It's worth testing this again with HTTP/2 with OkHttp 5.X

[baywet]

It's worth testing this again with HTTP/2 with OkHttp 4.10.0-RC1

https://square.github.io/okhttp/changelog/#version-4100-rc1

Fix: Attempt to read the response body even if the server canceled the request. This will cause some calls to return nice error codes like HTTP/1.1 429 Too Many Requests instead of transport errors like SocketException: Connection reset and StreamResetException: stream was reset: CANCEL.

Originally posted by @yschimke in microsoftgraph/msgraph-sdk-java#473 (comment)
AB#7113

[baywet]

additionally, AGS is rolling out HTTP/2 support. So this should be less of a problem for platform trying to connect with HTTP/2 first.

Note: the reason why this was disabled to begin with is because some platforms would try to connect using HTTP/2 first, AGS would not know how to handle the connection and the client would not fallback on 1.1 by default.

[yschimke]

This specific bug fix is in 5.0.0-alpha1, should make next release. Any testing here would help us confirm the fix.

https://github.com/square/okhttp/blob/master/CHANGELOG.md#version-500-alpha1

[baywet]

@yschimke thanks for the additional information. Two questions:

• it seems to me the only reason why they major bumped is because of the dependency reorg, which should have a minimal impact on us. Is it your reading as well?
• Do you know what are the timelines for the release of v5?

[yschimke]

Yep, we will be making sure it's binary compatible forwards (from 3.12) and backwards (to 4.9), excluding anything listed as new API. 5.0.0 is basically 4.10. But it does reorg mockwebserver dependencies while remaining compatible, and expand the Okio functionality (new file system API). We also expanded to commit to supporting GraalVM as a platform.

Not sure about timelines, basically when it's ready, but month-ish. No promises though.

[baywet]

great! thanks for the insights @yschimke !! So far the upgrade path should be straightforward for us then as we implemented our own mocks without relying on mockwebserver.
We also have "-ish" timelines here, on the same page in terms of expectations 😉