[Epic] Standardize on provider-ci for all providers

[mjeffryes]

Overview

Only bridged providers benefit from the consistency and automatic rollout of workflows generated from provider-ci today. This creates a lot of work when we have to make a change to all providers. Native provider ci also helps automate changes for some non-bridged providers but has also varied in structure to provider-ci. We'd like to drive greater consistency for across the whole fleet for things like package publish to gain the benefits of investment here across the whole fleet.

Key KPIs

• Number of implementations required to roll out a publishing change to all providers.
• Number of duplicate implementations where behaviour is the same.

Key Stakeholders

• Provider team
• Community provider authors

Key Deliverables

• New provider-ci templates for EKS, AWSx and API Gateway
• Integrate provider-ci into boilerplates
• Migrate native-provider-ci into provider-ci
• Minimise duplication & fragmentation

References 📔

• Project View
• PR/FAQ
• Design Doc
• UX Designs
• Decision Log

[t0yv0]

Why does the title say Standardize on CI-mgmt for all providers and the overview says "Bridged providers currently display unusable or misleading results in pulumi preview." ?

[mjeffryes]

Probably because I failed a copy-paste! thanks for the catch!

[danielrbradley]

Looks like this probably supersedes:

• Prepare provider-ci for non-bridged providers #936

Most of that issue is complete at this point. The only dangling issues are:

• Remove use of actions hooks #1039
• Extract shared test workflows and action #1037

[blampe]

This week we pulled most of the differences in the generic template into the bridged one #1140. In particular sharding and some workflow consolidation that makes subsequent changes easier.

• Consolidate test workflow #1214
• Allow bridged providers to opt-in to sharding #1217

We've also cleaned up a bunch of instances where providers had been using custom workflows:

• Remove unused sshPrivateKey config #1176
• Consolidate integration tests pulumi-aws#4910
• Test shim as a normal unit test pulumi-aws#4917
• Lint upstream as a normal unit test pulumi-aws#4918
• Remove unused extraTests #1230

In each case it's been relatively straightforward to translate the GH action into some additional test logic. This is not only simpler but it also moves us in a direction where things are more easily reproducible locally.

Remaining preTest steps seem to be mostly AWS login (easily handled in test_main.go) and make upstream (now handled automatically by make targets).

I think we're actually much closer to #936 now. There's still probably some need for the generic template, but there's definitely a path towards consolidating.

As a next step we need to rebase the generic template on top of all the recent work. We've onboarding the pulumiservice provider and can revisit EKS.

I'd also like to look into getting k8s onto it -- that has a number of exceptions (for example it spins up its own k8s cluster) which will need codifying, but I don't think it will be too bad.