Skip to content

V
VisualDL

PaddlePaddle / VisualDL
大约 1 年 前同步成功

通知 88
Star 4655
Fork 642
V
VisualDL
未验证 提交 9583bf7c 编写于 作者: C chenjian 提交者： GitHub
浏览文件
操作

add English version for fastdeploy client (#1197) 

* add english page for fastdeploy client

* fix a bug
上级 a418dd44
展开全部 隐藏空白更改
内联 并排
Showing with 598 addition and 80 deletion
visualdl/component/inference/fastdeploy_client/http_client_manager.py
浏览文件 @ 9583bf7c
...@@ -19,7 +19,6 @@ import numpy as np ...@@ -19,7 +19,6 @@ import numpy as np
import requests import requests
import tritonclient.http as httpclient import tritonclient.http as httpclient
from attrdict import AttrDict from attrdict import AttrDict
from tritonclient.utils import InferenceServerException
def convert_http_metadata_config(metadata): def convert_http_metadata_config(metadata):
...@@ -118,8 +117,63 @@ table, th {{ ...@@ -118,8 +117,63 @@ table, th {{
</div> </div>
""" """
metrics_table_head_en = """
<style>
table, th {{
border:0.1px solid black;
}}
</style>
<div>
<table style="width:100%">
<tr>
<th rowspan="2">Model name</th>
<th colspan="4">Execution metric</th>
<th colspan="5">Delay metric</th>
</tr>
<tr>
<th>inference request success</th>
<th>inference request failure</th>
<th>inference count</th>
<th>inference exec count</th>
<th>inference request duration(ms)</th>
<th>inference queue duration(ms)</th>
<th>inference comput input duration(ms)</th>
<th>inference compute infer duration
(ms)</th>
<th>inference compute output duration(ms)</th>
</tr>
{}
</table>
</div>
<br>
<br>
<br>
<br>
<br>
<div>
<table style="width:100%">
<tr>
<th rowspan="2">GPU</th>
<th colspan="4">Performance metric</th>
<th colspan="2">Memory</th>
</tr>
<tr>
<th>utilization(%)</th>
<th>power usage(W)</th>
<th>power limit(W)</th>
<th>energy consumption(W)</th>
<th>total(GB)</th>
<th>used(GB)</th>
</tr>
{}
</table>
</div>
"""
def get_metric_data(server_addr, metric_port): # noqa:C901 def get_metric_data(server_addr, metric_port, lang='zh'): # noqa:C901
''' '''
Get metrics data from fastdeploy server, and transform it into html table. Get metrics data from fastdeploy server, and transform it into html table.
Args: Args:
...@@ -235,6 +289,8 @@ def get_metric_data(server_addr, metric_port): # noqa:C901 ...@@ -235,6 +289,8 @@ def get_metric_data(server_addr, metric_port): # noqa:C901
for item in data]) + "</tr>" for item in data]) + "</tr>"
for data in gpu_data_list for data in gpu_data_list
]) ])
if lang == 'en':
return metrics_table_head_en.format(model_data, gpu_data)
return metrics_table_head.format(model_data, gpu_data) return metrics_table_head.format(model_data, gpu_data)
...@@ -294,7 +350,7 @@ class HttpClientManager: ...@@ -294,7 +350,7 @@ class HttpClientManager:
try: try:
model_metadata = fastdeploy_client.get_model_metadata( model_metadata = fastdeploy_client.get_model_metadata(
model_name=model_name, model_version=model_version) model_name=model_name, model_version=model_version)
except InferenceServerException as e: except Exception as e:
raise RuntimeError("Failed to retrieve the metadata: " + str(e)) raise RuntimeError("Failed to retrieve the metadata: " + str(e))
model_metadata = convert_http_metadata_config(model_metadata) model_metadata = convert_http_metadata_config(model_metadata)
......
visualdl/component/inference/fastdeploy_server.py
浏览文件 @ 9583bf7c
...@@ -25,6 +25,7 @@ from pathlib import Path ...@@ -25,6 +25,7 @@ from pathlib import Path
import requests import requests
from .fastdeploy_client.client_app import create_gradio_client_app from .fastdeploy_client.client_app import create_gradio_client_app
from .fastdeploy_client.client_app import create_gradio_client_app_en
from .fastdeploy_lib import analyse_config from .fastdeploy_lib import analyse_config
from .fastdeploy_lib import check_process_zombie from .fastdeploy_lib import check_process_zombie
from .fastdeploy_lib import copy_config_file_to_default_config from .fastdeploy_lib import copy_config_file_to_default_config
...@@ -53,7 +54,8 @@ class FastDeployServerApi(object): ...@@ -53,7 +54,8 @@ class FastDeployServerApi(object):
self.root_dir = Path(os.getcwd()) self.root_dir = Path(os.getcwd())
self.opened_servers = { self.opened_servers = {
} # Use to store the opened server process pid and process itself } # Use to store the opened server process pid and process itself
self.client_port = None self.client_port = None # Chinese version
self.client_en_port = None # English version
@result() @result()
def get_directory(self, cur_dir): def get_directory(self, cur_dir):
...@@ -351,34 +353,43 @@ class FastDeployServerApi(object): ...@@ -351,34 +353,43 @@ class FastDeployServerApi(object):
version_filenames_dict_for_frontend) version_filenames_dict_for_frontend)
return version_info_for_frontend return version_info_for_frontend
def create_fastdeploy_client(self): def create_fastdeploy_client(self, lang='zh'):
if self.client_port is None: def get_free_tcp_port():
tcp = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
def get_free_tcp_port(): # tcp.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEPORT, 1)
tcp = socket.socket(socket.AF_INET, socket.SOCK_STREAM) tcp.bind(('localhost', 0))
# tcp.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEPORT, 1) addr, port = tcp.getsockname()
tcp.bind(('localhost', 0)) tcp.close()
addr, port = tcp.getsockname() return port
tcp.close()
return port def check_alive(client_port):
while True:
self.client_port = get_free_tcp_port() try:
app = create_gradio_client_app() requests.get('http://localhost:{}/'.format(client_port))
thread = Process( break
target=app.launch, kwargs={'server_port': self.client_port}) except Exception:
thread.start() time.sleep(1)
def check_alive(): if lang == 'en':
while True: if self.client_en_port is None:
try: self.client_en_port = get_free_tcp_port()
requests.get('http://localhost:{}/'.format( app = create_gradio_client_app_en()
self.client_port)) thread = Process(
break target=app.launch,
except Exception: kwargs={'server_port': self.client_en_port})
time.sleep(1) thread.start()
check_alive(self.client_en_port)
check_alive() return self.client_en_port
return self.client_port else:
if self.client_port is None:
self.client_port = get_free_tcp_port()
app = create_gradio_client_app()
thread = Process(
target=app.launch,
kwargs={'server_port': self.client_port})
thread.start()
check_alive(self.client_port)
return self.client_port
def _poll_zombie_process(self): def _poll_zombie_process(self):
# check if there are servers killed by other vdl app instance and become zoombie # check if there are servers killed by other vdl app instance and become zoombie
...@@ -410,7 +421,7 @@ def create_fastdeploy_api_call(): ...@@ -410,7 +421,7 @@ def create_fastdeploy_api_call():
'start_server': (api.start_server, ['config']), 'start_server': (api.start_server, ['config']),
'stop_server': (api.stop_server, ['server_id']), 'stop_server': (api.stop_server, ['server_id']),
'get_server_output': (api.get_server_output, ['server_id', 'length']), 'get_server_output': (api.get_server_output, ['server_id', 'length']),
'create_fastdeploy_client': (api.create_fastdeploy_client, []), 'create_fastdeploy_client': (api.create_fastdeploy_client, ['lang']),
'get_server_list': (api.get_server_list, []), 'get_server_list': (api.get_server_list, []),
'get_server_metric': (api.get_server_metric, ['server_id']), 'get_server_metric': (api.get_server_metric, ['server_id']),
'get_server_config': (api.get_server_config, ['server_id']), 'get_server_config': (api.get_server_config, ['server_id']),
......
visualdl/server/app.py
浏览文件 @ 9583bf7c
...@@ -181,6 +181,7 @@ def create_app(args): # noqa: C901 ...@@ -181,6 +181,7 @@ def create_app(args): # noqa: C901
error_msg = '{}'.format(e) error_msg = '{}'.format(e)
return make_response(error_msg) return make_response(error_msg)
args = urllib.parse.urlencode(request_args) args = urllib.parse.urlencode(request_args)
if args: if args:
return redirect( return redirect(
api_path + "/fastdeploy/fastdeploy_client/app?{}".format(args), api_path + "/fastdeploy/fastdeploy_client/app?{}".format(args),
...@@ -201,14 +202,30 @@ def create_app(args): # noqa: C901 ...@@ -201,14 +202,30 @@ def create_app(args): # noqa: C901
Returns: Returns:
Any thing from gradio server. Any thing from gradio server.
''' '''
lang = 'zh'
if request.method == 'POST': if request.method == 'POST':
if request.mimetype == 'application/json':
request_args = request.json
else:
request_args = request.form.to_dict()
if 'data' in request_args:
lang = request_args['data'][-1]
request_args['lang'] = lang
elif 'lang' in request_args:
lang = request_args['lang']
port = fastdeploy_api_call('create_fastdeploy_client', port = fastdeploy_api_call('create_fastdeploy_client',
request.form) request_args)
request_args = request.form
else: else:
request_args = request.args.to_dict()
if 'data' in request_args:
lang = request_args['data'][-1]
request_args['lang'] = lang
elif 'lang' in request_args:
lang = request_args['lang']
port = fastdeploy_api_call('create_fastdeploy_client', port = fastdeploy_api_call('create_fastdeploy_client',
request.args) request_args)
request_args = request.args
if path == 'app': if path == 'app':
proxy_url = request.url.replace( proxy_url = request.url.replace(
request.host_url.rstrip('/') + api_path + request.host_url.rstrip('/') + api_path +
...@@ -239,38 +256,82 @@ def create_app(args): # noqa: C901 ...@@ -239,38 +256,82 @@ def create_app(args): # noqa: C901
model_name = start_args.get('default_model_name', '') model_name = start_args.get('default_model_name', '')
content = content.decode() content = content.decode()
try: try:
default_server_addr = re.search( if request_args.get('lang', 'zh') == 'en':
'"label": {}.*?"value": "".*?}}'.format( default_server_addr = re.search(
json.dumps("服务ip", ensure_ascii=True).replace( '"label": {}.*?"value": "".*?}}'.format(
'\\', '\\\\')), content).group(0) json.dumps(
cur_server_addr = default_server_addr.replace( "server ip", ensure_ascii=True).replace(
'"value": ""', '"value": "localhost"') '\\', '\\\\')), content).group(0)
default_http_port = re.search( cur_server_addr = default_server_addr.replace(
'"label": {}.*?"value": "".*?}}'.format( '"value": ""', '"value": "localhost"')
json.dumps("推理服务端口", ensure_ascii=True).replace( default_http_port = re.search(
'\\', '\\\\')), content).group(0) '"label": {}.*?"value": "".*?}}'.format(
cur_http_port = default_http_port.replace( json.dumps(
'"value": ""', '"value": "{}"'.format(http_port)) "server port", ensure_ascii=True).replace(
default_metrics_port = re.search( '\\', '\\\\')), content).group(0)
'"label": {}.*?"value": "".*?}}'.format( cur_http_port = default_http_port.replace(
json.dumps("性能服务端口", ensure_ascii=True).replace( '"value": ""', '"value": "{}"'.format(http_port))
'\\', '\\\\')), content).group(0) default_metrics_port = re.search(
cur_metrics_port = default_metrics_port.replace( '"label": {}.*?"value": "".*?}}'.format(
'"value": ""', '"value": "{}"'.format(metrics_port)) json.dumps(
default_model_name = re.search( "metrics port", ensure_ascii=True).replace(
'"label": {}.*?"value": "".*?}}'.format( '\\', '\\\\')), content).group(0)
json.dumps("模型名称", ensure_ascii=True).replace( cur_metrics_port = default_metrics_port.replace(
'\\', '\\\\')), content).group(0) '"value": ""',
cur_model_name = default_model_name.replace( '"value": "{}"'.format(metrics_port))
'"value": ""', '"value": "{}"'.format(model_name)) default_model_name = re.search(
default_model_version = re.search( '"label": {}.*?"value": "".*?}}'.format(
'"label": {}.*?"value": "".*?}}'.format( json.dumps(
json.dumps("模型版本", ensure_ascii=True).replace( "model name", ensure_ascii=True).replace(
'\\', '\\\\')), content).group(0) '\\', '\\\\')), content).group(0)
cur_model_version = default_model_version.replace( cur_model_name = default_model_name.replace(
'"value": ""', '"value": "{}"'.format('1')) '"value": ""', '"value": "{}"'.format(model_name))
content = content.replace(default_server_addr, default_model_version = re.search(
cur_server_addr) '"label": {}.*?"value": "".*?}}'.format(
json.dumps("model version",
ensure_ascii=True).replace(
'\\', '\\\\')),
content).group(0)
cur_model_version = default_model_version.replace(
'"value": ""', '"value": "{}"'.format('1'))
content = content.replace(default_server_addr,
cur_server_addr)
else:
default_server_addr = re.search(
'"label": {}.*?"value": "".*?}}'.format(
json.dumps("服务ip", ensure_ascii=True).replace(
'\\', '\\\\')), content).group(0)
cur_server_addr = default_server_addr.replace(
'"value": ""', '"value": "localhost"')
default_http_port = re.search(
'"label": {}.*?"value": "".*?}}'.format(
json.dumps(
"推理服务端口", ensure_ascii=True).replace(
'\\', '\\\\')), content).group(0)
cur_http_port = default_http_port.replace(
'"value": ""', '"value": "{}"'.format(http_port))
default_metrics_port = re.search(
'"label": {}.*?"value": "".*?}}'.format(
json.dumps(
"性能服务端口", ensure_ascii=True).replace(
'\\', '\\\\')), content).group(0)
cur_metrics_port = default_metrics_port.replace(
'"value": ""',
'"value": "{}"'.format(metrics_port))
default_model_name = re.search(
'"label": {}.*?"value": "".*?}}'.format(
json.dumps("模型名称", ensure_ascii=True).replace(
'\\', '\\\\')), content).group(0)
cur_model_name = default_model_name.replace(
'"value": ""', '"value": "{}"'.format(model_name))
default_model_version = re.search(
'"label": {}.*?"value": "".*?}}'.format(
json.dumps("模型版本", ensure_ascii=True).replace(
'\\', '\\\\')), content).group(0)
cur_model_version = default_model_version.replace(
'"value": ""', '"value": "{}"'.format('1'))
content = content.replace(default_server_addr,
cur_server_addr)
if http_port: if http_port:
content = content.replace(default_http_port, content = content.replace(default_http_port,
cur_http_port) cur_http_port)
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或