fix: Skip unnecessary query node health check in proxy (#36491)

issue: #36490
After the query node changes from a delegator to a worker, proxy should
skip this querynode's health check.

Signed-off-by: Wei Liu <[email protected]>

internal/proxy/look_aside_balancer.go

@@ -247,6 +247,7 @@ func (b *LookAsideBalancer) checkQueryNodeHealthLoop(ctx context.Context) {
 						qn, err := b.clientMgr.GetClient(ctx, node)
 						if err != nil {
 							// get client from clientMgr failed, which means this qn isn't a shard leader anymore, skip it's health check
+							b.trySetQueryNodeUnReachable(node, err)
 							log.RatedInfo(10, "get client failed", zap.Int64("node", node), zap.Error(err))
 							return struct{}{}, nil
 						}

internal/proxy/look_aside_balancer_test.go

@@ -25,7 +25,6 @@ import (
 	"github.com/cockroachdb/errors"
 	"github.com/stretchr/testify/mock"
 	"github.com/stretchr/testify/suite"
-	"go.uber.org/atomic"
 
 	"github.com/milvus-io/milvus-proto/go-api/v2/commonpb"
 	"github.com/milvus-io/milvus-proto/go-api/v2/milvuspb"
@@ -344,12 +343,10 @@ func (suite *LookAsideBalancerSuite) TestGetClientFailed() {
 	// test get shard client from client mgr return nil
 	suite.clientMgr.ExpectedCalls = nil
 	suite.clientMgr.EXPECT().GetClient(mock.Anything, int64(2)).Return(nil, errors.New("shard client not found"))
-	failCounter := atomic.NewInt64(0)
-	suite.balancer.failedHeartBeatCounter.Insert(2, failCounter)
-
-	// slepp 10s, wait for checkNodeHealth execute for more than one round
-	time.Sleep(10 * time.Second)
-	suite.True(failCounter.Load() == 0)
+	// expected stopping the health check after failure times reaching the limit
+	suite.Eventually(func() bool {
+		return !suite.balancer.metricsMap.Contain(2)
+	}, 30*time.Second, 1*time.Second)
 }
 
 func (suite *LookAsideBalancerSuite) TestNodeRecover() {