Grpcio pre system tests

[crwilcox]

No description provided.

[crwilcox]

Logging and BigTable will likely fail. This is due to #5303 (Logging) and #5302 (BigTable)

[tseaver]

@crwilcox How long do we expect to need the pre-release version of grpcio? Also, it would be useful to link the issue(s) here that the new pre-release solves, to give us scope about whether / when to revert.

[tswast]

@tseaver I assumed this was proactive on our part so that we could catch errors in gRPC that cause issues in the client libs before the release. I agree it would be good to get some clarification as to why we are adding this (ideally as comments in the code).

[tseaver]

@tswast I'm actually opposed to doing our normal CI against pre-releases of dependencies: we need more stability in our everyday runs, not less. It would be fine if we had an additional (maybe nightly?) "smoke test" for such things.

[tswast]

I'd be on board with splitting risky / flakey / slow & only moderate value system tests into a nightly build.

[theacodes]

That sounds lovely, unfortunately, our current CI is working against us here. Let's defer that until we move to Kokoro.

[tseaver]

@crwilcox You merged this one without getting consensus about the tradeoffs of using prerelease dependencies. I will ask again: Is there a specifc issue addressed by a particular grpcio prerlease that you are trying to fix? If not, then I think we should reconsider making our everyday CI more fragile.

[tseaver]

Note that #5333 (addressing https://circleci.com/gh/GoogleCloudPlatform/google-cloud-python/6613) is the kind of flakiness I'm talking about: we are now in a mode where we can't run against both the released and the unreleased grpcio libraries, which is not a Good Thing(TM).