ext_proc: Implement STREAMED body processing mode

[gbrail]

Commit Message: Implement STREAMED body processing mode in ext_proc

Additional Description: Implement this mode, which allows an external processor to receive the body one chunk at a time and examine or modify the body while it is being processed.

Risk Level: Medium. The mode is only engaged if used. However an external processor must be coded carefully, particularly if it expects to stream both request and response bodies simultaneously, since each will proceed in its own ordering.

Testing: New unit and integration tests.

Docs Changes: Updated proto docs to reflect that the mode is now supported.

Release Notes: External processing servers may now use the STREAMED processing mode. In this mode, chunks of the body are forwarded to the external processing server when they arrive. Depending on how the upstream system is implemented, request body chunks may be delivered before or after the response headers, and request and response body chunks may be interleaved if the upstream system delivers them that way. An external processor should be carefully coded so that it does not assume that a particular ordering will be implemented.

[repokitteh-read-only]

CC @envoyproxy/api-shepherds: Your approval is needed for changes made to api/envoy/.
envoyproxy/api-shepherds assignee is @htuch
CC @envoyproxy/api-watchers: FYI only for changes made to api/envoy/.

🐱

Caused by: #17069 was synchronize by gbrail.

see: more, trace.

[gbrail]

@ikepolinsky

[htuch]

Thanks, this is super cool.
/wait

[htuch]

@ikepolinsky would appreciate your take on this PR; since you will be writing the fuzzing against this, it's probably a good idea to make sure it's testable and matches your understanding.

[gbrail]

Thanks for your comments -- I have some improvements I'm working on but I have one test failure to address before they're ready.

[htuch]

Sorry about the delay, holiday week.
/wait

[htuch]

Thanks, this reads well to me now, the only two things before shipping are nits and maybe getting some better docs on the thread around HTTP filter chain resumption while chunks are out for processing.

[htuch]

LGTM, thanks!

[gbrail]

The original discussion about how to do this was covered in #16760 and with this PR, that issue can be closed.

[machaval]

@gbrail do you have a working example of Streamed mode working I'm not able to make it work even on the most basic example... only on buffered mode

[gbrail]

Sure -- take a look at: GoogleCloudPlatform/envoy-processor-examples (github.com) <https://github.com/GoogleCloudPlatform/envoy-processor-examples> In particular, there are two examples in the Go code of using streaming. This one base64-encodes the HTTP response: https://github.com/GoogleCloudPlatform/envoy-processor-examples/blob/f53b0d8698b76a0a586b60a42d3098d6b845dd14/go/cmd/processor/service.go#L611C9-L611C9 And this one calculates the hash of both the request and response by reading the streaming body without modifying it: https://github.com/GoogleCloudPlatform/envoy-processor-examples/blob/f53b0d8698b76a0a586b60a42d3098d6b845dd14/go/cmd/processor/service.go#L398

…

On Mon, Dec 11, 2023 at 1:55 PM Mariano de Achaval ***@***.***> wrote: @gbrail <https://github.com/gbrail> do you have a working example of Streamed mode working I'm not able to make it work even on the most basic example... only on buffered mode — Reply to this email directly, view it on GitHub <#17069 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAD7I27B5WQ5LL42ZPKNY2DYI56NJAVCNFSM47CF7742U5DIOJSWCZC7NNSXTN2JONZXKZKDN5WW2ZLOOQ5TCOBVGA4TKMRUGYZQ> . You are receiving this because you were mentioned.Message ID: ***@***.***>