From: "Lukas Wagner" <l.wagner@proxmox.com>
To: "Dominik Csapak" <d.csapak@proxmox.com>,
"Proxmox Datacenter Manager development discussion"
<pdm-devel@lists.proxmox.com>
Subject: Re: [pdm-devel] [RFC proxmox-datacenter-manager 1/2] server: add convenience helper to fetch results from many remote nodes in parallel
Date: Fri, 29 Aug 2025 16:05:41 +0200 [thread overview]
Message-ID: <DCEYWBLN6265.5IZA3NFZYZ1J@proxmox.com> (raw)
In-Reply-To: <5fca5c4c-cdbb-4cbf-b3a0-475f2d7f1f44@proxmox.com>
On Fri Aug 29, 2025 at 10:57 AM CEST, Dominik Csapak wrote:
> some minor comments inline, but looks good in general to me
>
> On 8/28/25 4:42 PM, Lukas Wagner wrote:
>> Implementing this over and over again is a lot of work; many parts of
>> PDM need to fetch some kind of resource from all remotes and all nodes
>> of these remotes.
>>
>> The general approach is the same as for remote task fetching (JoinSet
>> and semaphores to control concurrency), it will be refactored in a later
>> commit to use this new helper.
>>
>> Signed-off-by: Lukas Wagner <l.wagner@proxmox.com>
>> ---
>> server/src/lib.rs | 1 +
>> server/src/parallel_fetcher.rs | 236 +++++++++++++++++++++++++++++++++
>> 2 files changed, 237 insertions(+)
>> create mode 100644 server/src/parallel_fetcher.rs
>>
>> diff --git a/server/src/lib.rs b/server/src/lib.rs
>> index 485bc792..3f8b7708 100644
>> --- a/server/src/lib.rs
>> +++ b/server/src/lib.rs
>> @@ -6,6 +6,7 @@ pub mod auth;
>> pub mod context;
>> pub mod env;
>> pub mod metric_collection;
>> +pub mod parallel_fetcher;
>> pub mod remote_cache;
>> pub mod remote_tasks;
>> pub mod resource_cache;
>> diff --git a/server/src/parallel_fetcher.rs b/server/src/parallel_fetcher.rs
>> new file mode 100644
>> index 00000000..3fad68a0
>> --- /dev/null
>> +++ b/server/src/parallel_fetcher.rs
>> @@ -0,0 +1,236 @@
>> +use std::{
>> + collections::HashMap,
>> + fmt::Debug,
>> + future::Future,
>> + sync::Arc,
>> + time::{Duration, Instant},
>> +};
>> +
>> +use anyhow::Error;
>> +use pdm_api_types::remotes::{Remote, RemoteType};
>> +use pve_api_types::ClusterNodeIndexResponse;
>> +use tokio::{
>> + sync::{OwnedSemaphorePermit, Semaphore},
>> + task::JoinSet,
>> +};
>> +
>> +use crate::connection;
>> +
>> +pub struct ParallelFetcher<C> {
>> + pub max_connections: usize,
>> + pub max_connections_per_remote: usize,
>> + pub context: C,
>> +}
>> +
>> +pub struct FetchResults<T> {
>> + pub remote_results: HashMap<String, Result<RemoteResult<T>, Error>>,
>> +}
>> +
>> +impl<T> Default for FetchResults<T> {
>> + fn default() -> Self {
>> + Self {
>> + remote_results: Default::default(),
>> + }
>> + }
>> +}
>> +
>> +#[derive(Debug)]
>> +pub struct RemoteResult<T> {
>> + pub node_results: HashMap<String, Result<NodeResults<T>, Error>>,
>> +}
>> +
>> +impl<T> Default for RemoteResult<T> {
>> + fn default() -> Self {
>> + Self {
>> + node_results: Default::default(),
>> + }
>> + }
>> +}
>> +
>> +#[derive(Debug)]
>> +pub struct NodeResults<T> {
>> + pub data: T,
>> + pub api_response_time: Duration,
>> +}
>> +
>> +impl<C: Clone + Send + 'static> ParallelFetcher<C> {
>> + pub fn new(context: C) -> Self {
>> + Self {
>> + max_connections: 20,
>> + max_connections_per_remote: 5,
>
> you probably thought of it already, but we probably want to
> make this settable from outside in a seperate constructor, or at least
> put the values in const values for now
I've extracted these into constants for v2, but I think for the future a
builder pattern could be nice for this one.
Stefan and I are playing around with some ideas on how to make the
whole thing a bit nicer to use in the call sites, but that's something
for a future series; since we both need this helper for our code we've
settled for something basic first, which can then be improved/refactored
later.
>
>> + context,
>> + }
>> + }
>> +
>> + pub async fn do_for_all_remote_nodes<A, F, T, Ft>(self, remotes: A, func: F) -> FetchResults<T>
>> + where
>> + A: Iterator<Item = Remote>,
>> + F: Fn(C, Remote, String) -> Ft + Clone + Send + 'static,
>> + Ft: Future<Output = Result<T, Error>> + Send + 'static,
>> + T: Send + Debug + 'static,
>> + {
>> + let total_connections_semaphore = Arc::new(Semaphore::new(self.max_connections));
>> +
>> + let mut remote_join_set = JoinSet::new();
>> +
>> + for remote in remotes {
>> + let semaphore = Arc::clone(&total_connections_semaphore);
>> +
>> + let f = func.clone();
>> +
>> + remote_join_set.spawn(Self::fetch_remote(
>> + remote,
>> + self.context.clone(),
>> + semaphore,
>> + f,
>> + self.max_connections_per_remote,
>> + ));
>
> not sure if it would work properly, but couldn't we put the semaphores
> into self, and clone that instead?
>
> then `fetch_remote` could become a normal method on self
> and we could implement a helper for getting permits and
> we wouldn't have to pass the semaphores around all the time.
>
> (maybe that does not work as i think though..)
Good idea actually, but I'd also revisit that a bit later; I'd prefer to
get this applied as soon as possible, so that Stefan and I can build on
it.
>
>> + }
>> +
>> + let mut results = FetchResults::default();
>> +
>> + while let Some(a) = remote_join_set.join_next().await {
>> + match a {
>> + Ok((remote_name, remote_result)) => {
>> + results.remote_results.insert(remote_name, remote_result);
>> + }
>> + Err(err) => {
>> + log::error!("join error when waiting for future: {err}")
>> + }
>> + }
>> + }
>> +
>> + results
>> + }
>> +
>> + async fn fetch_remote<F, Ft, T>(
>> + remote: Remote,
>> + context: C,
>> + semaphore: Arc<Semaphore>,
>> + func: F,
>> + max_connections_per_remote: usize,
>> + ) -> (String, Result<RemoteResult<T>, Error>)
>> + where
>> + F: Fn(C, Remote, String) -> Ft + Clone + Send + 'static,
>> + Ft: Future<Output = Result<T, Error>> + Send + 'static,
>> + T: Send + Debug + 'static,
>> + {
>> + let mut per_remote_results = RemoteResult::default();
>> +
>> + let mut permit = Some(Arc::clone(&semaphore).acquire_owned().await.unwrap());
>> + let per_remote_semaphore = Arc::new(Semaphore::new(max_connections_per_remote));
>> +
>> + match remote.ty {
>> + RemoteType::Pve => {
>> + let remote_clone = remote.clone();
>> +
>> + let nodes = match async move {
>> + let client = connection::make_pve_client(&remote_clone)?;
>> + let nodes = client.list_nodes().await?;
>> +
>> + Ok::<Vec<ClusterNodeIndexResponse>, Error>(nodes)
>> + }
>> + .await
>> + {
>> + Ok(nodes) => nodes,
>> + Err(err) => return (remote.id.clone(), Err(err)),
>> + };
>> +
>> + let mut nodes_join_set = JoinSet::new();
>> +
>> + for node in nodes {
>> + let permit = if let Some(permit) = permit.take() {
>> + permit
>> + } else {
>> + Arc::clone(&semaphore).acquire_owned().await.unwrap()
>> + };
>> +
>> + let per_remote_connections_permit = Arc::clone(&per_remote_semaphore)
>> + .acquire_owned()
>> + .await
>> + .unwrap();
>> +
>> + let func_clone = func.clone();
>> + let remote_clone = remote.clone();
>> + let node_name = node.node.clone();
>> + let context_clone = context.clone();
>> +
>> + nodes_join_set.spawn(Self::fetch_pve_node(
>> + func_clone,
>> + context_clone,
>> + remote_clone,
>> + node_name,
>> + permit,
>> + per_remote_connections_permit,
>> + ));
>> + }
>> +
>> + while let Some(join_result) = nodes_join_set.join_next().await {
>> + match join_result {
>> + Ok((node_name, per_node_result)) => {
>> + per_remote_results
>> + .node_results
>> + .insert(node_name, per_node_result);
>> + }
>> + Err(e) => {
>> + log::error!("join error when waiting for future: {e}")
>> + }
>> + }
>> + }
>> + }
>> + RemoteType::Pbs => {
>> + let node = "localhost".to_string();
>> +
>> + let now = Instant::now();
>> + let result = func(context, remote.clone(), node.clone()).await;
>> + let api_response_time = now.elapsed();
>> +
>> + match result {
>> + Ok(data) => {
>> + per_remote_results.node_results.insert(
>> + node,
>> + Ok(NodeResults {
>> + data,
>> + api_response_time,
>> + }),
>> + );
>> + }
>> + Err(err) => {
>> + per_remote_results.node_results.insert(node, Err(err));
>> + }
>> + }
>> + }
>> + }
>> +
>> + (remote.id, Ok(per_remote_results))
>> + }
>
> probably a `fn do_for_all_remotes` that only does one thing per remote
> would also be nice? (i think stefan said he'll need something like this)
Will be added for v2, I adopted (and slightly modified, after some
discussion) Stefan's patches for that
>> +
>> + async fn fetch_pve_node<F, Ft, T>(
>> + func: F,
>> + context: C,
>> + remote: Remote,
>> + node: String,
>> + _permit: OwnedSemaphorePermit,
>> + _per_remote_connections_permit: OwnedSemaphorePermit,
>> + ) -> (String, Result<NodeResults<T>, Error>)
>> + where
>> + F: Fn(C, Remote, String) -> Ft + Clone + Send + 'static,
>> + Ft: Future<Output = Result<T, Error>> + Send + 'static,
>> + T: Send + Debug + 'static,
>> + {
>> + let now = Instant::now();
>> + let result = func(context, remote.clone(), node.clone()).await;
>> + let api_response_time = now.elapsed();
>> +
>> + match result {
>> + Ok(data) => (
>> + node,
>> + Ok(NodeResults {
>> + data,
>> + api_response_time,
>> + }),
>> + ),
>> + Err(err) => (node, Err(err)),
>> + }
>> + }
>> +}
_______________________________________________
pdm-devel mailing list
pdm-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pdm-devel
next prev parent reply other threads:[~2025-08-29 14:05 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-08-28 14:42 [pdm-devel] [RFC proxmox-datacenter-manager 0/2] add helper to fetch " Lukas Wagner
2025-08-28 14:42 ` [pdm-devel] [RFC proxmox-datacenter-manager 1/2] server: add convenience helper to fetch results " Lukas Wagner
2025-08-29 8:57 ` Dominik Csapak
2025-08-29 9:02 ` Stefan Hanreich
2025-08-29 9:17 ` Lukas Wagner
2025-08-29 14:05 ` Lukas Wagner [this message]
2025-08-28 14:42 ` [pdm-devel] [RFC proxmox-datacenter-manager 2/2] remote task fetching: use ParallelFetcher helper Lukas Wagner
2025-08-29 9:15 ` Dominik Csapak
2025-08-29 9:19 ` Lukas Wagner
2025-08-29 14:11 ` [pdm-devel] superseded: [RFC proxmox-datacenter-manager 0/2] add helper to fetch from many remote nodes in parallel Lukas Wagner
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=DCEYWBLN6265.5IZA3NFZYZ1J@proxmox.com \
--to=l.wagner@proxmox.com \
--cc=d.csapak@proxmox.com \
--cc=pdm-devel@lists.proxmox.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox