Allow deframer errors to close stream with a status code

[ryanpbrewster]

Relevant to #3996

Today, if the deframer encounters an error it will trigger a stream reset. This does not provide the peer with much information about why the stream was reset. In particular, if a client sends an oversized message, the server will
reject the request, but the client will see that as a CANCELLED status code rather than a RESOURCE_EXHAUSTED.

grpc-go servers do provide this information to clients, and I think that's helpful.

This PR is an attempt to provide best-effort status codes to peers on deframer errors. @ejona86 explained (here) that it will not always be possible for the stream to gracefully close with properly encoded trailers. In these cases the peer will see garbled data. But in some circumstances — including the very prevalent case of a unary request payload that is too large — the status will be well-formed and useful.

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: ryanpbrewster / name: Ryan Brewster (fc49d80)

[ryanpbrewster]

Tests & linters finally seem like they're passing.

@ejona86 can you recommend someone who could review this PR and help audit for other parts of the Netty implementation that need to be adjusted?

[ryanpbrewster]

I missed the bit in CONTRIBUTING.md about maintaining a clean commit history. Tidied this up.

This change did not affect any existing tests. I added a new test specifically for this behavior, but it's fairly low-level. I tested manually this end-to-end by sending an oversized request and validating that I now get a RESOURCE_EXHAUSTED instead of a CANCEL

[ryanpbrewster]

I think that :grpc-servlet:undertowTest is flaky, that test seems unrelated to any of the changes in this PR, and it passes locally.

[ejona86]

This looks good, with one null check needed

[ejona86]

I think that :grpc-servlet:undertowTest is flaky

Yeah, don't worry about it.

[ejona86]

Thank you!

[ejona86]

Easy build failure:

Line is longer than 100 characters

[ryanpbrewster]

Someday I'll push a commit that doesn't have a formatting error, but apparently not today :)

[ryanpbrewster]

I'm used to squash-merging. Anything I need to do on my end to get this PR merged?

[ejona86]

We would squash when we merge. But you can squash yourself if you want more control of the commit message. (And this is a fine time to squash, as it won't make the review harder.) Looks like the commit message on your first commit is still accurate. I would probably end up prefixing the git commit title with "netty: " and append Fixes #3996. But that's minor stuff.

[ryanpbrewster]

Squashed everything down and made the formatting suggestions 👍