Allow site wide configuration settings for all users

[lgarrison]

Occasionally users on our cluster will open a large directory tree (10s of millions of files) on a shared network filesystem (CephFS) as a workspace. This causes noticeable latency for all users of the filesystem, as unfortunately network filesystems have notoriously bad performance isolation. I believe this happens during the initial indexing of the files in the directory tree, since setting

"files.watcherExclude": {
    "**/ceph/**": true,
}

doesn't help. I can watch the fileWatcher process indexing the directory tree via strace regardless of this setting. (Even if it did help, I don't think we have a way to deploy this as a site-wide setting.) For what it's worth, CephFS doesn't support inotify, but if I'm understanding what I'm seeing correctly, that doesn't really matter since the issue happens during the initial crawl.

While we continue to identify these issues case-by-case and try to educate users about workspaces, I'm wondering if there could be a technical solution. I think the ideal behavior might be that we could set a site-wide "blacklist" of directories to never crawl preemptively. I realize this might affect the behavior of extensions that want to discover properties of the workspace, but in our case this would be acceptable -- the CephFS filesystem is for storing large data files, not code, so it doesn't need to function properly as a part of a workspace.

Is such a feature be possible? Or is there another approach that you would recommend for issues like this? Thanks!

[bpasero]

Can you do a test by entirely disabling the file watcher via:

"files.watcherExclude": {
    "**": true
}

[lgarrison]

That does seem to stop the crawl! And changing it back to **/ceph/** immediately triggers it. Here's what I observed:

1. Set "files.watcherExclude": { "**": true }
2. On a cluster node via Remote-SSH, open /mnt/home/lgarrison/ceph as a workspace, which is a symlink to /mnt/ceph/users/lgarrison
3. Check strace -f -p <fileWatcher PID>, don't see stats
4. Set "files.watcherExclude": { "**/ceph/**": true }
5. Immediately see stat flood in strace

This happens regardless of whether I open the workspace as /mnt/home/lgarrison/ceph (the symlink) or /mnt/ceph/users/lgarrison (the target), for what it's worth.

However, taking a step back, even if watcherExclude could be set to stop watching ceph paths, is there any way to deploy such a setting site-wide? We have new users connecting all the time from their personal devices where we don't control the VS Code settings.

[bpasero]

No, not to my knowledge.

[lgarrison]

Okay, in that case I think this issue has 2 parts, a bug/question and a feature request.

Bug/question: why doesn't "files.watcherExclude": { "**/ceph/**": true } stop crawls of /mnt/ceph/users/lgarrison as a workspace? What's the right syntax?

Feature request: a way to deploy a files.watcherExclude setting site-wide, or perhaps more generically inherit a site-wide settings.json. I don't mind if the user has the capability to override site-wide settings (in fact, that's probably better); we're just trying to set smart defaults.

[bpasero]

I hope this clarifies it:

Turning into a feature request for the second part.

[vs-code-engineering]

This feature request is now a candidate for our backlog. The community has 60 days to upvote the issue. If it receives 20 upvotes we will move it to our backlog. If not, we will close it. To learn more about how we handle feature requests, please see our documentation.

Happy Coding!

[lgarrison]

Thanks! And just to confirm for anyone else since it wasn't immediately obvious to me: setting "files.watcherExclude": { "/mnt/ceph/": true } does appear to effectively exclude the entire directory tree rooted at that point, including any paths that point into that tree via symlinks.