Milos Gajdos

Closing as outdated but more importantly the basics have been put in place in https://github.com/distribution/distribution/pull/4188

This introduces a recursion...which makes me think....I haven't properly dived into the code, but the fact this never cropped up even in `2.8` branch/release (https://github.com/distribution/distribution/blob/8728c52ef24fc1f4bb17dba625750cd2c9d53f75/registry/storage/driver/s3-aws/s3.go#L1340-L1371) makes this a bit peculiar...given...

@tpoxa would it be possible for you to write a test that causes S3 storage driver test to fail? We can take it from there I reckon 🤔

Circling back to this quickly. Is there any reason why you are settings this param to `false`? ``` multipartcombinesmallpart: false ```

We've recently merged a rework of S3 push path https://github.com/distribution/distribution/pull/4066 I wonder if that fixes the issue this PR attempts to address.

@pditommaso do you experience that issue when using `distribution/distribution:edge` image? `2.8.x` image is way behind the `main`

Hmm, interesting. Something still seems broken when `multipartcombinesmallpart` is set to `false`. I am still unconvinced by this patch, though, maybe @corhere has better ideas

> I'm also unconvinced by the patch. If the problem is that pending bytes aren't flushed when committing/closing the writer, why not just flush twice when committing or closing the...

> We are flushing as soon as both buffers fill up. That's what I said in point 2 > ...and while it keeps writing we are maintaining the "overflow" data...

> I'm guessing you mean the enforcing of Conventional Commit messaging. It does create some friction which I also don't really like. Yes, that's what I meant. There are already...