Merge branch 'develop' into lifecycle_managerment

PaddlePaddle · Mar 6, 2018 · d8cf99e · d8cf99e
2 parents 189e26f + 99a82a9
commit d8cf99e
Show file tree

Hide file tree

Showing 3 changed files with 10 additions and 7 deletions.
diff --git a/docker/paddle_k8s b/docker/paddle_k8s
@@ -65,20 +65,23 @@ start_fluid_process() {
   task_index=""
 
   stdbuf -oL python /root/k8s_tools.py wait_pods_running ${pserver_label} ${PSERVERS}
-  stdbuf -oL python /root/k8s_tools.py wait_pods_running ${trainer_label} ${TRAINERS}
 
-  ps_hosts=$(python /root/k8s_tools.py fetch_endpoints  ${pserver_label} ${PADDLE_INIT_PORT})
-  trainer_hosts=$(python /root/k8s_tools.py fetch_endpoints  ${trainer_label} ${PADDLE_INIT_PORT})
+  if [ "${TRAINING_ROLE}" == "TRAINER" ]; then
+    stdbuf -oL python /root/k8s_tools.py wait_pods_running ${trainer_label} ${TRAINERS}
+  fi
+
+  export PADDLE_INIT_PSERVERS=$(python /root/k8s_tools.py fetch_ips ${pserver_label} ${PADDLE_INIT_PORT})
 
   if [ "${TRAINING_ROLE}" == "TRAINER" ]; then
     check_failed_cnt ${TRAINERS}
     task_index=$(python /root/k8s_tools.py fetch_id ${trainer_label})
   else
     task_index=$(python /root/k8s_tools.py fetch_id ${pserver_label})
   fi
+
+  export PADDLE_INIT_TRAINER_ID=${task_index}
 
-  stdbuf -oL sh -c "${ENTRY} --ps_hosts=${ps_hosts} --trainer_hosts=${trainer_hosts} \
-      --task_index=${task_index}"
+  stdbuf -oL sh -c "${ENTRY}"
   check_trainer_ret $?
 }
 

diff --git a/go/apis/paddlepaddle/v1/types.go b/go/apis/paddlepaddle/v1/types.go
@@ -84,7 +84,7 @@ type TrainerSpec struct {
 	MinInstance  int                        `json:"min-instance"`
 	MaxInstance  int                        `json:"max-instance"`
 	Resources    apiv1.ResourceRequirements `json:"resources"`
-	ReplicaSpec  *batchv1.JobSpec           `json:"replicaSpec"`
+	ReplicaSpec  *batchv1.Job               `json:"replicaSpec"`
 }
 
 // TrainingJobPhase is the phase of TrainingJob

diff --git a/go/apis/paddlepaddle/v1/zz_generated.deepcopy.go b/go/apis/paddlepaddle/v1/zz_generated.deepcopy.go
@@ -149,7 +149,7 @@ func (in *TrainerSpec) DeepCopyInto(out *TrainerSpec) {
 		if *in == nil {
 			*out = nil
 		} else {
-			*out = new(batch_v1.JobSpec)
+			*out = new(batch_v1.Job)
 			(*in).DeepCopyInto(*out)
 		}
 	}