diff --git a/jobUpdate.go b/jobUpdate.go
index d880c29..f73b946 100644
--- a/jobUpdate.go
+++ b/jobUpdate.go
@@ -294,3 +294,8 @@ func (j *JobUpdate) PartitionPolicy(reschedule bool, delay int64) *JobUpdate {
 	})
 	return j
 }
+
+func (j *JobUpdate) SlaPolicy(sla *aurora.SlaPolicy) *JobUpdate {
+	j.task.task.SlaPolicy = sla
+	return j
+}
diff --git a/realis.go b/realis.go
index a7c536e..5d9b4fa 100644
--- a/realis.go
+++ b/realis.go
@@ -522,6 +522,52 @@ func (c *Client) RestartInstances(key aurora.JobKey, instances ...int32) error {
 	return nil
 }
 
+// Restarts specific instances specified with slaAware.
+// If slaPolicy is nil, uses the existing slaPolicy of taskConfig.
+func (c *Client) SlaRestartInstances(jobKey aurora.JobKey,
+	slaPolicy *aurora.SlaPolicy,
+	instances ...int32) (*aurora.StartJobUpdateResult_, error) {
+	c.logger.DebugPrintf("SlaRestartInstances Thrift Payload: %v %+v %v\n", slaPolicy, jobKey, instances)
+
+	if len(instances) == 0 {
+		c.logger.DebugPrintf("it is not necessary to restart 0 instances")
+		return nil, nil
+	}
+
+	jobSummary, err := c.GetJobSummary(jobKey.Role)
+	if err != nil {
+		return nil, err
+	}
+	var jobConfig *aurora.JobConfiguration
+	for _, s := range jobSummary.Summaries {
+		if s.Job.Key.Environment == jobKey.Environment && s.Job.Key.Name == jobKey.Name {
+			jobConfig = s.Job
+		}
+	}
+	if jobConfig == nil {
+		return nil, fmt.Errorf("failed to find %v", jobKey)
+	}
+
+	// create job update request
+	jobUpdate := JobUpdateFromConfig(jobConfig.TaskConfig)
+	jobUpdate.
+		SlaAware(true).
+		InstanceCount(jobConfig.InstanceCount)
+	if slaPolicy != nil {
+		jobUpdate.SlaPolicy(slaPolicy)
+	}
+	for _, v := range instances {
+		jobUpdate.AddInstanceRange(v, v)
+	}
+
+	msg := fmt.Sprintf("SlaRestartInstances %v-%v via StartJobUpdate", jobKey, instances)
+	if result, err := c.StartJobUpdate(jobUpdate, fmt.Sprintf("restart instances %v", instances)); err != nil {
+		return nil, errors.Wrap(err, msg)
+	} else {
+		return result, nil
+	}
+}
+
 // Restarts all active tasks under a job configuration.
 func (c *Client) RestartJob(key aurora.JobKey) error {
 
diff --git a/realis_e2e_test.go b/realis_e2e_test.go
index 55fab19..dc1e946 100644
--- a/realis_e2e_test.go
+++ b/realis_e2e_test.go
@@ -925,3 +925,59 @@ func TestRealisClient_GetJobSummary(t *testing.T) {
 	err = r.KillJob(job.JobKey())
 	assert.NoError(t, err)
 }
+
+func TestRealisClient_SlaRestartInstances(t *testing.T) {
+	// Create a single job
+	role := "vagrant"
+	env := "prod"
+	name := "slaRestartInstances"
+
+	job := realis.NewJob().
+		Environment(env).
+		Role(role).
+		Name(name).
+		ThermosExecutor(thermosExec).
+		CPU(.01).
+		RAM(4).
+		Disk(10).
+		Tier("preferred").
+		InstanceCount(3).
+		IsService(true)
+
+	// Needed to populate the task config correctly
+	assert.NoError(t, job.BuildThermosPayload())
+
+	var cpu = 3.5
+	var ram int64 = 20480
+	var disk int64 = 10240
+	err := r.SetQuota(role, &cpu, &ram, &disk)
+	assert.NoError(t, err)
+
+	err = r.CreateJob(job)
+	assert.NoError(t, err)
+
+	// waiting until all instances running
+	success, err := r.MonitorScheduleStatus(job.JobKey(),
+		job.GetInstanceCount(),
+		[]aurora.ScheduleStatus{aurora.ScheduleStatus_RUNNING},
+		1*time.Second,
+		150*time.Second)
+	assert.True(t, success)
+	assert.NoError(t, err)
+
+	slaPolicy := &aurora.SlaPolicy{
+		PercentageSlaPolicy: &aurora.PercentageSlaPolicy{
+			Percentage:   50,
+			DurationSecs: 0,
+		},
+	}
+
+	t.Run("TestRealisClient_SlaRestartInstances", func(t *testing.T) {
+		result, err := r.SlaRestartInstances(job.JobKey(), slaPolicy, 0)
+		assert.NoError(t, err)
+		assert.NotNil(t, result)
+
+		assert.NoError(t, r.AbortJobUpdate(*result.GetKey(), "abort update to kill the job"))
+		assert.NoError(t, r.KillJob(job.JobKey()))
+	})
+}