peer and metadata: Implement the Stringer interface for Peer and Metadata

[AnomalRoil]

Effectively allowing to print contextes containing a Peer properly.
Instead of printing with %v as:

context.Background.WithValue(type peer.peerKey, val <not Stringer>)

contexts containing a peer will now print as:

context.Background.WithValue(type peer.peerKey, val Peer{Addr: '1.2.3.4:1234', AuthInfo: 'testAuthInfo'})

This is also satisfying Stringer interface for Metadata, allowing to print them easily too. Since Metadata are basically a map, notice that String output is not stable. I could change it to print it in alphabetical order if you prefer.

RELEASE NOTES:

• peer/peer: implement the Stringer interface for pretty printing Peer
• metadata/metadata: implement the Stringer interface for pretty printing MD

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: AnomalRoil / name: AnomalRoil (83118ab, 985347e, 20ee25c, 172603c, d36bfd5, c908bad, edfd323, 067b07e, e482bcc)

[arvindbr8]

The change overall looks good to me. I took a pass at it. please take a look

[arvindbr8]

The change overall looks good to me. I took a pass at it. please take a look

[arvindbr8]

Thanks, I've responded to our conversations here. Please take a look

[arvindbr8]

Just FYI, go test std library already does something similar by replacing \ss with _s.

[AnomalRoil]

I was trying to convey information about whether the test case included + or didn't include - that specific field.

[easwars]

The use of + and - to convey the presence or absence of certain fields within the Peer isn't very readable. How about the following:

• withAddrAndValidAuth
• withAddrLocalAddrAndValidAuth
• withAddrAndEmptyAuth
• withAddrLocalAddrAndEmptyAuth
• nilPeer

The above names in the test output are more readable I feel.

[arvindbr8]

Per policy we need one more review before we can check this in. I will ping a co-maintainer to take a look at this.

[zasweq]

LGTM. Thank you for the contribution.

[serathius]

Making String() non-deterministic broke etcd stream selection mechanism.

[dims]

@serathius @zasweq the following patch to make the string deterministic solves our issue:

diff --git a/vendor/google.golang.org/grpc/metadata/metadata.go b/vendor/google.golang.org/grpc/metadata/metadata.go
index 6c01a9b359cc..ec1055f6ccf0 100644
--- a/vendor/google.golang.org/grpc/metadata/metadata.go
+++ b/vendor/google.golang.org/grpc/metadata/metadata.go
@@ -24,6 +24,7 @@ package metadata // import "google.golang.org/grpc/metadata"
 import (
 	"context"
 	"fmt"
+	"sort"
 	"strings"
 
 	"google.golang.org/grpc/internal"
@@ -95,11 +96,16 @@ func Pairs(kv ...string) MD {
 func (md MD) String() string {
 	var sb strings.Builder
 	fmt.Fprintf(&sb, "MD{")
-	for k, v := range md {
+	keys := make([]string, 0, len(md))
+	for k := range md {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	for _, k := range keys {
 		if sb.Len() > 3 {
 			fmt.Fprintf(&sb, ", ")
 		}
-		fmt.Fprintf(&sb, "%s=[%s]", k, strings.Join(v, ", "))
+		fmt.Fprintf(&sb, "%s=[%s]", k, strings.Join(md[k], ", "))
 	}
 	fmt.Fprintf(&sb, "}")
 	return sb.String()

[serathius]

Context why deterministic String() is important for etcd.

Etcd uses grpc metadata to change properties of the bi-directional watch stream. Example of that is WithRequireLeader() which will ask server to break the stream if leader is no longer available in the cluster. As client can make multiple watch requests within a bi-directional stream, we need to create a separate stream for clients that provide different metadata in the request context. Etcd client get the identifiers of stream by formatting the context metadata, so it's crucial to have deterministic stringer to properly reuse the streams.

Code: https://github.com/etcd-io/etcd/blob/21e5876f7f0539509c277b4c2a3ba1b1599b1721/client/v3/watch.go#L1038-L1043

[easwars]

What do you think about sorting the keys before generating the string representation and ensuring that the output is deterministic. This will also make the unit test much simpler than what it is currently.

[easwars]

How about defining a bool first and using that as the conditional here, instead of this magic check which took me sometime to understand.

[easwars]

The use of + and - to convey the presence or absence of certain fields within the Peer isn't very readable. How about the following:

• withAddrAndValidAuth
• withAddrLocalAddrAndValidAuth
• withAddrAndEmptyAuth
• withAddrLocalAddrAndEmptyAuth
• nilPeer

The above names in the test output are more readable I feel.

[easwars]

Phew.... Didn't realize that the PR was already approved and merged.