Derek McGowan

We haven't listed registries that return a 405 in the past because 405 is the correct HTTP response code for not supporting the method. The non-405s are the non-standard ones...

This will be a painful one to keep rebasing. Either one is fine but `imagespec` is probably more consistent, can we update that and just get this one merged since...

`time="2024-01-16T19:13:35.389541880Z" level=warning msg="failed to load plugin" error="failed to get \"io.containerd.transfer.v1\" plugin: no plugins registered for io.containerd.transfer.v1: plugin: not found" id=io.containerd.internal.v1.restart type=io.containerd.internal.` transfer service likely needed in `integration/build_local_containerd_helper_test.go` now, since cri...

@fangn2 I see something similar locally as well, I'll try to help you look into it more. Either way it is the right change to get the transfer service in...

@helen-frank I'm curious if the difference in that benchmark is just from moving the allocation of `addrError`. If you just make that change, is it the same result? I think...

I see, I didn't look at the benchmarks closer. The result is faster because it returns quickly when the strings are `localhost` or `127.0.0.1`. It feels like a bit of...

Are there specific errors you were encountering originally that made this necessary? The `ErrUnexpectedEOF` is treated as ignore and retry, but other errors we may not want to ignore (either...

> Because we are writing the full content length, net/http is going to implicitly make the request. Doesn't seem like we should rely on that and should actually Close the...

The discussion this one needs might be better in the OCI group rather than here. The OCI spec defines the chunked uploading technique using `PATCH`, as mentioned in #3533. However,...

@z4ce I think the best long term solution is to use PATCH for larger blobs and break up the upload into individual PATCH requests of pre-defined static size. The advantage...