Call configure in sequence (instead of in parallel)

[iwahbe]

It looks like GCP has introduced a dependency on configuring SDK and then PF into their provider. This commit is being used to check if ordering will fix the panics that the GCP upgrade is seeing in tests.

pulumi/pulumi-gcp#2733 shows that this is a necessary change.

[codecov]

Codecov Report

Attention: Patch coverage is 81.81818% with 2 lines in your changes missing coverage. Please review.

Project coverage is 69.64%. Comparing base (7b3ace6) to head (0f94385).
Report is 4 commits behind head on master.

Files with missing lines	Patch %	Lines
pkg/x/muxer/muxer.go	81.81%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2738   +/-   ##
=======================================
  Coverage   69.63%   69.64%           
=======================================
  Files         301      301           
  Lines       38725    38721    -4     
=======================================
- Hits        26967    26966    -1     
+ Misses      10240    10238    -2     
+ Partials     1518     1517    -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[VenelinMartinov]

How does this impact AWS? I remember there were some performance issues with Configure there. Very fuzzy on the details but I believe one of the Configure calls might hang sometimes

[iwahbe]

I believe that AWS solved that by not verifying twice. This will result in a Configure call no slower then TF. If Configure hung, then it would have hung before this change as well.

[iwahbe]

This will be marginally slower, but it should still be pretty fast. AWS had major issues with start-up time, but this doesn't effect that at all.

[iwahbe]

@t0yv0 Might have a better memory for AWS's performance issues.

[t0yv0]

Yeah we might pay few more milliseconds on startup for this in AWS. I guess the behavior here could be opt-in so only GCP opts in? Or do you think it's generally the right thing to do for most providers?

I suspect in the case of AWS what we might actually want is removing the PF Configure call entirely or making it a no-op in AWS. I think it might still work if both PF and SDKv2 providers in the AWS internals are wrapping the same state bag for the configuration information.

[t0yv0]

I think we need to own gating CI on pulumi-aws wrt the performance issues, so we just come back to this convo as needed when breaching thresholds. I'm taking a quick look.

[pulumi-bot]

This PR has been shipped in release v3.98.0.