Report gRPC connnection errors to the caller

By default `grpc.DialContext()` is non-blocking so any connection issue
will not be surfaced to the user. This change makes it blocking and
configures the gRPC dialer to report the underlying error if any
happens.

Signed-off-by: Simon Pasquier <spasquie@redhat.com>

CHANGELOG.md

@@ -37,6 +37,7 @@ We use *breaking :warning:* to mark changes that are not backward compatible (re
 - [#6342](https://github.com/thanos-io/thanos/pull/6342) Cache/Redis: Upgrade `rueidis` to v1.0.2 to to improve error handling while shrinking a redis cluster.
 - [#6325](https://github.com/thanos-io/thanos/pull/6325) Store: return gRPC resource exhausted error for byte limiter.
 - [#6399](https://github.com/thanos-io/thanos/pull/6399) *: Fix double-counting bug in http_request_duration metric
+- [#6428](https://github.com/thanos-io/thanos/pull/6428) Report gRPC connnection errors in the logs.
 
 ### Changed
 - [#6168](https://github.com/thanos-io/thanos/pull/6168) Receiver: Make ketama hashring fail early when configured with number of nodes lower than the replication factor.

pkg/query/endpointset.go

@@ -636,6 +636,11 @@ type endpointRef struct {
 // The call to newEndpointRef will return an error if establishing the channel fails.
 func (e *EndpointSet) newEndpointRef(ctx context.Context, spec *GRPCEndpointSpec) (*endpointRef, error) {
 	dialOpts := append(e.dialOpts, spec.dialOpts...)
+	// By default DialContext is non-blocking which means that any connection
+	// failure won't be reported/logged. Instead block until the connection is
+	// successfully established and return the details of the connection error
+	// if any.
+	dialOpts = append(dialOpts, grpc.WithReturnConnectionError())
 	conn, err := grpc.DialContext(ctx, spec.Addr(), dialOpts...)
 	if err != nil {
 		return nil, errors.Wrap(err, "dialing connection")