Spaces:
Running
on
Zero
Running
on
Zero
File size: 9,885 Bytes
df6c67d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 |
import os
import signal
import socket
import sys
from functools import partial
from multiprocessing import Process, Queue
from socketserver import BaseRequestHandler, BaseServer
from types import FrameType
from typing import Any, Dict, Optional, Tuple
from uuid import uuid4
from inference.core import logger
from inference.enterprise.stream_management.manager.communication import (
receive_socket_data,
send_data_trough_socket,
)
from inference.enterprise.stream_management.manager.entities import (
PIPELINE_ID_KEY,
STATUS_KEY,
TYPE_KEY,
CommandType,
ErrorType,
OperationStatus,
)
from inference.enterprise.stream_management.manager.errors import MalformedPayloadError
from inference.enterprise.stream_management.manager.inference_pipeline_manager import (
InferencePipelineManager,
)
from inference.enterprise.stream_management.manager.serialisation import (
describe_error,
prepare_error_response,
prepare_response,
)
from inference.enterprise.stream_management.manager.tcp_server import RoboflowTCPServer
PROCESSES_TABLE: Dict[str, Tuple[Process, Queue, Queue]] = {}
HEADER_SIZE = 4
SOCKET_BUFFER_SIZE = 16384
HOST = os.getenv("STREAM_MANAGER_HOST", "127.0.0.1")
PORT = int(os.getenv("STREAM_MANAGER_PORT", "7070"))
SOCKET_TIMEOUT = float(os.getenv("STREAM_MANAGER_SOCKET_TIMEOUT", "5.0"))
class InferencePipelinesManagerHandler(BaseRequestHandler):
def __init__(
self,
request: socket.socket,
client_address: Any,
server: BaseServer,
processes_table: Dict[str, Tuple[Process, Queue, Queue]],
):
self._processes_table = processes_table # in this case it's required to set the state of class before superclass init - as it invokes handle()
super().__init__(request, client_address, server)
def handle(self) -> None:
pipeline_id: Optional[str] = None
request_id = str(uuid4())
try:
data = receive_socket_data(
source=self.request,
header_size=HEADER_SIZE,
buffer_size=SOCKET_BUFFER_SIZE,
)
data[TYPE_KEY] = CommandType(data[TYPE_KEY])
if data[TYPE_KEY] is CommandType.LIST_PIPELINES:
return self._list_pipelines(request_id=request_id)
if data[TYPE_KEY] is CommandType.INIT:
return self._initialise_pipeline(request_id=request_id, command=data)
pipeline_id = data[PIPELINE_ID_KEY]
if data[TYPE_KEY] is CommandType.TERMINATE:
self._terminate_pipeline(
request_id=request_id, pipeline_id=pipeline_id, command=data
)
else:
response = handle_command(
processes_table=self._processes_table,
request_id=request_id,
pipeline_id=pipeline_id,
command=data,
)
serialised_response = prepare_response(
request_id=request_id, response=response, pipeline_id=pipeline_id
)
send_data_trough_socket(
target=self.request,
header_size=HEADER_SIZE,
data=serialised_response,
request_id=request_id,
pipeline_id=pipeline_id,
)
except (KeyError, ValueError, MalformedPayloadError) as error:
logger.error(
f"Invalid payload in processes manager. error={error} request_id={request_id}..."
)
payload = prepare_error_response(
request_id=request_id,
error=error,
error_type=ErrorType.INVALID_PAYLOAD,
pipeline_id=pipeline_id,
)
send_data_trough_socket(
target=self.request,
header_size=HEADER_SIZE,
data=payload,
request_id=request_id,
pipeline_id=pipeline_id,
)
except Exception as error:
logger.error(
f"Internal error in processes manager. error={error} request_id={request_id}..."
)
payload = prepare_error_response(
request_id=request_id,
error=error,
error_type=ErrorType.INTERNAL_ERROR,
pipeline_id=pipeline_id,
)
send_data_trough_socket(
target=self.request,
header_size=HEADER_SIZE,
data=payload,
request_id=request_id,
pipeline_id=pipeline_id,
)
def _list_pipelines(self, request_id: str) -> None:
serialised_response = prepare_response(
request_id=request_id,
response={
"pipelines": list(self._processes_table.keys()),
STATUS_KEY: OperationStatus.SUCCESS,
},
pipeline_id=None,
)
send_data_trough_socket(
target=self.request,
header_size=HEADER_SIZE,
data=serialised_response,
request_id=request_id,
)
def _initialise_pipeline(self, request_id: str, command: dict) -> None:
pipeline_id = str(uuid4())
command_queue = Queue()
responses_queue = Queue()
inference_pipeline_manager = InferencePipelineManager.init(
command_queue=command_queue,
responses_queue=responses_queue,
)
inference_pipeline_manager.start()
self._processes_table[pipeline_id] = (
inference_pipeline_manager,
command_queue,
responses_queue,
)
command_queue.put((request_id, command))
response = get_response_ignoring_thrash(
responses_queue=responses_queue, matching_request_id=request_id
)
serialised_response = prepare_response(
request_id=request_id, response=response, pipeline_id=pipeline_id
)
send_data_trough_socket(
target=self.request,
header_size=HEADER_SIZE,
data=serialised_response,
request_id=request_id,
pipeline_id=pipeline_id,
)
def _terminate_pipeline(
self, request_id: str, pipeline_id: str, command: dict
) -> None:
response = handle_command(
processes_table=self._processes_table,
request_id=request_id,
pipeline_id=pipeline_id,
command=command,
)
if response[STATUS_KEY] is OperationStatus.SUCCESS:
logger.info(
f"Joining inference pipeline. pipeline_id={pipeline_id} request_id={request_id}"
)
join_inference_pipeline(
processes_table=self._processes_table, pipeline_id=pipeline_id
)
logger.info(
f"Joined inference pipeline. pipeline_id={pipeline_id} request_id={request_id}"
)
serialised_response = prepare_response(
request_id=request_id, response=response, pipeline_id=pipeline_id
)
send_data_trough_socket(
target=self.request,
header_size=HEADER_SIZE,
data=serialised_response,
request_id=request_id,
pipeline_id=pipeline_id,
)
def handle_command(
processes_table: Dict[str, Tuple[Process, Queue, Queue]],
request_id: str,
pipeline_id: str,
command: dict,
) -> dict:
if pipeline_id not in processes_table:
return describe_error(exception=None, error_type=ErrorType.NOT_FOUND)
_, command_queue, responses_queue = processes_table[pipeline_id]
command_queue.put((request_id, command))
return get_response_ignoring_thrash(
responses_queue=responses_queue, matching_request_id=request_id
)
def get_response_ignoring_thrash(
responses_queue: Queue, matching_request_id: str
) -> dict:
while True:
response = responses_queue.get()
if response[0] == matching_request_id:
return response[1]
logger.warning(
f"Dropping response for request_id={response[0]} with payload={response[1]}"
)
def execute_termination(
signal_number: int,
frame: FrameType,
processes_table: Dict[str, Tuple[Process, Queue, Queue]],
) -> None:
pipeline_ids = list(processes_table.keys())
for pipeline_id in pipeline_ids:
logger.info(f"Terminating pipeline: {pipeline_id}")
processes_table[pipeline_id][0].terminate()
logger.info(f"Pipeline: {pipeline_id} terminated.")
logger.info(f"Joining pipeline: {pipeline_id}")
processes_table[pipeline_id][0].join()
logger.info(f"Pipeline: {pipeline_id} joined.")
logger.info(f"Termination handler completed.")
sys.exit(0)
def join_inference_pipeline(
processes_table: Dict[str, Tuple[Process, Queue, Queue]], pipeline_id: str
) -> None:
inference_pipeline_manager, command_queue, responses_queue = processes_table[
pipeline_id
]
inference_pipeline_manager.join()
del processes_table[pipeline_id]
if __name__ == "__main__":
signal.signal(
signal.SIGINT, partial(execute_termination, processes_table=PROCESSES_TABLE)
)
signal.signal(
signal.SIGTERM, partial(execute_termination, processes_table=PROCESSES_TABLE)
)
with RoboflowTCPServer(
server_address=(HOST, PORT),
handler_class=partial(
InferencePipelinesManagerHandler, processes_table=PROCESSES_TABLE
),
socket_operations_timeout=SOCKET_TIMEOUT,
) as tcp_server:
logger.info(
f"Inference Pipeline Processes Manager is ready to accept connections at {(HOST, PORT)}"
)
tcp_server.serve_forever()
|