balancer/pickfirstleaf: Fix handling of Addresses with Metadata

[arjan-bal]

Addresses: #8225

This PR introduces a new data structure that wraps AddressMap but considers Address.Metadata while comparing map keys. This is required since pickfirst must consider addresses with different Metadata as separate.

RELEASE NOTES:

• balancer/pickfirstleaf: Fix handling of the deprecated Address.Metadata field.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 82.03%. Comparing base (ce35fd4) to head (c8fb55c).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #8227      +/-   ##
==========================================
- Coverage   82.16%   82.03%   -0.13%     
==========================================
  Files         410      411       +1     
  Lines       40248    40284      +36     
==========================================
- Hits        33068    33047      -21     
- Misses       5830     5875      +45     
- Partials     1350     1362      +12     

Files with missing lines	Coverage Δ
balancer/pickfirst/pickfirstleaf/pickfirstleaf.go	86.57% <100.00%> (-0.03%)	⬇️
balancer/pickfirst/pickfirstleaf/subconn_map.go	100.00% <100.00%> (ø)

... and 32 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[easwars]

Do we still want to do this?

The most recent comment on etcd-io/etcd#19700 (comment) says that etcd itself is not affected by this and that they are going to prioritize the long term fix of not using the Metadata field over any short term fixes.

But I also see that your note saying pickfirstleaf is affected and that it does not work with Addresses that contain the Metadata field. Can we say that the new pick_first policy will not support the Metadata field since it has now been marked as deprecated for over 5 years and thereby simplify our lives?

[arjan-bal]

Waiting to discuss this with @dfawley. If we think is fine not to support Metadata to differentiate addresses in the new pickfirst, we would need to change the following condition. Having a test for the behaviour would be good.

grpc-go/balancer/pickfirst/pickfirstleaf/pickfirstleaf.go

Lines 924 to 928 in 732f3f3

	func equalAddressIgnoringBalAttributes(a, b *resolver.Address) bool {
	return a.Addr == b.Addr && a.ServerName == b.ServerName &&
	a.Attributes.Equal(b.Attributes) &&
	a.Metadata == b.Metadata
	}