fix: improve piptar upload - sequential uploads via background task queue (#5994)

* feat: improve piptar upload with sequential background task queue

Replace simultaneous piptar uploads with sequential processing via background job queue while keeping dependency installation parallelized.

Key changes:
- Add PiptarUpload job kind and database migration  
- Queue piptar uploads instead of using tokio::spawn
- Implement job handler for sequential S3 uploads
- Maintain parallel dependency installation as requested

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Ruben Fiszel <rubenfiszel@users.noreply.github.com>

* Remove PiptarUpload job kind migration files

Refactoring piptar uploads to use tokio channel instead of Windmill jobs.
Migration files are no longer needed.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-authored-by: Ruben Fiszel <rubenfiszel@users.noreply.github.com>

* Refactor piptar uploads from Windmill jobs to tokio channel

Replace the complex job queue infrastructure for piptar uploads with a simple 
tokio channel approach as requested. This maintains sequential upload behavior
while keeping dependency installation parallel.

Key changes:
- Add PIPTAR_UPLOAD_CHANNEL global channel for sequential processing
- Replace JobPayload::PiptarUpload with simple channel send
- Remove PiptarUpload from JobKind enum and all job handling code
- Remove job dispatcher case from worker.rs
- Simplify upload logic while maintaining same functionality

Benefits:
- Reduced complexity by removing unnecessary job infrastructure
- Sequential uploads without blocking dependency installation
- Better separation of concerns

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-authored-by: Ruben Fiszel <rubenfiszel@users.noreply.github.com>

* Update worker.rs

* Update python_executor.rs

---------

Co-authored-by: claude[bot] <209825114+claude[bot]@users.noreply.github.com>
Co-authored-by: Ruben Fiszel <rubenfiszel@users.noreply.github.com>
Co-authored-by: Ruben Fiszel <ruben@windmill.dev>
This commit is contained in:
claude[bot]
2025-06-19 15:32:21 +02:00
committed by GitHub
parent e4255e6276
commit c4adaeeabd
2 changed files with 53 additions and 3 deletions

View File

@@ -4028,6 +4028,7 @@ pub async fn push<'c, 'd>(
),
};
let final_priority: Option<i16>;
#[cfg(not(feature = "enterprise"))]
{

View File

@@ -62,12 +62,52 @@ lazy_static::lazy_static! {
static ref EPHEMERAL_TOKEN_CMD: Option<String> = var("EPHEMERAL_TOKEN_CMD").ok();
}
#[cfg(all(feature = "enterprise", feature = "parquet", unix))]
lazy_static::lazy_static! {
static ref PIPTAR_UPLOAD_CHANNEL: tokio::sync::mpsc::UnboundedSender<PiptarUploadTask> = {
let (tx, rx) = tokio::sync::mpsc::unbounded_channel();
// Spawn background task to handle uploads sequentially
tokio::spawn(handle_piptar_uploads(rx));
tx
};
}
#[cfg(all(feature = "enterprise", feature = "parquet", unix))]
#[derive(Debug)]
struct PiptarUploadTask {
venv_path: String,
cache_dir: String,
}
#[cfg(all(feature = "enterprise", feature = "parquet", unix))]
async fn handle_piptar_uploads(mut rx: tokio::sync::mpsc::UnboundedReceiver<PiptarUploadTask>) {
use crate::global_cache::build_tar_and_push;
use windmill_common::s3_helpers::get_object_store;
while let Some(task) = rx.recv().await {
if let Some(os) = get_object_store().await {
match build_tar_and_push(os, task.venv_path.clone(), task.cache_dir, None, false).await {
Ok(()) => {
tracing::info!("Successfully uploaded piptar for {}", task.venv_path);
}
Err(e) => {
tracing::error!("Failed to upload piptar for {}: {}", task.venv_path, e);
}
}
} else {
tracing::warn!("S3 object store not available for piptar upload: {}", task.venv_path);
}
}
}
const NSJAIL_CONFIG_DOWNLOAD_PY_CONTENT: &str = include_str!("../nsjail/download.py.config.proto");
const NSJAIL_CONFIG_RUN_PYTHON3_CONTENT: &str = include_str!("../nsjail/run.python3.config.proto");
const RELATIVE_PYTHON_LOADER: &str = include_str!("../loader.py");
#[cfg(all(feature = "enterprise", feature = "parquet", unix))]
use crate::global_cache::{build_tar_and_push, pull_from_tar};
use crate::global_cache::pull_from_tar;
#[cfg(all(feature = "enterprise", feature = "parquet", unix))]
use windmill_common::s3_helpers::OBJECT_STORE_SETTINGS;
@@ -1897,8 +1937,16 @@ pub async fn handle_python_reqs(
#[cfg(all(feature = "enterprise", feature = "parquet", unix))]
if s3_push {
if let Some(os) = windmill_common::s3_helpers::get_object_store().await {
tokio::spawn(build_tar_and_push(os, venv_p.clone(), py_version.to_cache_dir_top_level(false), None, false));
// Send to upload channel for sequential processing
let upload_task = PiptarUploadTask {
venv_path: venv_p.clone(),
cache_dir: py_version.to_cache_dir_top_level(false),
};
if let Err(e) = PIPTAR_UPLOAD_CHANNEL.send(upload_task) {
tracing::warn!("Failed to queue piptar upload for {venv_p}: {e}");
} else {
tracing::info!("Queued piptar upload for {venv_p}");
}
}
@@ -2200,3 +2248,4 @@ for line in sys.stdin:
)
.await
}