/retest

These oss-bot failures don't look like they have anything to do with my changes.

/retest

This was implemented with okhttp 3, because the newer version: 1) lets us specify a single timeout for the whole call, instead of separate timeouts for connect, read, and write, and 2) lets us easily set timeouts for particular calls, rather than having to create a new client for each callable.

https://github.com/square/okhttp/wiki/Recipes#timeouts

The downside is this may increase library size for developers using both Firestore and Functions, since they will no longer share the same okhttp version.

Is 60s the best number on the client? If that's measured by the moment the client attempts to connect, to the moment it receives the response, the function on the server may not actually have its own full 60s to yield a response, since the client could also be waiting for network latency and such.

60s is what had been agreed upon for the API, and it's a big improvement over the previous, unspecified default. But I agree it would be reasonable to have a longer default to account for connection time. We should make the default the same across all client SDKs. What length default timeout would you propose?

I'd guess that a buffer of 5 extra seconds is enough to cover 98% of the connections out there.

After some discussion offline, I've upped the default timeout to 70s. Thanks for the feedback @CodingDoug.

/retest

@bklimt: The following test failed, say /retest to rerun them all: