mirror of
https://github.com/tencentmusic/cube-studio.git
synced 2024-12-15 06:09:57 +08:00
482 lines
20 KiB
Python
482 lines
20 KiB
Python
import ray
|
|
import re
|
|
|
|
import os
|
|
import sys
|
|
import time
|
|
|
|
from job.pkgs.k8s.py_k8s import K8s
|
|
k8s_client = K8s()
|
|
|
|
import argparse
|
|
import datetime, time
|
|
import pysnooper
|
|
|
|
# print(os.environ)
|
|
base_dir = os.path.split(os.path.realpath(__file__))[0]
|
|
KFJ_NAMESPACE = os.getenv('KFJ_NAMESPACE', '')
|
|
KFJ_TASK_ID = os.getenv('KFJ_TASK_ID', '')
|
|
KFJ_TASK_NAME = os.getenv('KFJ_TASK_NAME', '')
|
|
task_node_selectors = re.split(',|;|\n|\t', os.getenv('KFJ_TASK_NODE_SELECTOR', 'cpu=true,train=true'))
|
|
KFJ_TASK_NODE_SELECTOR = {}
|
|
for task_node_selector in task_node_selectors:
|
|
KFJ_TASK_NODE_SELECTOR[task_node_selector.split('=')[0]] = task_node_selector.split('=')[1]
|
|
|
|
KFJ_PIPELINE_ID = os.getenv('KFJ_PIPELINE_ID', '')
|
|
KFJ_RUN_ID = os.getenv('KFJ_RUN_ID', '')
|
|
KFJ_CREATOR = os.getenv('KFJ_CREATOR', '')
|
|
KFJ_RUNNER = os.getenv('KFJ_RUNNER','')
|
|
KFJ_PIPELINE_NAME = os.getenv('KFJ_PIPELINE_NAME', '')
|
|
KFJ_TASK_IMAGES = os.getenv('KFJ_TASK_IMAGES', '')
|
|
KFJ_TASK_VOLUME_MOUNT = os.getenv('KFJ_TASK_VOLUME_MOUNT', '')
|
|
KFJ_TASK_RESOURCE_CPU = os.getenv('KFJ_TASK_RESOURCE_CPU', '')
|
|
KFJ_TASK_RESOURCE_MEMORY = os.getenv('KFJ_TASK_RESOURCE_MEMORY', '')
|
|
NUM_WORKER = 3
|
|
|
|
os.environ['RAY_HOST'] = 'ray-header-' + KFJ_PIPELINE_NAME + '-' + KFJ_TASK_ID
|
|
HEADER_NAME = os.getenv('RAY_HOST', '')
|
|
WORKER_NAME = HEADER_NAME.replace('header', 'worker')
|
|
INIT_FILE=''
|
|
|
|
|
|
k8s_volumes, k8s_volume_mounts = k8s_client.get_volume_mounts(KFJ_TASK_VOLUME_MOUNT,KFJ_CREATOR)
|
|
|
|
|
|
print(k8s_volumes)
|
|
print(k8s_volume_mounts)
|
|
|
|
GPU_TYPE= os.getenv('KFJ_GPU_TYPE', 'NVIDIA')
|
|
GPU_RESOURCE= os.getenv('KFJ_TASK_RESOURCE_GPU', '0')
|
|
print(GPU_TYPE,GPU_RESOURCE)
|
|
|
|
|
|
def create_header_service(name):
|
|
service_json = {
|
|
"apiVersion": "v1",
|
|
"kind": "Service",
|
|
"metadata": {
|
|
"namespace": KFJ_NAMESPACE,
|
|
"name": name,
|
|
"labels":{
|
|
"run-id":os.getenv('KFJ_RUN_ID','unknown'),
|
|
"run-rtx":os.getenv('KFJ_RUNNER','unknown'),
|
|
"pipeline-rtx": os.getenv('KFJ_CREATOR', 'unknown'),
|
|
"task-id":os.getenv('KFJ_TASK_ID','unknown'),
|
|
"pipeline-id": os.getenv('KFJ_PIPELINE_ID', 'unknown')
|
|
}
|
|
},
|
|
"spec": {
|
|
"ports": [
|
|
{
|
|
"name": "client",
|
|
"protocol": "TCP",
|
|
"port": 10001,
|
|
"targetPort": 10001
|
|
},
|
|
{
|
|
"name": "dashboard",
|
|
"protocol": "TCP",
|
|
"port": 8265,
|
|
"targetPort": 8265
|
|
},
|
|
{
|
|
"name": "redis",
|
|
"protocol": "TCP",
|
|
"port": 6379,
|
|
"targetPort": 6379
|
|
}
|
|
],
|
|
"selector": {
|
|
"component": name
|
|
}
|
|
}
|
|
}
|
|
return service_json
|
|
|
|
# @pysnooper.snoop()
|
|
def create_header_deploy(name):
|
|
header_deploy = {
|
|
"apiVersion": "apps/v1",
|
|
"kind": "Deployment",
|
|
"metadata": {
|
|
"namespace": KFJ_NAMESPACE,
|
|
"name": name,
|
|
"labels":{
|
|
"run-id":os.getenv('KFJ_RUN_ID','unknown'),
|
|
"run-rtx":os.getenv('KFJ_RUNNER','unknown'),
|
|
"pipeline-rtx": os.getenv('KFJ_CREATOR', 'unknown'),
|
|
"task-id":os.getenv('KFJ_TASK_ID','unknown'),
|
|
"pipeline-id": os.getenv('KFJ_PIPELINE_ID', 'unknown')
|
|
}
|
|
},
|
|
"spec": {
|
|
"replicas": 1,
|
|
"selector": {
|
|
"matchLabels": {
|
|
"component": name,
|
|
"type": "ray"
|
|
}
|
|
},
|
|
"template": {
|
|
"metadata": {
|
|
"labels": {
|
|
"pipeline-id": KFJ_PIPELINE_ID,
|
|
"pipeline-name": KFJ_PIPELINE_NAME,
|
|
"task-name": KFJ_TASK_NAME,
|
|
'rtx-user': KFJ_RUNNER,
|
|
"component": name,
|
|
"type": "ray",
|
|
"run-id": os.getenv('KFJ_RUN_ID', 'unknown'),
|
|
}
|
|
},
|
|
"spec": {
|
|
"restartPolicy": "Always",
|
|
"volumes": k8s_volumes,
|
|
"imagePullSecrets": [
|
|
{
|
|
"name": "hubsecret"
|
|
}
|
|
],
|
|
"affinity": {
|
|
"nodeAffinity": {
|
|
"requiredDuringSchedulingIgnoredDuringExecution": {
|
|
"nodeSelectorTerms": [
|
|
{
|
|
"matchExpressions": [
|
|
{
|
|
"key": node_selector_key,
|
|
"operator": "In",
|
|
"values": [
|
|
KFJ_TASK_NODE_SELECTOR[node_selector_key]
|
|
]
|
|
} for node_selector_key in KFJ_TASK_NODE_SELECTOR
|
|
]
|
|
}
|
|
]
|
|
}
|
|
},
|
|
"podAntiAffinity": {
|
|
"preferredDuringSchedulingIgnoredDuringExecution": [
|
|
{
|
|
"weight": 5,
|
|
"podAffinityTerm": {
|
|
"topologyKey": "kubernetes.io/hostname",
|
|
"labelSelector": {
|
|
"matchLabels": {
|
|
"component": name,
|
|
"type":"ray"
|
|
}
|
|
}
|
|
}
|
|
}
|
|
]
|
|
}
|
|
},
|
|
"containers": [
|
|
{
|
|
"name": "ray-head",
|
|
"image": KFJ_TASK_IMAGES,
|
|
"imagePullPolicy": "Always",
|
|
"command": [
|
|
"/bin/bash",
|
|
"-c",
|
|
"%s ray start --head --port=6379 --redis-shard-ports=6380,6381 --num-cpus=$MY_CPU_REQUEST --object-manager-port=12345 --node-manager-port=12346 --block"%INIT_FILE
|
|
],
|
|
"ports": [
|
|
{
|
|
"containerPort": 6379
|
|
},
|
|
{
|
|
"containerPort": 10001
|
|
},
|
|
{
|
|
"containerPort": 8265
|
|
}
|
|
],
|
|
"volumeMounts": k8s_volume_mounts,
|
|
"env": [
|
|
{
|
|
"name": "MY_CPU_REQUEST",
|
|
"valueFrom": {
|
|
"resourceFieldRef": {
|
|
"resource": "requests.cpu"
|
|
}
|
|
}
|
|
}
|
|
],
|
|
"resources": {
|
|
"requests": {
|
|
"cpu": KFJ_TASK_RESOURCE_CPU,
|
|
"memory": KFJ_TASK_RESOURCE_MEMORY,
|
|
},
|
|
"limits": {
|
|
"cpu": KFJ_TASK_RESOURCE_CPU,
|
|
"memory": KFJ_TASK_RESOURCE_MEMORY
|
|
}
|
|
}
|
|
}
|
|
]
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if GPU_TYPE=='NVIDIA' and GPU_RESOURCE:
|
|
header_deploy['spec']['template']['spec']['containers'][0]['resources']['requests']['nvidia.com/gpu'] = GPU_RESOURCE.split(',')[0]
|
|
header_deploy['spec']['template']['spec']['containers'][0]['resources']['limits']['nvidia.com/gpu'] = GPU_RESOURCE.split(',')[0]
|
|
|
|
if GPU_TYPE=='TENCENT' and GPU_RESOURCE:
|
|
if len(GPU_RESOURCE.split(','))==2:
|
|
gpu_core,gpu_mem = GPU_RESOURCE.split(',')[0],str(4*int(GPU_RESOURCE.split(',')[1]))
|
|
if gpu_core and gpu_mem:
|
|
header_deploy['spec']['template']['spec']['containers'][0]['resources']['requests'][
|
|
'tencent.com/vcuda-core'] = gpu_core
|
|
header_deploy['spec']['template']['spec']['containers'][0]['resources']['requests'][
|
|
'tencent.com/vcuda-memory'] = gpu_mem
|
|
header_deploy['spec']['template']['spec']['containers'][0]['resources']['limits'][
|
|
'tencent.com/vcuda-core'] = gpu_core
|
|
header_deploy['spec']['template']['spec']['containers'][0]['resources']['limits'][
|
|
'tencent.com/vcuda-memory'] = gpu_mem
|
|
|
|
return header_deploy
|
|
|
|
|
|
def create_worker_deploy(header_name,worker_name):
|
|
worker_deploy = {
|
|
"apiVersion": "apps/v1",
|
|
"kind": "Deployment",
|
|
"metadata": {
|
|
"namespace": KFJ_NAMESPACE,
|
|
"name": worker_name,
|
|
"labels": {
|
|
"run-id":os.getenv('KFJ_RUN_ID','unknown'),
|
|
"run-rtx":os.getenv('KFJ_RUNNER','unknown'),
|
|
"pipeline-rtx": os.getenv('KFJ_CREATOR', 'unknown'),
|
|
"task-id":os.getenv('KFJ_TASK_ID','unknown'),
|
|
"pipeline-id": os.getenv('KFJ_PIPELINE_ID', 'unknown')
|
|
}
|
|
},
|
|
"spec": {
|
|
"replicas": NUM_WORKER,
|
|
"selector": {
|
|
"matchLabels": {
|
|
"component": worker_name,
|
|
"type": "ray"
|
|
}
|
|
},
|
|
"template": {
|
|
"metadata": {
|
|
"labels": {
|
|
"pipeline-id": KFJ_PIPELINE_ID,
|
|
"pipeline-name": KFJ_PIPELINE_NAME,
|
|
"task-name": KFJ_TASK_NAME,
|
|
'rtx-user': KFJ_RUNNER,
|
|
"component": worker_name,
|
|
"type": "ray",
|
|
"run-id": os.getenv('KFJ_RUN_ID', 'unknown'),
|
|
|
|
}
|
|
},
|
|
|
|
"spec": {
|
|
"affinity": {
|
|
"nodeAffinity": {
|
|
"requiredDuringSchedulingIgnoredDuringExecution": {
|
|
"nodeSelectorTerms": [
|
|
{
|
|
"matchExpressions": [
|
|
{
|
|
"key": node_selector_key,
|
|
"operator": "In",
|
|
"values": [
|
|
KFJ_TASK_NODE_SELECTOR[node_selector_key]
|
|
]
|
|
} for node_selector_key in KFJ_TASK_NODE_SELECTOR
|
|
]
|
|
}
|
|
]
|
|
}
|
|
},
|
|
"podAntiAffinity": {
|
|
"preferredDuringSchedulingIgnoredDuringExecution": [
|
|
{
|
|
"weight": 5,
|
|
"podAffinityTerm": {
|
|
"topologyKey": "kubernetes.io/hostname",
|
|
"labelSelector": {
|
|
"matchLabels": {
|
|
"component": worker_name
|
|
}
|
|
}
|
|
}
|
|
}
|
|
]
|
|
}
|
|
},
|
|
"imagePullSecrets": [
|
|
{
|
|
"name": "hubsecret"
|
|
}
|
|
],
|
|
"restartPolicy": "Always",
|
|
"volumes": k8s_volumes,
|
|
"containers": [
|
|
{
|
|
"name": "ray-worker",
|
|
"image": KFJ_TASK_IMAGES,
|
|
"imagePullPolicy": "Always",
|
|
"command": [
|
|
"/bin/bash",
|
|
"-c",
|
|
"%s ray start --num-cpus=$MY_CPU_REQUEST --address=$RAY_HEAD_SERVICE_HOST:$RAY_HEAD_SERVICE_PORT_REDIS --object-manager-port=12345 --node-manager-port=12346 --block"%INIT_FILE
|
|
],
|
|
"volumeMounts": k8s_volume_mounts,
|
|
"env": [
|
|
{
|
|
"name": "MY_CPU_REQUEST",
|
|
"valueFrom": {
|
|
"resourceFieldRef": {
|
|
"resource": "requests.cpu"
|
|
}
|
|
}
|
|
},
|
|
{
|
|
"name": "RAY_HEAD_SERVICE_HOST",
|
|
"value": header_name
|
|
},
|
|
{
|
|
"name": "RAY_HEAD_SERVICE_PORT_REDIS",
|
|
"value": "6379"
|
|
}
|
|
],
|
|
"resources": {
|
|
"requests": {
|
|
"cpu": KFJ_TASK_RESOURCE_CPU,
|
|
"memory": KFJ_TASK_RESOURCE_MEMORY
|
|
},
|
|
"limits": {
|
|
"cpu": KFJ_TASK_RESOURCE_CPU,
|
|
"memory": KFJ_TASK_RESOURCE_MEMORY
|
|
}
|
|
}
|
|
}
|
|
]
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if GPU_TYPE=='NVIDIA' and GPU_RESOURCE:
|
|
worker_deploy['spec']['template']['spec']['containers'][0]['resources']['requests']['nvidia.com/gpu'] = GPU_RESOURCE.split(',')[0]
|
|
worker_deploy['spec']['template']['spec']['containers'][0]['resources']['limits']['nvidia.com/gpu'] = GPU_RESOURCE.split(',')[0]
|
|
|
|
if GPU_TYPE=='TENCENT' and GPU_RESOURCE:
|
|
if len(GPU_RESOURCE.split(','))==2:
|
|
gpu_core,gpu_mem = GPU_RESOURCE.split(',')[0],str(4*int(GPU_RESOURCE.split(',')[1]))
|
|
if gpu_core and gpu_mem:
|
|
worker_deploy['spec']['template']['spec']['containers'][0]['resources']['requests'][
|
|
'tencent.com/vcuda-core'] = gpu_core
|
|
worker_deploy['spec']['template']['spec']['containers'][0]['resources']['requests'][
|
|
'tencent.com/vcuda-memory'] = gpu_mem
|
|
worker_deploy['spec']['template']['spec']['containers'][0]['resources']['limits'][
|
|
'tencent.com/vcuda-core'] = gpu_core
|
|
worker_deploy['spec']['template']['spec']['containers'][0]['resources']['limits'][
|
|
'tencent.com/vcuda-memory'] = gpu_mem
|
|
|
|
|
|
return worker_deploy
|
|
|
|
|
|
# @pysnooper.snoop()
|
|
def wait_for_nodes():
|
|
# Wait for all nodes to join the cluster.
|
|
while True:
|
|
resources = ray.cluster_resources()
|
|
node_keys = [key for key in resources if "node" in key]
|
|
num_nodes = sum(resources[node_key] for node_key in node_keys)
|
|
if num_nodes < NUM_WORKER:
|
|
print("{} nodes have joined so far, waiting for {} more.".format(num_nodes, NUM_WORKER - num_nodes))
|
|
sys.stdout.flush()
|
|
time.sleep(1)
|
|
else:
|
|
break
|
|
|
|
|
|
# @pysnooper.snoop()
|
|
def launcher_cluster(deal=None):
|
|
# 清理一下之前存在的
|
|
try:
|
|
print('begin delete old header service')
|
|
k8s_client.v1.delete_namespaced_service(HEADER_NAME, KFJ_NAMESPACE)
|
|
except Exception as e1:
|
|
pass
|
|
print(e1)
|
|
|
|
try:
|
|
print('begin delete old header deployment')
|
|
k8s_client.AppsV1Api.delete_namespaced_deployment(HEADER_NAME, KFJ_NAMESPACE)
|
|
except Exception as e1:
|
|
pass
|
|
print(e1)
|
|
|
|
try:
|
|
print('begin delete old worker deployment')
|
|
k8s_client.AppsV1Api.delete_namespaced_deployment(WORKER_NAME, KFJ_NAMESPACE)
|
|
except Exception as e1:
|
|
pass
|
|
print(e1)
|
|
time.sleep(3)
|
|
|
|
if deal=='create':
|
|
header_service = create_header_service(HEADER_NAME)
|
|
header_deploy = create_header_deploy(HEADER_NAME)
|
|
worker_deploy = create_worker_deploy(HEADER_NAME,WORKER_NAME)
|
|
try:
|
|
print(KFJ_NAMESPACE)
|
|
print(header_service)
|
|
print('begin create ray header service,%s ' % datetime.datetime.now())
|
|
k8s_client.v1.create_namespaced_service(KFJ_NAMESPACE, header_service, pretty='true')
|
|
print('begin create ray header deployment,%s ' % datetime.datetime.now())
|
|
print(header_deploy)
|
|
k8s_client.AppsV1Api.create_namespaced_deployment(KFJ_NAMESPACE, header_deploy, pretty='true')
|
|
print('begin create ray worker deployment,%s ' % datetime.datetime.now())
|
|
print(worker_deploy)
|
|
k8s_client.AppsV1Api.create_namespaced_deployment(KFJ_NAMESPACE, worker_deploy, pretty='true')
|
|
# 等待创建完成
|
|
time.sleep(20)
|
|
header_host = "%s:10001" % HEADER_NAME
|
|
print('begin connect ray cluster %s,%s ' % (header_host,datetime.datetime.now()))
|
|
|
|
ray.util.connect(header_host,connection_retries=20)
|
|
wait_for_nodes()
|
|
print('ray cluster all node ready,%s ' % datetime.datetime.now())
|
|
|
|
except Exception as e:
|
|
print(e)
|
|
try:
|
|
print('begin delete error header service')
|
|
k8s_client.v1.delete_namespaced_service(HEADER_NAME, KFJ_NAMESPACE)
|
|
except Exception as e1:
|
|
pass
|
|
# print(e1)
|
|
try:
|
|
print('begin delete error header deployment')
|
|
k8s_client.AppsV1Api.delete_namespaced_deployment(HEADER_NAME, KFJ_NAMESPACE)
|
|
except Exception as e1:
|
|
pass
|
|
# print(e1)
|
|
try:
|
|
print('begin delete error worker deployment')
|
|
k8s_client.AppsV1Api.delete_namespaced_deployment(WORKER_NAME, KFJ_NAMESPACE)
|
|
except Exception as e1:
|
|
pass
|
|
print(e1)
|
|
# 如果出现错误，报错退出。不进行下一步代码
|
|
raise e
|
|
|
|
def ray_launcher(num_workers, init_file, deal):
|
|
NUM_WORKER = int(num_workers)
|
|
INIT_FILE = "bash "+init_file.strip()+" && "
|
|
launcher_cluster(deal=deal)
|
|
return "%s:10001" % HEADER_NAME
|