chore: update grpc runtime library

[leoluz]

Signed-off-by: Leonardo Luz Almeida leonardo_almeida@intuit.com

TL;DR

This PR updates the gRPC runtime library (to the current latest version v1.45.0) used in ArgoCD.

Analysis

In ArgoCD we have a mix of proto3 and proto2 used for gRPC messages:

• proto2 is used by the Application gRPC APIs and uses gogo-protobuf annotations.
• proto3 is used for all other operations provided by the API server.

With this scenario, updating the grpc runtime library works for most of UI/CLI operations but fail in Application ones with the error:

09:13:53  api-server | panic: protobuf tag not enough fields in Empty.state: 
09:13:53  api-server | goroutine 403 [running]:
09:13:53  api-server | github.com/gogo/protobuf/proto.(*unmarshalInfo).computeUnmarshalInfo(0xc000fa2500)
09:13:53  api-server | 	/Users/lalmeida1/dev/go/src/github.com/argoproj/argo-cd/vendor/github.com/gogo/protobuf/proto/table_unmarshal.go:341 +0x138a

Error explanation details can be found here. As this is coming from the gogo marshaling code, one other possibility would be to remove gogo plugin from the project and use the standard provided by google.protobuf. Unfortunately this is still not possible because of Kubernetes proto files incompatibility issues described here.

Solution

With this scenario, the only possibility to update gRPC runtime library in ArgoCD is by removing all gogo annotations from the application.proto so we don't need to rely on generated Marshallers code and avoid the incompatibility issue. However we still need to use gogo plugin for go stubs generation to maintain the compatibility with Kubernetes protos.

This PR is an attempt to implement the solution above.

Relates to #4972

Checklist:

• Either (a) I've created an enhancement proposal and discussed it with the community, (b) this is a bug fix, or (c) this does not need to be in the release notes.
• The title of the PR states what changed and the related issues number (used for the release note).
• I've included "Closes [ISSUE #]" or "Fixes [ISSUE #]" in the description to automatically close the associated issue.
• I've updated both the CLI and UI to expose my feature, or I plan to submit a second PR with them.
• Does this PR require documentation updates?
• I've updated documentation as required by this PR.
• Optional. My organization is added to USERS.md.
• I have signed off all my commits as required by DCO
• I have written unit and/or e2e tests for my change. PRs without these are unlikely to be merged.
• My build is green (troubleshooting builds).

[codecov]

Codecov Report

Merging #8929 (df0e4b5) into master (0e92705) will increase coverage by 1.48%.
The diff coverage is 13.90%.

@@            Coverage Diff             @@
##           master    #8929      +/-   ##
==========================================
+ Coverage   43.40%   44.88%   +1.48%     
==========================================
  Files         186      212      +26     
  Lines       23373    25310    +1937     
==========================================
+ Hits        10145    11361    +1216     
- Misses      11779    12341     +562     
- Partials     1449     1608     +159     

Impacted Files	Coverage Δ
cmd/argocd/commands/app.go	9.15% <0.00%> (-0.03%)	⬇️
cmd/argocd/commands/app_actions.go	0.00% <0.00%> (ø)
util/grpc/grpc.go	0.00% <0.00%> (ø)
server/application/application.go	30.13% <21.50%> (-1.24%)	⬇️
server/server.go	55.57% <100.00%> (+0.16%)	⬆️
pkg/apis/application/v1alpha1/hack.go	0.00% <0.00%> (ø)
...licationset/generators/generator_spec_processor.go	57.57% <0.00%> (ø)
applicationset/utils/utils.go	77.10% <0.00%> (ø)
... and 27 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0e92705...df0e4b5. Read the comment docs.

[alexmt]

LGTM.

Code changes look good. Tested CLI locally as well and could not find any bugs. Lets merge it !

[victorboissiere]

FYI, this change deleted all of our apps since we used project=XX. The filter on project was not there and resulted in all our apps being listed and deleted. This was not in the changelog.

[crenshaw-dev]

@victorboissiere would you mind opening an issue requesting that the upgrade guide be changed to include a note about this?

We probably also need to establish a stricter policy about API backwards-compatibility.

[jannfis]

We probably also need to establish a stricter policy about API backwards-compatibility.

I think we already promise backwards compatibility for the API. And I think this one just slipped through the cracks.

[leoluz]

This file is auto-generated. The breaking change originated here. This was probably a misunderstanding about how the gogoproto.customname worked. Unfortunately this wasn't caught by our automated tests on that occasion. I agree with @jannfis that a stricter policy isn't the answer to this incident. Our API is exposed at /api/v1 and that shouldn't break. What we could do to avoid this type of problem in the future is improving our automated tests. Probably having some sort of contract testing would be beneficial.

[leoluz]

@victorboissiere I am sorry about the incident that this change caused. To avoid this type of issue to happen again in the future, I created the following proposal: #12589

[crenshaw-dev]

@victorboissiere what was your client? I'm investigating whether this is a problem with the CLI.

[victorboissiere]

Thanks for the issue! it was not the CLI, we’ve built an internal tool that cleanup from time to time some project applications. The filter stopped working and we’ve lost ~8000 pods.
Thanks for the effort to avoid this in the future 🙏

[crenshaw-dev]

Sweet, thanks! PR to fix the issue: #12594

I had written up some warnings about the CLI, but after investigation, I don't think the CLI is impacted. Only the JSON API.