cvat: CVAT fails health check using >90% disk

@harrystuart , you can adjust health check limits (https://github.com/revsys/django-health-check/blob/master/health_check/conf.py)

Please try to define in cvat/settings/base.py the following variable:

HEALTH_CHECK = {
    "DISK_USAGE_MAX": 99
}

@harrystuart, @Harsturomai, @Ghaitharar, hi, could you please share and check the output of the following commands in the terminal:

docker exec -it -u django cvat_server python -c 'import os; print(os.statvfs("/"))'
docker exec -it -u django cvat_server python -c 'import psutil; print(psutil.disk_usage("/"))'
docker exec -it -u django cvat_server df -h /

I used CVAT on windows 10 to annotated a number of .mp4 videos that are collectively less than 1GB in size. When trying to extract the dataset, CVAT consumes more than 100GB and then report

@Ghaitharar, it can be related to video frame unpacking. If I understand correctly, you’re trying to export a task with images. In this case the video images will be saved on disk and then packed into an archive. If you’re exporting a task, please try to turn off the Save images option.

In general, I see few ways to handle the problem:

• export just annotations, turn the Save images checkbox off
• turn off the use chunks switch during task creation
• check for the exported dataset cache and clean it with this command:

  docker exec -it cvat_server bash -c 'rm -r data/*/*/export_cache/'
  

  (if you see rm: cannot remove 'data/*/*/export_cache/': No such file or directory, everything is OK)
• try to clear the cache with this command:

  docker exec -it cvat_server bash -c 'rm -r data/cache/'
  

• turn off the use cache checkbox (note that this may slow down the workflow with videos)

• Need to add into documentation more information about health check and its configuration
• Need to fix creation of tasks with video files. By default, they are created with zip chunks with images

HEALTH_CHECK = { “DISK_USAGE_MAX”: 99 }

Are these settings somehow injectable without baking a new cvat docker image? I am “suffering” from the same issue, since my K8s nodes have little disk space, thus running into the limit preventing the webserver to start, even though cvat-backend-data volume is placed on another machine via nfs volume and has plenty space.

Is there a way to turn off use chunks after a task is created? I’v already created a number of tasks with this option enabled.

Unfortunately, no. I can suggest you to export the annotations without images, and then you can download images separately using SDK or CLI: cvat-cli frames --quality original --outdir task_<N>_images <task_id> (modify the command to fit your case).