Fix retry stat measures to match those in grpc-java exactly

[mackenziestarr]

• fixes the retry stat measure to match those in grpc-java exactly as defined in CensusStatsModule
  • it wasn't obvious to me that the description field for a measure has a meaningful purpose beyond describing the measure and because they didn't match what was in grpc-java #2084 did not have the intended effect
• corrects type of GRPC_CLIENT_RETRY_DELAY_PER_CALL from MeasureLong to MeasureDouble

In the medium term grpc-java should reference the retry stat Measures defined in opencensus-java instead of creating its own, although these are referenced in a file called DeprecatedCensusConstants.java so i'm not sure what the long-term goal there is.

[mackenziestarr]

cc: @punya @dapengzhang0

[punya]

Would it be possible to add a test that verifies that the measures match?

@dapengzhang0 do you know why the measure constants are considered deprecated on the OpenCensus side? E.g., was there a desire at some point for the definitions in this repo to be considered the source of truth?

[mackenziestarr]

hey @punya added a test in 05e8256, hopefully it is what you had in mind.

[punya]

hey @punya added a test in 05e8256, hopefully it is what you had in mind.

Thanks for putting in the effort to write this. I was actually hoping that the test would check against the source of truth (grpc-java), not against hardcoded constants which simply repeat the code. If accessing the source of truth is impossible (or even very kludgey), I'd rather revert this test commit and merge without it.

[dapengzhang0]

@dapengzhang0 do you know why the measure constants are considered deprecated on the OpenCensus side? E.g., was there a desire at some point for the definitions in this repo to be considered the source of truth?

@punya, that might be for some historical reasons mentioned in the comment here https://github.com/grpc/grpc-java/pull/4494#discussion_r190654824

I don't if the current status inside google has changed. cc\ @zhangkun83 who might have more context.

[mackenziestarr]

@punya I think any efforts towards preventing a regression would better be spent opening a pull request in grpc-java to upgrade to the eventual opencensus version that contains this commit and reference the measures defined here directly - I'd be happy to contribute that once this is merged. I'm fine with you reverting the superfluous unit test, I agree it adds nothing.

[mackenziestarr]

hi @punya would it be possible to cut this into a 0.31.1 release since its a backwards compatible bug fix

[punya]

@mackenziestarr I'd be happy to, but probably no sooner than next week (unfortunately the release process isn't as automated as we'd like it to be). Would that work for you?

[mackenziestarr]

@punya sounds great! thanks for responding so quickly

[mackenziestarr]

hi @punya would it be possible to cut a release this week or the next?

[punya]

@mackenziestarr my apologies for dropping the ball on this. I'll get this out today.

[mackenziestarr]

no problem @punya, thank you!

[punya]

FYI: I ran into a few CI snags, but I'm still working on this. We haven't done a patch release in a while.

[mackenziestarr]

hi @punya is the patch release still blocked on CI?

[punya]

Sorry for the long wait, @mackenziestarr. This release is finally out (Github | Sonatype).

[mackenziestarr]

@punya no worries, thank you! i just pulled down the release artifact and tested it out, works great