Run Theia with bigger UV_THREADPOOL_SIZE for better fileio performance?

[jonahgraham]

Question

Hi folks,

TL;DR is there any issue with running Theia's node process with a bigger than default UV_THREADPOOL_SIZE?

---

I am hitting some performance limitations in part because Theia's open project is on an NFS mount in my use case. The per-file operation performance when doing large number of file operations can be very slow.

A solution to this is to increase UV_THREADPOOL_SIZE so that more parallel file operations can be performed so that the iowait state can be spread.

In one of the more extreme cases I have I want to copy ~500 files (~15MB). With normal cp that takes 15-20 seconds. If I parallelize the copy, it still takes 4-5 seconds with default UV_THREADPOOL_SIZE of 4. However if I increase the size I get progressively faster, with diminishing returns after 30-60. The tradeoff is more memory usage for libuv and simply more threads.

UV_THREADPOOL_SIZE	Copy 500 files from memfs to nfs (seconds)
4	4.5
8	2.9
16	2.0
32	1.3
64	1.0
128	0.9
500	0.9

So I want to increase UV_THREADPOOL_SIZE for Theia, but before I do, I want to ensure I haven't overlooked something that others have figured out about potentially why increasing UV_THREADPOOL_SIZE in the context of Theia is a bad idea?

The only concern I have so far is that child node processes (like extension host or DAP server) inherit UV_THREADPOOL_SIZE and each of them needs too much memory or threads. Perhaps I can set process.env.UV_THREADPOOL_SIZE back to 4 after startup so that child nodes aren't too big.

See node docs and posts like this on linkedin for more info

[github-actions[bot]]

Hi @jonahgraham, thanks for starting this discussion! 👋

The Theia community will take a look soon. In the meantime, you might find helpful information in:

• 📖 Theia Documentation
• ❓ FAQ
• 💬 Previous Discussions

---

💙 Eclipse Theia is built and maintained by a community of contributors and sponsors. If Theia is valuable to your work, consider sponsoring the project. For professional support, training, or consulting services, learn more about available options.

[jonahgraham]

Instead of increasing UV_THREADPOOL_SIZE all the time, we have split up some short-lived but heavy I/O operations into a separate script and start only that sub-node with the needed increased UV_THREADPOOL_SIZE to achieve the performance results we want.

I'm going to keep an eye on overall Theia performance when running with the workspace opened on an NFS to see if there are other places that may benefit from higher levels of concurrent I/O operations.