Performance issues while exporting a project ~5000 images

[R-Peleg]

Describe the bug
When trying to export a Bounding Box labeling project contains 4982 images, a gateway timeout occurred.
Raising the Nginx timeout did not resolve the problem, with more than 10 minutes passed without a response.

To Reproduce
Steps to reproduce the behavior:

1. create a project
2. Import several thousands of annotated images
3. Hit "Export", then select "COCO" and click "export"
4. See error

Expected behavior
A COCO dataset should be downloaded

Screenshots
If applicable, add screenshots to help explain your problem.

Environment (please complete the following information):

• OS: mac OS, LS run with docker-compose up
• Label Studio Version master (commit 185ed8e), with the branch fix/export-performance merged.

Additional context
Using py-spy I collected some profiling info during the request. It seems that much of the time is spent downloading the image files, so a solution might involve handling those downloads better (caching, parallel downloading). The flame SVG reports generated attached.
profiles.zip

[makseq]

@R-Peleg Could you try to use the export from the latest master branch?

[R-Peleg]

Tried on commit 1a3b1677, still got timeout

[makseq]

We made some improvements here: #1151
Could you change the latest master please? @R-Peleg

[R-Peleg]

Still timeouts.
I suspect it can be much faster, as the images are referenced from an external server, and downloading 4,000 images just take time. Maybe the solution is to make the export asynchronous so it won't be limited to the HTTP timeout and stuck the GUI meanwhile.

[twsl]

I had the same issue just exporting to COCO format.
There should be a option only to export the label file and keep the images in the cloud.

[smoreface]

This should be addressed with the version 1.3.0 release.

[makseq]

You can use download_resources in API request /api/projects//export?download_resources=0 to avoid image downloading. More info: https://labelstud.io/api#operation/api_projects_export_read