025-08-14 06:06:12 +00:00
Code Issues Projects Releases Wiki Activity

Merge pull request #5729 from openanolis/netnsref

Browse Source 
runtime-rs: block on the current thread when setup the network to avoid be take over by other task
This commit is contained in:
Zhongtao Hu 2022-11-25 08:09:10 +08:00 committed by GitHub
commit f02bb1a9cb
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 29 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
src/runtime-rs/crates/resource/src/manager_inner.rs
View File

@ -4,17 +4,18 @@
// SPDX-License-Identifier: Apache-2.0
//
use std::sync::Arc;
use std::{sync::Arc, thread};
use crate::resource_persist::ResourceState;
use agent::{Agent, Storage};
use anyhow::{Context, Result};
use anyhow::{anyhow, Context, Result};
use async_trait::async_trait;
use hypervisor::Hypervisor;
use kata_types::config::TomlConfig;
use kata_types::mount::Mount;
use oci::LinuxResources;
use persist::sandbox_persist::Persist;
use tokio::runtime;
use crate::{
cgroups::{CgroupArgs, CgroupsResource},
@ -88,11 +89,32 @@ impl ResourceManagerInner {
};
}
ResourceConfig::Network(c) => {
let d = network::new(&c).await.context("new network")?;
d.setup(self.hypervisor.as_ref())
.await
.context("setup network")?;
self.network = Some(d)
// 1. When using Rust asynchronous programming, we use .await to
// allow other task to run instead of waiting for the completion of the current task.
// 2. Also, when handling the pod network, we need to set the shim threads
// into the network namespace to perform those operations.
// However, as the increase of the I/O intensive tasks, two issues could be caused by the two points above:
// a. When the future is blocked, the current thread (which is in the pod netns)
// might be take over by other tasks. After the future is finished, the thread take over
// the current task might not be in the pod netns. But the current task still need to run in pod netns
// b. When finish setting up the network, the current thread will be set back to the host namespace.
// In Rust Async, if the current thread is taken over by other task, the netns is dropped on another thread,
// but it is not in netns. So, the previous thread would still remain in the pod netns.
// The solution is to block the future on the current thread, it is enabled by spawn an os thread, create a
// tokio runtime, and block the task on it.
let hypervisor = self.hypervisor.clone();
let network = thread::spawn(move || -> Result<Arc<dyn Network>> {
let rt = runtime::Builder::new_current_thread().enable_io().build()?;
let d = rt.block_on(network::new(&c)).context("new network")?;
rt.block_on(d.setup(hypervisor.as_ref()))
.context("setup network")?;
Ok(d)
})
.join()
.map_err(|e| anyhow!("{:?}", e))
.context("Couldn't join on the associated thread")?
.context("failed to set up network")?;
self.network = Some(network);
}
};
}