mirror/cube-studio
2
0
Fork 0
mirror of https://github.com/tencentmusic/cube-studio.git synced 2024-12-21 06:19:31 +08:00
Code Issues Packages Projects Releases Wiki Activity
660db81f29
cube-studio/myapp/views/view_inferenceserving.py

999 lines
44 KiB
Python
Raw Normal View History

fix
2022-02-26 22:36:57 +08:00
from flask import render_template,redirect
from flask_appbuilder.models.sqla.interface import SQLAInterface
from flask import Blueprint, current_app, jsonify, make_response, request
# 将model添加成视图，并控制在前端的显示
from myapp.models.model_serving import InferenceService
from myapp.models.model_team import Project,Project_User
from myapp.utils import core
from flask_babel import gettext as __
from flask_babel import lazy_gettext as _
from flask_appbuilder.actions import action
from myapp import app, appbuilder,db,event_logger
import logging
from flask_babel import lazy_gettext,gettext
import re
import copy
import uuid
import requests
from myapp.exceptions import MyappException
from flask_appbuilder.security.decorators import has_access
from myapp.models.model_job import Repository
from flask_wtf.file import FileAllowed, FileField, FileRequired
from werkzeug.datastructures import FileStorage
from wtforms.ext.sqlalchemy.fields import QuerySelectField
from myapp import security_manager
import os,sys
from wtforms.validators import DataRequired, Length, NumberRange, Optional,Regexp
from wtforms import BooleanField, IntegerField, SelectField, StringField,FloatField,DateField,DateTimeField,SelectMultipleField,FormField,FieldList
from flask_appbuilder.fieldwidgets import BS3TextFieldWidget,BS3PasswordFieldWidget,DatePickerWidget,DateTimePickerWidget,Select2ManyWidget,Select2Widget
from myapp.forms import MyBS3TextAreaFieldWidget,MySelect2Widget,MyCodeArea,MyLineSeparatedListField,MyJSONField,MyBS3TextFieldWidget,MySelectMultipleField
from myapp.utils.py import py_k8s
import os, zipfile
import shutil
from myapp.views.view_team import filter_join_org_project
from flask import (
current_app,
abort,
flash,
g,
Markup,
make_response,
redirect,
render_template,
request,
send_from_directory,
Response,
url_for,
)
from .base import (
DeleteMixin,
api,
BaseMyappView,
check_ownership,
data_payload_response,
DeleteMixin,
generate_download_headers,
get_error_msg,
get_user_roles,
handle_api_exception,
json_error_response,
json_success,
MyappFilter,
MyappModelView,
)
from sqlalchemy import and_, or_, select
from .baseApi import (
MyappModelRestApi
)
from flask_appbuilder import CompactCRUDMixin, expose
import pysnooper,datetime,time,json
conf = app.config
class InferenceService_Filter(MyappFilter):
# @pysnooper.snoop()
def apply(self, query, func):
if g.user.is_admin():
return query
join_projects_id = security_manager.get_join_projects_id(db.session)
return query.filter(self.model.project_id.in_(join_projects_id))
class InferenceService_ModelView(MyappModelView):
datamodel = SQLAInterface(InferenceService)
check_redirect_list_url = '/inferenceservice_modelview/list/'
help_url = conf.get('HELP_URL', {}).get('inferenceservice','')
# 外层的add_column和edit_columns 还有show_columns 一定要全，不然在gunicorn形式下get的不一定能被翻译
# add_columns = ['service_type','project','name', 'label','images','resource_memory','resource_cpu','resource_gpu','min_replicas','max_replicas','ports','host','hpa','metrics','health']
add_columns = ['service_type', 'project', 'label', 'model_name', 'model_version', 'images', 'model_path', 'model_input', 'model_output', 'resource_memory', 'resource_cpu', 'resource_gpu', 'min_replicas', 'max_replicas', 'hpa', 'canary', 'shadow', 'host', 'command', 'working_dir', 'env', 'ports', 'metrics', 'health', 'expand']
show_columns = ['service_type','project', 'name', 'label','model_name', 'model_version', 'images', 'model_path', 'input_html', 'output_html', 'images', 'volume_mount','working_dir', 'command', 'env', 'resource_memory',
'resource_cpu', 'resource_gpu', 'min_replicas', 'max_replicas', 'ports', 'inference_host_url','hpa', 'canary', 'shadow', 'health','model_status', 'expand_html','metrics_html','deploy_history' ]
list_columns = ['project','service_type','model_name_url','model_version','inference_host_url','model_status','creator','modified','operate_html']
edit_columns = add_columns
label_title = '推理服务'
base_order = ('id','desc')
order_columns = ['id']
base_filters = [["id",InferenceService_Filter, lambda: []]] # 设置权限过滤器
custom_service = 'serving'
# service_type_choices= ['',custom_service,'tfserving','torch-server','onnxruntime','triton-server','kfserving-tf','kfserving-torch','kfserving-onnx','kfserving-sklearn','kfserving-xgboost','kfserving-lightgbm','kfserving-paddle']
add stand-alone deployment
2022-04-24 14:35:13 +08:00
service_type_choices= ['',custom_service,'tfserving','torch-server','onnxruntime','triton-server']
fix
2022-02-26 22:36:57 +08:00
# label_columns = {
# "host": _("域名：测试环境test.xx，调试环境 debug.xx"),
# }
service_type_choices = [x.replace('_','-') for x in service_type_choices]
add_form_extra_fields={
"project": QuerySelectField(
_(datamodel.obj.lab('project')),
query_factory=filter_join_org_project,
allow_blank=True,
widget=Select2Widget(),
validators=[DataRequired()]
),
"resource_memory":StringField(_(datamodel.obj.lab('resource_memory')),default='5G',description='内存的资源使用限制，示例1G，10G， 最大100G，如需更多联系管路员',widget=BS3TextFieldWidget(),validators=[DataRequired()]),
"resource_cpu":StringField(_(datamodel.obj.lab('resource_cpu')), default='5',description='cpu的资源使用限制(单位核)，示例 0.4，10，最大50核，如需更多联系管路员',widget=BS3TextFieldWidget(), validators=[DataRequired()]),
"min_replicas": StringField(_(datamodel.obj.lab('min_replicas')), default=InferenceService.min_replicas.default.arg,description='最小副本数，用来配置高可用，流量变动自动伸缩',widget=BS3TextFieldWidget(), validators=[DataRequired()]),
"max_replicas": StringField(_(datamodel.obj.lab('max_replicas')), default=InferenceService.max_replicas.default.arg,
description='最大副本数，用来配置高可用，流量变动自动伸缩', widget=BS3TextFieldWidget(),
validators=[DataRequired()]),
"host": StringField(_(datamodel.obj.lab('host')), default=InferenceService.host.default.arg,description='访问域名，xx.serving.%s'%conf.get('ISTIO_INGRESS_DOMAIN',''),widget=BS3TextFieldWidget()),
"transformer":StringField(_(datamodel.obj.lab('transformer')), default=InferenceService.transformer.default.arg,description='前后置处理逻辑，用于原生开源框架的请求预处理和响应预处理，目前仅支持kfserving下框架',widget=BS3TextFieldWidget()),
'resource_gpu':StringField(_(datamodel.obj.lab('resource_gpu')), default=0,
description='gpu的资源使用限制(单位卡)，示例:1，2，训练任务每个容器独占整卡。申请具体的卡型号，可以类似 1(V100),目前支持T4/V100/A100/VGPU',
widget=BS3TextFieldWidget()),
'sidecar': MySelectMultipleField(
_(datamodel.obj.lab('sidecar')), default='',
description='容器的agent代理',
widget=Select2ManyWidget(),
choices=[['L5', 'L5'], ['DC', 'DC']]
)
}
edit_form_extra_fields = add_form_extra_fields
# edit_form_extra_fields['name']=StringField(_(datamodel.obj.lab('name')), description='英文名(字母、数字、- 组成)，最长50个字符',widget=MyBS3TextFieldWidget(readonly=True), validators=[Regexp("^[a-z][a-z0-9\-]*[a-z0-9]$"),Length(1,54)]),
# @pysnooper.snoop()
def set_column(self, service=None):
# 对编辑进行处理
request_data = request.args.to_dict()
service_type = request_data.get('service_type', '')
if service:
service_type = service.service_type
if service:
self.add_form_extra_fields['service_type'] = SelectField(
_(self.datamodel.obj.lab('service_type')),
description="推理框架类型",
choices=[[x,x] for x in self.service_type_choices],
widget=MySelect2Widget(extra_classes="readonly",value=service_type),
validators=[DataRequired()]
)
else:
self.add_form_extra_fields['service_type'] = SelectField(
_(self.datamodel.obj.lab('service_type')),
description="推理框架类型",
widget=MySelect2Widget(new_web=True,value=service_type),
choices=[[x,x] for x in self.service_type_choices],
validators=[DataRequired()]
)
self.add_form_extra_fields['model_name'] = StringField(
_('模型名称'),
default=service.model_name if service else '',
description='英文名(字母、数字、- 组成)，最长50个字符',
widget=BS3TextFieldWidget(),
validators=[DataRequired(),Regexp("^[a-z][a-z0-9\-]*[a-z0-9]$"), Length(1, 54)]
)
self.add_form_extra_fields['model_version'] = StringField(
_('模型版本号'),
default=service.model_version if service else datetime.datetime.now().strftime('v%Y.%m.%d.1'),
description='版本号，时间格式',
widget=BS3TextFieldWidget(),
validators=[DataRequired(),Regexp("^v[0-9.]*$"), Length(1, 54)]
)
self.add_form_extra_fields['model_path'] = StringField(
_('模型地址'),
default=service.model_path if service else '',
description='英文名(字母、数字、- 组成)，最长50个字符',
widget=BS3TextFieldWidget()
# validators=[DataRequired()]
)
# 下面是公共配置，特定化值
images = conf.get('INFERNENCE_IMAGES',{}).get(service_type,[])
command = conf.get('INFERNENCE_COMMAND',{}).get(service_type,'')
env = conf.get('INFERNENCE_ENV',{}).get(service_type,[])
ports = conf.get('INFERNENCE_PORTS', {}).get(service_type, '80')
metrics = conf.get('INFERNENCE_METRICS', {}).get(service_type, '')
health = conf.get('INFERNENCE_HEALTH', {}).get(service_type, '')
if service_type==self.custom_service:
self.add_form_extra_fields['images'] = StringField(
_(self.datamodel.obj.lab('images')),
default=service.images if service else '',
description="推理服务镜像",
widget=BS3TextFieldWidget(),
validators=[DataRequired()]
)
else:
self.add_form_extra_fields['images'] = SelectField(
_(self.datamodel.obj.lab('images')),
default=service.images if service else '',
description="推理服务镜像",
widget=Select2Widget(),
choices=[[x,x] for x in images]
)
self.add_form_extra_fields['command'] = StringField(
_(self.datamodel.obj.lab('command')),
default=service.command if service else command,
description='启动命令，支持多行命令，留空时将被自动重置',
widget=MyBS3TextAreaFieldWidget(rows=3)
)
self.add_form_extra_fields['env'] = StringField(
_(self.datamodel.obj.lab('env')),
default=service.env if service else '\n'.join(env),
description='使用模板的task自动添加的环境变量，支持模板变量。书写格式:每行一个环境变量env_key=env_value',
widget=MyBS3TextAreaFieldWidget()
)
self.add_form_extra_fields['ports'] = StringField(
_(self.datamodel.obj.lab('ports')),
default=service.ports if service else ports,
description='监听端口号，逗号分隔',
widget=BS3TextFieldWidget(),
validators=[DataRequired()]
)
self.add_form_extra_fields['metrics'] = StringField(
_(self.datamodel.obj.lab('metrics')),
default=service.metrics if service else metrics,
description='请求指标采集，配置端口+url，示例：8080:/metrics',
widget=BS3TextFieldWidget()
)
self.add_form_extra_fields['health'] = StringField(
_(self.datamodel.obj.lab('health')),
default=service.health if service else health,
description='健康检查接口，使用http接口或者shell命令，示例：8080:/health或者 shell:python health.py',
widget=BS3TextFieldWidget()
)
# self.add_form_extra_fields['name'] = StringField(
# _(self.datamodel.obj.lab('name')),
# default=g.user.username+"-"+service_type+'-xx-v1',
# description='英文名(字母、数字、- 组成)，最长50个字符',
# widget=BS3TextFieldWidget(),
# validators=[DataRequired(),Regexp("^[a-z][a-z0-9\-]*[a-z0-9]$"), Length(1, 54)]
# )
self.add_form_extra_fields['label'] = StringField(
_(self.datamodel.obj.lab('label')),
default="xx模型，%s框架，xx版"%service_type,
description='中文描述',
widget=BS3TextFieldWidget(),
validators=[DataRequired()]
)
self.add_form_extra_fields["hpa"]=StringField(
_(self.datamodel.obj.lab('hpa')),
default=service.hpa if service else 'cpu:50%,gpu:50%',
description='弹性伸缩容的触发条件：可以使用cpu/mem/gpu/qps等信息，可以使用其中一个指标或者多个指标，示例：cpu:50%,mem:%50,gpu:50%',
widget=BS3TextFieldWidget()
)
self.add_form_extra_fields['expand'] = StringField(
_(self.datamodel.obj.lab('expand')),
default=service.expand if service else '{}',
description='扩展字段',
widget=MyBS3TextAreaFieldWidget(rows=12)
)
self.add_form_extra_fields['canary'] = StringField(
_('流量分流'),
default=json.loads(service.expand).get('canary','') if service else '',
description='流量分流，将该服务的所有请求，按比例分流到目标服务上。格式 service1:20%,service2:30%，表示分流20%流量到service1，30%到service2',
widget=BS3TextFieldWidget()
)
self.add_form_extra_fields['shadow'] = StringField(
_('流量镜像'),
default=json.loads(service.expand).get('shadow','') if service else '',
description='流量镜像，将该服务的所有请求，按比例复制到目标服务上，格式 service1:20%,service2:30%，表示复制20%流量到service1，30%到service2',
widget=BS3TextFieldWidget()
)
model_columns = ['service_type', 'project', 'label', 'model_name', 'model_version', 'images', 'model_path']
service_columns = ['resource_memory', 'resource_cpu','resource_gpu', 'min_replicas', 'max_replicas', 'hpa','canary','shadow','host','sidecar']
admin_columns = ['command','working_dir','env','ports','metrics','health','expand']
if service_type=='tfserving' or service_type=='kfserving-tf':
self.add_form_extra_fields['model_path'] = StringField(
_('模型地址'),
default=service.model_path if service else '/mnt/.../saved_model',
description='仅支持tf save_model的模型存储方式',
widget=BS3TextFieldWidget(),
validators=[DataRequired()]
)
if service_type=='torch-server' or service_type=='kfserving-torch':
self.add_form_extra_fields['model_path'] = StringField(
_('模型地址'),
default=service.model_path if service else '/mnt/.../$model_name.mar',
description='需保存完整模型信息，包括模型结构和模型参数，或者使用torch-model-archiver编译后的mar模型文件',
widget=BS3TextFieldWidget(),
validators=[DataRequired()]
)
self.add_form_extra_fields['model_type'] = SelectField(
_('模型类型'),
default=service.model_type if service else 'image_classifier',
description='模型的功能类型',
widget=Select2Widget(),
choices=[[x, x] for x in ["image_classifier","image_segmenter","object_detector","text_classifier"]],
validators=[DataRequired()]
)
model_columns.append('model_type')
if service_type=='onnxruntime' or service_type=='kfserving-onnx':
self.add_form_extra_fields['model_path'] = StringField(
_('模型地址'),
default=service.model_path if service else '/mnt/.../$model_name.onnx',
description='onnx模型文件的地址',
widget=BS3TextFieldWidget(),
validators=[DataRequired()]
)
if service_type=='triton-server':
self.add_form_extra_fields['model_path'] = StringField(
_('模型地址'),
default=service.model_path if service else 'onnx:/mnt/.../model.onnx(model.plan,model.bin,model.savedmodel/,model.pt,model.dali)',
description='框架:地址。onnx:模型文件地址model.onnx，pytorch:torchscript模型文件地址model.pt，tf:模型目录地址saved_model，tensorrt:模型文件地址model.plan',
widget=BS3TextFieldWidget(),
validators=[DataRequired()]
)
input_demo='''
[
{
name: "input_name"
data_type: TYPE_FP32
format: FORMAT_NCHW
dims: [ 3, 224, 224 ]
reshape: {
shape: [ 1, 3, 224, 224 ]
}
}
]
'''
output_demo ='''
[
{
name: "output_name"
data_type: TYPE_FP32
dims: [ 1000 ]
reshape: {
shape: [ 1, 1000 ]
}
}
]
'''
self.add_form_extra_fields['model_input'] = StringField(
_('模型输入'),
default=service.model_input if service else input_demo,
description='目前仅支持onnx/tensorrt/torch模型的triton gpu推理加速',
widget=MyBS3TextAreaFieldWidget(rows=5),
validators=[DataRequired()]
)
self.add_form_extra_fields['model_output'] = StringField(
_('模型输出'),
default=service.model_output if service else output_demo,
description='目前仅支持onnx/tensorrt/torch模型的triton gpu推理加速',
widget=MyBS3TextAreaFieldWidget(rows=5),
validators=[DataRequired()]
)
model_columns.append('model_input')
model_columns.append('model_output')
# if 'kfserving' in service_type:
# model_columns.append('transformer')
add_fieldsets = [
(
lazy_gettext('模型配置'),
{"fields": model_columns, "expanded": True},
),
(
lazy_gettext('推理配置'),
{"fields": service_columns, "expanded": True},
),
(
lazy_gettext('管理员配置'),
{"fields": admin_columns, "expanded": service_type==self.custom_service},
)
]
add_columns=model_columns+service_columns+admin_columns
self.add_columns=add_columns
self.edit_columns=self.add_columns
self.add_fieldsets=add_fieldsets
self.edit_fieldsets=self.add_fieldsets
self.edit_form_extra_fields=self.add_form_extra_fields
# self.show_columns=list(set(self.show_columns+add_columns+self.edit_columns+self.list_columns))
# print('----------')
# print(self.add_columns)
# print(self.show_columns)
# print('----------')
pre_add_get=set_column
pre_update_get=set_column
# @pysnooper.snoop()
def tfserving_model_config(self,model_name,model_version,model_path):
config_str='''
model_config_list {
config {
name: "%s"
base_path: "/%s/"
model_platform: "tensorflow"
model_version_policy {
specific {
versions: %s
}
}
}
}
'''%(model_name,model_path.strip('/'),model_version)
return config_str
def tfserving_monitoring_config(self):
config_str='''
prometheus_config {
enable: true
path: "/metrics"
}
'''
return config_str
def tfserving_platform_config(self):
config_str = '''
platform_configs {
key: "tensorflow"
value {
source_adapter_config {
[type.googleapis.com/tensorflow.serving.SavedModelBundleSourceAdapterConfig] {
legacy_config {
session_config {
gpu_options {
allow_growth: true
}
}
}
}
}
}
}
'''
return config_str
# 这些配置可在环境变量中 TS_<PROPERTY_NAME>中实现
def torch_config(self):
config_str='''
inference_address=http://0.0.0.0:8080
management_address=http://0.0.0.0:8081
metrics_address=http://0.0.0.0:8082
cors_allowed_origin=*
cors_allowed_methods=GET, POST, PUT, OPTIONS
cors_allowed_headers=X-Custom-Header
number_of_netty_threads=32
enable_metrics_api=true
job_queue_size=1000
enable_envvars_config=true
async_logging=true
default_response_timeout=120
max_request_size=6553500
vmargs=-Dlog4j.configurationFile=file:///config/log4j2.xml
'''
return config_str
def torch_log(self):
config_str='''
<RollingFile
name="access_log"
fileName="${env:LOG_LOCATION:-logs}/access_log.log"
filePattern="${env:LOG_LOCATION:-logs}/access_log.%d{dd-MMM}.log.gz">
<PatternLayout pattern="%d{ISO8601} - %m%n"/>
<Policies>
<SizeBasedTriggeringPolicy size="100 MB"/>
<TimeBasedTriggeringPolicy/>
</Policies>
<DefaultRolloverStrategy max="5"/>
</RollingFile>
<RollingFile
name="ts_log"
fileName="${env:LOG_LOCATION:-logs}/ts_log.log"
filePattern="${env:LOG_LOCATION:-logs}/ts_log.%d{dd-MMM}.log.gz">
<PatternLayout pattern="%d{ISO8601} [%-5p] %t %c - %m%n"/>
<Policies>
<SizeBasedTriggeringPolicy size="100 MB"/>
<TimeBasedTriggeringPolicy/>
</Policies>
<DefaultRolloverStrategy max="5"/>
</RollingFile>
'''
return config_str
def triton_config(self,model_name,model_input,model_output,model_type):
plat_form={
"onnx":"onnxruntime_onnx",
"tensorrt":"tensorrt_plan",
"torch":"pytorch_libtorch",
"pytorch":"pytorch_libtorch",
"tf":"tensorflow_savedmodel"
}
parameters=''
if model_type == 'tf':
parameters = '''
optimization { execution_accelerators {
gpu_execution_accelerator : [ {
name : "tensorrt"
parameters { key: "precision_mode" value: "FP16" }}]
}}
'''
if model_type=='onnx':
parameters = '''
parameters { key: "intra_op_thread_count" value: { string_value: "0" } }
parameters { key: "execution_mode" value: { string_value: "1" } }
parameters { key: "inter_op_thread_count" value: { string_value: "0" } }
'''
if model_type=='pytorch' or model_type=='torch':
parameters = '''
parameters: { key: "DISABLE_OPTIMIZED_EXECUTION" value: { string_value:"true" } }
parameters: { key: "INFERENCE_MODE" value: { string_value: "false" } }
'''
config_str = '''
name: "%s"
platform: "%s"
max_batch_size: 0
input %s
output %s
%s
instance_group [
{
count: 1
kind: KIND_GPU
}
]
'''%(model_name,plat_form[model_type],model_input,model_output,parameters)
return config_str
# @pysnooper.snoop(watch_explode=('item'))
def use_expand(self, item):
# 先存储特定参数到expand
expand = json.loads(item.expand) if item.expand else {}
print(self.src_item_json)
expand['service_token'] = json.loads(self.src_item_json['expand']).get('service_token','') if item.id else ''
expand['alias_token'] = json.loads(self.src_item_json['expand']).get('alias_token', '') if item.id else ''
expand['alias_l5'] = json.loads(self.src_item_json['expand']).get('alias_l5', '') if item.id else ''
model_version = item.model_version.replace('v','').replace('.','').replace(':','')
if item.service_type=='tfserving':
model_path=item.model_path.strip('/')
des_model_path = "/models/%s/" % (item.model_name,)
des_version_path = "/models/%s/%s/"%(item.model_name,model_version)
if not item.id or not item.command:
item.command='mkdir -p %s && cp -r /%s/* %s && /usr/bin/tf_serving_entrypoint.sh --model_config_file=/config/models.config --monitoring_config_file=/config/monitoring.config --platform_config_file=/config/platform.config'%(des_version_path,model_path,des_version_path)
item.health='8501:/v1/models/%s/versions/%s/metadata'%(item.model_name,model_version)
expand['models.config']=expand['models.config'] if expand.get('models.config','') else self.tfserving_model_config(item.model_name,model_version,des_model_path)
expand['monitoring.config']=expand['monitoring.config'] if expand.get('monitoring.config','') else self.tfserving_monitoring_config()
expand['platform.config'] = expand['platform.config'] if expand.get('platform.config','') else self.tfserving_platform_config()
if item.service_type=='torch-server':
if '.mar' not in item.model_path:
tar_command = 'torch-model-archiver --model-name %s --version %s --handler %s --serialized-file %s --export-path /models -f'%(item.model_name,model_version,item.transformer or item.model_type,item.model_path)
else:
tar_command='cp %s /models/'%item.model_path
if not item.id or not item.command:
item.command='mkdir -p /models && cp /config/* /models/ && '+tar_command+' && torchserve --start --model-store /models --models %s=%s.mar --foreground --ts-config=/config/config.properties'%(item.model_name,item.model_name)
if not item.working_dir:
item.working_dir='/models'
expand['config.properties'] = expand['config.properties'] if expand.get('config.properties','') else self.torch_config()
expand['log4j2.xml'] = expand['log4j2.xml'] if expand.get('log4j2.xml','') else self.torch_log()
if item.service_type=='triton-server':
# 识别模型类型
model_type = item.model_path.split(":")[0]
model_path = item.model_path.split(":")[1]
if not item.id or not item.command:
if model_type=='tf':
item.command='mkdir -p /models/{model_name}/{model_version}/model.savedmodel && cp /config/* /models/{model_name}/ && cp -r /{model_path}/* /models/{model_name}/{model_version}/model.savedmodel && tritonserver --model-repository=/models --strict-model-config=true --log-verbose=1'.format(model_path=model_path.strip('/'),model_name=item.model_name,model_version=model_version)
else:
model_file_ext = item.model_path.split(".")[-1]
item.command='mkdir -p /models/{model_name}/{model_version}/ && cp /config/* /models/{model_name}/ && cp -r {model_path} /models/{model_name}/{model_version}/model.{model_file_ext} && tritonserver --model-repository=/models --strict-model-config=true --log-verbose=1'.format(model_path=model_path,model_name=item.model_name,model_version=model_version,model_file_ext=model_file_ext)
config_str = self.triton_config(item.model_name,item.model_input,item.model_output,model_type)
old_config_str = json.loads(self.src_item_json['expand']).get('config.pbtxt','') if item.id else ''
new_config_str = expand.get('config.pbtxt','')
if not item.id:
expand['config.pbtxt']=config_str
elif new_config_str==old_config_str and new_config_str!=config_str:
expand['config.pbtxt']=config_str
elif not new_config_str:
expand['config.pbtxt'] = config_str
if item.service_type=='onnxruntime':
if not item.id or not item.command:
item.command='./onnxruntime_server --log_level info --model_path %s'%item.model_path
item.name=item.service_type+"-"+item.model_name+"-"+model_version
item.expand = json.dumps(expand,indent=4,ensure_ascii=False)
# @pysnooper.snoop()
def pre_add(self, item):
if not item.volume_mount:
item.volume_mount=item.project.volume_mount
self.use_expand(item)
def delete_old_service(self,service_name,cluster):
from myapp.utils.py.py_k8s import K8s
k8s_client = K8s(cluster['KUBECONFIG'])
service_namespace = conf.get('SERVICE_NAMESPACE')
kfserving_namespace = conf.get('KFSERVING_NAMESPACE')
for namespace in [service_namespace,kfserving_namespace]:
for name in [service_name,'debug-'+service_name,'test-'+service_name]:
service_external_name = (name + "-external").lower()[:60].strip('-')
k8s_client.delete_deployment(namespace=namespace, name=name)
k8s_client.delete_service(namespace=namespace, name=name)
k8s_client.delete_service(namespace=namespace, name=service_external_name)
k8s_client.delete_istio_ingress(namespace=namespace, name=name)
k8s_client.delete_hpa(namespace=namespace, name=name)
k8s_client.delete_configmap(namespace=namespace, name=name)
isvc_crd=conf.get('CRD_INFO')['inferenceservice']
k8s_client.delete_crd(isvc_crd['group'],isvc_crd['version'],isvc_crd['plural'],namespace=namespace,name=name)
# @pysnooper.snoop(watch_explode=('item',))
def pre_update(self, item):
# 修改了名称的话，要把之前的删掉
self.use_expand(item)
def pre_delete(self, item):
self.delete_old_service(item.name,item.project.cluster)
flash('服务已清理完成', category='warning')
@expose('/clear/<service_id>', methods=['POST', "GET"])
# @pysnooper.snoop()
def clear(self, service_id):
service = db.session.query(InferenceService).filter_by(id=service_id).first()
self.delete_old_service(service.name,service.project.cluster)
service.model_status='offline'
if not service.deploy_history:
service.deploy_history=''
service.deploy_history = service.deploy_history + "\n" + "clear：%s" % datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')
db.session.commit()
flash('服务清理完成', category='warning')
return redirect('/inferenceservice_modelview/list/')
# @pysnooper.snoop()
def create_polaris(self,service):
try:
# l5的值创建以后是不应该变的，一个服务对应一个固定的l5，不然客户端需要修改代码
from myapp.utils.py.py_polaris import Polaris
polaris = Polaris()
alias_token = json.loads(service.expand).get('alias_token','') if service.expand else ''
alias_l5 = json.loads(service.expand).get('alias_l5','') if service.expand else '' # 只创建一次
service_name = '%s.service' % (service.name)
username = service.created_by.username + "," + conf.get('ADMIN_USER')
if not alias_l5:
alias = polaris.get_alias(service_name)
if len(alias)>0 and alias[0]['alias']!=alias_l5:
flash('创建失败，存在系统无法识别的北极星别名，请先联系管理员手动处理','warning')
return
polaris.delete_instances(service_name)
polaris.delete_alias(service_name,alias_token)
polaris.delete_service(service_name)
polaris_service = polaris.register_service(username, service_name)
print(polaris_service)
service_token = polaris_service['token'] if polaris_service else ''
alias = polaris.register_alias(username, service_name)
expand = json.loads(service.expand) if service.expand else {}
expand.update(
{
"service_token": service_token,
"alias_token": alias['service_token'],
"alias_l5": alias['alias'],
}
)
service.expand = json.dumps(expand, indent=4, ensure_ascii=False)
db.session.commit()
polaris.delete_instances(service_name)
instances = polaris.register_instances(service_name,conf.get('SERVICE_EXTERNAL_IP'),30000+10*service.id)
print(instances)
except Exception as e:
print(e)
flash('部署北极星失败:%s'%str(e),'warning')
#
# # 针对kfserving框架，单独的部署方式
# @pysnooper.snoop()
# def deploy_kfserving(self,service):
# from myapp.utils.py.py_k8s import K8s
# k8s_client = K8s(service.project.cluster['KUBECONFIG'])
# namespace = conf.get('KFSERVING_NAMESPACE')
#
# crd_info=conf.get('CRD_INFO')['inferenceservice']
#
#
# crd_list = k8s_client.get_crd(group=crd_info['group'], version=crd_info['version'], plural=crd_info['plural'],namespace=namespace)
# for vs_obj in crd_list:
# if vs_obj['name'] == service.name:
# k8s_client.delete_crd(group=crd_info['group'], version=crd_info['version'], plural=crd_info['plural'],namespace=namespace, name=vs_obj['name'])
# time.sleep(1)
#
# crd_json = {
# "apiVersion": "%s/%s"%(crd_info['group'],crd_info['version']),
# "kind": crd_info['kind'],
# "metadata": {
# "name": service.name,
# "namespace": namespace,
# "labels": {
# "app": service.name,
# "rtx-user": service.created_by.username
# }
# },
# "spec": {
# "predictor":
# {
# "min_replicas":service.min_replicas,
# "max_replicas":service.max_replicas,
# "pytorch": {
# "storageUri": "gs://kfserving-examples/models/torchserve/image_classifier"
# }
# }
# }
# }
#
# print(crd_json)
# crd = k8s_client.create_crd(group=crd_info['group'], version=crd_info['version'], plural=crd_info['plural'],namespace=namespace, body=crd_json)
#
#
# flash('服务部署完成', category='success')
# return redirect('/inferenceservice_modelview/list/')
@expose('/debug/<service_id>',methods=['POST',"GET"])
# @pysnooper.snoop()
def deploy_debug(self,service_id):
return self.deploy(service_id,env='debug')
@expose('/deploy/test/<service_id>',methods=['POST',"GET"])
# @pysnooper.snoop()
def deploy_test(self,service_id):
return self.deploy(service_id,env='test')
@expose('/deploy/prod/<service_id>', methods=['POST', "GET"])
# @pysnooper.snoop()
def deploy_prod(self, service_id):
return self.deploy(service_id,env='prod')
@pysnooper.snoop()
def deploy(self,service_id,env='prod'):
service = db.session.query(InferenceService).filter_by(id=service_id).first()
namespace = conf.get('SERVICE_NAMESPACE','service')
pre_namespace = conf.get('PRE_SERVICE_NAMESPACE','pre-service')
name = service.name
command = service.command
deployment_replicas = service.min_replicas
if env=='debug':
name = env+'-'+service.name
command = 'sleep 43200'
deployment_replicas = 1
# namespace=pre_namespace
if env =='test':
name = env+'-'+service.name
# namespace=pre_namespace
# if 'kfserving' in service.service_type:
# return self.deploy_kfserving(service)
image_secrets = conf.get('HUBSECRET', [])
user_hubsecrets = db.session.query(Repository.hubsecret).filter(Repository.created_by_fk == g.user.id).all()
if user_hubsecrets:
for hubsecret in user_hubsecrets:
if hubsecret[0] not in image_secrets:
image_secrets.append(hubsecret[0])
from myapp.utils.py.py_k8s import K8s
k8s_client = K8s(service.project.cluster['KUBECONFIG'])
expand=json.loads(service.expand)
config_data={}
if service.service_type=='tfserving':
config_data={
"models.config":expand.get('models.config').replace('\r\n','\n'),
"monitoring.config":expand.get('monitoring.config').replace('\r\n','\n'),
"platform.config": expand.get('platform.config').replace('\r\n','\n')
}
if service.service_type=='torch-server':
config_data={
"config.properties":expand.get('config.properties').replace('\r\n','\n'),
"log4j2.xml":expand.get('log4j2.xml').replace('\r\n','\n'),
}
if service.service_type=='triton-server':
config_data={
"config.pbtxt":expand.get('config.pbtxt').replace('\r\n','\n')
}
k8s_client.create_configmap(namespace=namespace,name=name,data=config_data,labels={'app':name})
volume_mount = service.volume_mount+",%s(configmap):/config/"%name
ports = [int(port) for port in service.ports.split(',')]
pod_env = service.env
pod_env+="\nKUBEFLOW_ENV="+env
pod_env+='\nKUBEFLOW_MODEL_PATH='+service.model_path if service.model_path else ''
pod_env+='\nKUBEFLOW_MODEL_VERSION='+service.model_version
pod_env+='\nKUBEFLOW_MODEL_IMAGES='+service.images
pod_env+='\nKUBEFLOW_MODEL_NAME='+service.model_name
pod_env=pod_env.strip(',')
if env=='test' or env =='debug':
try:
k8s_client.delete_deployment(namespace=namespace,name=name)
except Exception as e:
print(e)
# 因为所有的服务流量通过ingress实现，所以没有isito的envoy代理
try:
k8s_client.create_deployment(
namespace=namespace,
name=name,
replicas=deployment_replicas,
labels={"app":name,"username":service.created_by.username},
command=['sh','-c',command] if command else None,
args=None,
volume_mount=volume_mount,
working_dir=service.working_dir,
node_selector=service.get_node_selector(),
resource_memory=service.resource_memory,
resource_cpu=service.resource_cpu,
resource_gpu=service.resource_gpu if service.resource_gpu else '',
image_pull_policy='Always',
image_pull_secrets=image_secrets,
image=service.images,
hostAliases=conf.get('HOSTALIASES',''),
env=pod_env,
privileged=False,
accounts=None,
username=service.created_by.username,
ports=ports,
health=service.health if ':' in service.health else None
)
except Exception as e:
flash('deploymnet:'+str(e),'warning')
# 监控
if service.metrics:
annotations = {
"prometheus.io/scrape": "true",
"prometheus.io/port": service.metrics.split(":")[0],
"prometheus.io/path": service.metrics.split(":")[1]
}
else:
annotations={}
k8s_client.create_service(
namespace=namespace,
name=name,
username=service.created_by.username,
ports=ports,
annotations=annotations
)
# 如果域名配置的gateway，就用这个
if 'kfserving' in service.service_type:
host = service.name + "." + service.project.cluster.get('KFSERVING_DOMAIN', conf.get('KFSERVING_DOMAIN'))
else:
host = service.name+"."+ service.project.cluster.get('SERVICE_DOMAIN',conf.get('SERVICE_DOMAIN'))
if service.host:
host=service.host.replace('http://','').replace('https://','').strip()
if "/" in host:
host = host[:host.index("/")]
# 前缀来区分不同的环境服务
if env=='debug' or env=='test':
host=env+'.'+host
k8s_client.create_istio_ingress(
namespace=namespace,
name=name,
host = host,
ports=service.ports.split(','),
canary=service.canary,
shadow=service.shadow
)
# # 以ip形式访问的话，使用的代理ip。不然不好处理机器服务化机器扩容和缩容时ip变化
# SERVICE_EXTERNAL_IP = conf.get('SERVICE_EXTERNAL_IP',None)
# if SERVICE_EXTERNAL_IP:
# service_ports = [[20000+10*service.id+index,port] for index,port in enumerate(ports)]
# service_external_name = (service.name + "-external").lower()[:60].strip('-')
# k8s_client.create_service(
# namespace=namespace,
# name=service_external_name,
# username=service.created_by.username,
# ports=service_ports,
# selector={"app": service.name, 'user': service.created_by.username},
# externalIPs=conf.get('SERVICE_EXTERNAL_IP',None)
# )
# self.create_polaris(service)
if env!='debug':
hpas = re.split(',|;', service.hpa)
if not int(service.resource_gpu):
for hpa in copy.deepcopy(hpas):
if 'gpu' in hpa:
hpas.remove(hpa)
# 伸缩容
if int(service.max_replicas)>int(service.min_replicas) and service.hpa:
try:
k8s_client.create_hpa(
namespace=namespace,
name=name,
min_replicas=int(service.min_replicas),
max_replicas=int(service.max_replicas),
hpa=','.join(hpas)
)
except Exception as e:
flash('hpa:'+str(e),'warning')
# # 使用激活器
# if int(service.min_replicas)==0:
# flash('检测到最小副本为0，已加入激活器装置')
# pass
# 不记录部署测试的情况
if env =='debug' and service.model_status=='offline':
service.model_status = 'debug'
if env=='test' and service.model_status=='offline':
service.model_status = 'test'
if env=='prod':
service.model_status = 'online'
service.deploy_history=service.deploy_history+"\n"+"deploy %s：%s"%(env,datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S'))
db.session.commit()
if env=="debug":
time.sleep(2)
pods = k8s_client.get_pods(namespace=namespace,labels={"app":name})
if pods:
pod = pods[0]
return redirect("/myapp/web/debug/%s/%s/%s/%s" % (service.project.cluster['NAME'], namespace, pod['name'],name))
flash('服务部署完成',category='success')
return redirect('/inferenceservice_modelview/list/')
appbuilder.add_view(InferenceService_ModelView,"推理服务",icon = 'fa-space-shuttle',category = '服务化')
Reference in New Issue Copy Permalink