fixes for safe conversation and avoid panics (#1213)

* fixes for safe convertation

Signed-off-by: Vasiliy Tolstov <v.tolstov@unistack.org>

* fix client publish panic

If broker connect returns error we dont check it status and use
it later to publish message, mostly this is unexpected because
broker connection failed and we cant use it.
Also proposed solution have benefit - we flag connection status
only when we have succeseful broker connection

Signed-off-by: Vasiliy Tolstov <v.tolstov@unistack.org>

* api/handler/broker: fix possible broker publish panic

Signed-off-by: Vasiliy Tolstov <v.tolstov@unistack.org>

grpc.go

@@ -6,7 +6,7 @@ import (
 	"crypto/tls"
 	"fmt"
 	"os"
-	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/micro/go-micro/v2/broker"
@@ -24,9 +24,9 @@ import (
 )
 
 type grpcClient struct {
-	once sync.Once
 	opts client.Options
 	pool *pool
+	once atomic.Value
 }
 
 func init() {
@@ -570,9 +570,12 @@ func (g *grpcClient) Publish(ctx context.Context, p client.Message, opts ...clie
 		body = b
 	}
 
-	g.once.Do(func() {
-		g.opts.Broker.Connect()
-	})
+	if !g.once.Load().(bool) {
+		if err = g.opts.Broker.Connect(); err != nil {
+			return errors.InternalServerError("go.micro.client", err.Error())
+		}
+		g.once.Store(true)
+	}
 
 	topic := p.Topic()
 
@@ -641,9 +644,9 @@ func newClient(opts ...client.Option) client.Client {
 	}
 
 	rc := &grpcClient{
-		once: sync.Once{},
 		opts: options,
 	}
+	rc.once.Store(false)
 
 	rc.pool = newPool(options.PoolSize, options.PoolTTL, rc.poolMaxIdle(), rc.poolMaxStreams())