Does caching proxy work on unavailable upstream?

[Felixoid]

Hello. According to manual, the registry mirror requests if the images updated on the ustream.

We use docker very intensively in our CI, and quite often affected by issues like this:

Exception: Command ['docker-compose', '--env-file', '/ClickHouse/tests/integration/test_cleanup_dir_after_bad_zk_conn/_instances_0/.env', '--project-name', 'roottestcleanupdirafterbadzkconn', '--file', '/ClickHouse/tests/integration/test_cleanup_dir_after_bad_zk_conn/_instances_0/node1/docker-compose.yml', '--file', '/compose/docker_compose_keeper.yml', 'pull'] return non-zero code 1: Pulling zoo2  ... 
Pulling zoo3  ... 
Pulling zoo1  ... 
Pulling node1 ... 
Pulling zoo3  ... error
Pulling zoo1  ... error
Pulling zoo2  ... error
Pulling node1 ... error

ERROR: for zoo3  Get "https://registry-1.docker.io/v2/": net/http: request canceled while waiting for connection (Client.Timeout exceeded while awaiting headers)

ERROR: for zoo1  Get "https://registry-1.docker.io/v2/": net/http: request canceled while waiting for connection (Client.Timeout exceeded while awaiting headers)

ERROR: for zoo2  Get "https://registry-1.docker.io/v2/": context deadline exceeded

ERROR: for node1  Get "https://registry-1.docker.io/v2/": net/http: request canceled while waiting for connection (Client.Timeout exceeded while awaiting headers)
Get "https://registry-1.docker.io/v2/": net/http: request canceled while waiting for connection (Client.Timeout exceeded while awaiting headers)
Get "https://registry-1.docker.io/v2/": net/http: request canceled while waiting for connection (Client.Timeout exceeded while awaiting headers)
Get "https://registry-1.docker.io/v2/": context deadline exceeded
Get "https://registry-1.docker.io/v2/": net/http: request canceled while waiting for connection (Client.Timeout exceeded while awaiting headers)

And then a huge number of jobs fail.

Do I understand correctly, that registry:2 (the current latest) can't be setup to serve requests with what it has w/o failing?

p.s. It's a moved issue from distribution/distribution-library-image#141

[Felixoid]

The issue affects us pretty badly. We have a vast testing infrastructure, so the whole thing fails as a card house when there's a connection issue between proxy and docker hub. Here another case ClickHouse/ClickHouse#40241 (comment)

[alexey-milovidov]

This issue affects us a few times a day.

[milosgajdos]

Registry proxy has not seen much TLC in a while, unfortunately, but we are open to accepting PRs.

[Felixoid]

Can you point to a code where I could start my research, please?

As well, I've tried to find what TLC means, but it's still a bit unclear

[Jamstah]

TLC=Tender Love and Care, it basically means no one has put any work into making it better for a while.

The code for a proxying registry is here: https://github.com/distribution/distribution/tree/main/registry/proxy