New to Google Container Registry and Docker ecosystem in general. I'm pushing an existing image to gcr.io and I'd expect the time to complete the task to be close to 0 seconds, as all the bits are already on gcr.io. The context is running dev code in the Cloud, on lots of cores at the same time, as opposed to the 4 cores my Mac laptop has. I'm running a no-op to isolate the bottlenecks, the real usage has about 6M new bytes. It is slow, 14 seconds to perform a no-op. Is there a way to cut this no-op down to less than a second?

$ time gcloud docker push gcr.io/ai2-general/euclid:latest
WARNING: login credentials saved in /Users/cristipp/.docker/config.json
Login Succeeded
WARNING: login credentials saved in /Users/cristipp/.docker/config.json
Login Succeeded
WARNING: login credentials saved in /Users/cristipp/.docker/config.json
Login Succeeded
WARNING: login credentials saved in /Users/cristipp/.docker/config.json
Login Succeeded
WARNING: login credentials saved in /Users/cristipp/.docker/config.json
Login Succeeded
WARNING: login credentials saved in /Users/cristipp/.docker/config.json
Login Succeeded
WARNING: login credentials saved in /Users/cristipp/.docker/config.json
Login Succeeded
The push refers to a repository [gcr.io/ai2-general/euclid]
3a67b2b013f5: Layer already exists 
b7c8985fbf02: Layer already exists 
fef418d1a9e8: Layer already exists 
c58360ce048c: Layer already exists 
0030e912789f: Layer already exists 
5f70bf18a086: Layer already exists 
0ece0aa9121d: Layer already exists 
ef63204109e7: Layer already exists 
694ead1cbb4d: Layer already exists 
591569fa6c34: Layer already exists 
998608e2fcd4: Layer already exists 
c12ecfd4861d: Layer already exists 
latest: digest: sha256:04a831f4bf3e3033c40eaf424e447dd173e233329440a3c9796bf1515225546a size: 10321

real    0m14.742s
user    0m0.622s
sys 0m0.181s

14 seconds is long time. Using plain docker push is faster, but still wastes 5 precious seconds.

$ time docker push gcr.io/ai2-general/euclid:latest
The push refers to a repository [gcr.io/ai2-general/euclid]
3a67b2b013f5: Layer already exists 
b7c8985fbf02: Layer already exists 
fef418d1a9e8: Layer already exists 
c58360ce048c: Layer already exists 
0030e912789f: Layer already exists 
5f70bf18a086: Layer already exists 
0ece0aa9121d: Layer already exists 
ef63204109e7: Layer already exists 
694ead1cbb4d: Layer already exists 
591569fa6c34: Layer already exists 
998608e2fcd4: Layer already exists 
c12ecfd4861d: Layer already exists 
latest: digest: sha256:04a831f4bf3e3033c40eaf424e447dd173e233329440a3c9796bf1515225546a size: 10321

real    0m5.014s
user    0m0.030s
sys 0m0.011s

I suspect the difference is caused by the 7 login attempts, which take a while to process, afterwards it feels like the docker push overhead.

For reference:

$ gcloud --version
Google Cloud SDK 107.0.0

bq 2.0.24
bq-nix 2.0.24
core 2016.04.21
core-nix 2016.03.28
gcloud 
gsutil 4.19
gsutil-nix 4.18
kubectl 
kubectl-darwin-x86_64 1.2.2