I had a docker image build based on yolov5 with some additional code.
For the same build image, if the image is run in docker using
docker run --rm yolo-custom --env my_id=123
output
Yolo Prediction Done. (0.467s)
The prediction for each image takes less than 0.5 using cpu
However, on the same instance where the docker is run, if I create a kubernetes deployment running the same docker image. The prediction is at least >3s
output
Yolo Prediction Done. (3.300s)
Yolo Prediction Done. (11.197s)
And, if I run more than 1 instance of the kubernetes pod, the 2nd or 3rd pod will go as high as 40s or more.
Both container are run in the same instance of backend server.
I would expect the kubernetes pod to run close to the time run in a docker container