Support for retry?

[vsoch]

hey oras developers!

I've been using oras in CI, and I typically need to wrap it in some kind of retry with exponential backoff. To compare to another registry client, when I use docker it will retry pulling blobs, and same with Singularity. How could it work? I suspect there could be some Retry-After header returned from a server, and if not, the user could specify a number of retries. Would this be something oras could support? In practice I have seen a lot of push failures in my CI when using oras, and probably just the fault of the receiving end.

[shizhMSFT]

Thanks for mentioning the retry. The retry functionality is planned after the release of v2 as an enhancement.

Retrying seems to be a simple feature, but it is not. There are two levels of retry: HTTP level retry and application level retry.

• HTTP Level: Not all requests are retriable. Only GET and HEAD requests can be retried. Retrying PUT/PATCH/POST requests is problematic. For uploading the blob, the POST request is retriable but PATCH and PUT are not.
• Application Level: The retry logic is at the application side as the application as more context on the operation (e.g. blob upload). For instance, the retry logic is in the dockerd code based and is not in the registry client.

We will probably add retry logic in the remote package but before that, we need to have a proper design first.

[vsoch]

Excellent! And yes of course a good design goes without saying.

[souleb]

@shizhMSFT I am interested in implementing the retry logic in the remote pkg. What is the process to propose a design?

[FeynmanZhou]

@shizhMSFT I am interested in implementing the retry logic in the remote pkg. What is the process to propose a design?

@souleb Thanks for your interest! ORAS community is open to any kind of proposal and design. Now you can create a discussion to add your proposal.

Going forward, we will create a separate folder to store the proposal docs and template. What do you think of it?

[souleb]

Great! Thanks.

[shizhMSFT]

Discussed in #232