A registry configured in pull-through cache (mirror) mode is designed to handle multiple concurrent requests for the same image or layer efficiently.

For the first request for a specific image layer:

• The mirror will realize it does not have the content locally.
• It will make a single request to the upstream registry (e.g., Docker Hub) to fetch that layer.
• While the layer is being pulled, the registry typically employs mechanisms to block or wait for other concurrent requests for the exact same layer to prevent duplicate upstream pulls.

For subsequent requests for the same image or layer:

• Once the mirror has successfully fetched and stored the layer, all subsequent requests (either from different clients or the same client later on) will be served from its local cache.
• This greatly reduces the number of requests to the upstream registry and improves image pull speed for clients within your network.

In summary, for concurrent requests:

• The very first concurrent pull will trigger the fetch from the upstream.
• Other simultaneous pulls for the same content should typically wait for the first pull to complete, and then all will use the newly cached local content, resulting in only one pull to the upstream for that layer/blob from the mirror's perspective.

Find out more about how a registry can be configured as a pull-through cache in this video on Mirroring and Caching for Container Registries.