Merge pull request 'metric' (#82) from metric into master

Reviewed-on: #82

.deploy/deploy-dev.yaml

@@ -22,7 +22,6 @@ services:
     networks:
       - configurator
       - queues-development
-      - locks-development
     environment:
       STAGE: "development"
     command: mailbox
@@ -39,5 +38,3 @@ networks:
     external: true
   queues-development:
     external: true
-  locks-development:
-    external: true

.deploy/deploy-prod.yaml

@@ -22,7 +22,6 @@ services:
     networks:
       - configurator
       - queues
-      - locks
     environment:
       STAGE: "production"
     command: mailbox
@@ -39,5 +38,3 @@ networks:
     external: true
   queues:
     external: true
-  locks:
-    external: true

.gitignore

@@ -121,3 +121,4 @@ GitHub.sublime-settings
 local_platform.json
 
 *pb2*
+schemas

Dockerfile

@@ -4,6 +4,5 @@ WORKDIR /usr/src/app
 COPY requirements.txt requirements.txt
 RUN pip install -r requirements.txt
 COPY . .
-RUN make gen
 ENV PYTHONUNBUFFERED 1
 ENTRYPOINT ["python", "main.py"]

Makefile

@@ -1,2 +0,0 @@
-gen:
-	python -m grpc_tools.protoc --proto_path schemas --python_out=. --pyi_out=. --grpc_python_out=. ./schemas/tasks.proto

daemons/base.py

@@ -1,18 +1,3 @@
-import os
-import grpc
-import tasks_pb2_grpc
-
-
-stage = os.getenv("STAGE", 'local')
-if stage == 'local':
-    QUEUES_URL = 'localhost:50051'
-else:
-    QUEUES_URL = 'queues-grpc:50051'
-
-
 class Daemon:
-    def __init__(self):
-        self.channel = grpc.insecure_channel(QUEUES_URL)
-        self.stub = tasks_pb2_grpc.TasksStub(channel=self.channel)     
     def execute(self):
         raise NotImplemented

daemons/poll.py

@@ -1,6 +1,7 @@
 import telebot
-import threading
+import multiprocessing
 import time
+import json
 
 from daemons import base
 from utils import platform
@@ -9,45 +10,37 @@ from utils import queues
 
 class Daemon(base.Daemon):
     def __init__(self):
-        self.telegram_bots: dict[str, dict[str, telebot.TeleBot|None]] = {}
-        self.threads: dict[str, dict[str, threading.Thread|None]] = {}
+        self.processes: dict[str, multiprocessing.Process|None] = {}
 
     def execute(self):
         while True:
             bots = platform.platform_client.get_config('bots')
             for project_name, project in bots.items():
-                if project_name not in self.telegram_bots:
-                    self.telegram_bots[project_name] = {}
-                    self.threads[project_name] = {}
                 for bot_name, bot_info in project.items():
-                    if bot_name not in self.telegram_bots[project_name]:
-                        self.telegram_bots[project_name][bot_name] = None
-                        self.threads[project_name][bot_name] = None
-                    bot = self.telegram_bots[project_name][bot_name]
+                    key = f'{project_name}_{bot_name}'
+                    proc = self.processes.get(key)
                     if bot_info.get('poll_enabled'):
-                        if bot is not None and self.threads[project_name][bot_name].is_alive():
+                        if proc and proc.is_alive():
                             print(f'process for {project_name} {bot_name} is alive')
                             continue
                         print(f'starting process for {project_name} {bot_name}')
-                        bot = telebot.TeleBot(bot_info['secrets']['telegram_token'])
-                        thread = self.start_polling(bot, bot_info['queue'])
-                        self.telegram_bots[project_name][bot_name] = bot 
-                        self.threads[project_name][bot_name] = thread 
+                        process = multiprocessing.Process(target=self.start_polling, args=(bot_info['secrets']['telegram_token'], bot_info['queue']))
+                        process.start()
+                        self.processes[key] = process
                         print(f'started process for {project_name} {bot_name}')
                     else:
-                        if bot is None:
+                        if proc is None:
                             print(f'process for {project_name} {bot_name} is not alive')
                             continue
                         print(f'terminating process for {project_name} {bot_name}')
-                        bot.stop_bot()
-                        self.telegram_bots[project_name][bot_name] = None
+                        proc.terminate()
+                        self.processes[key] = None
                         print(f'terminated process for {project_name} {bot_name}')
             time.sleep(10)
 
-    def start_polling(self, bot: telebot.TeleBot, queue: str) -> threading.Thread:
+    def start_polling(self, token: str, queue: str):
+        bot = telebot.TeleBot(token)
         @bot.message_handler(content_types=['audio', 'photo', 'voice', 'video', 'document', 'animation', 'text', 'location', 'contact', 'sticker', 'video_note'])
         def do_action(message: telebot.types.Message):
-            queues.set_task(self.stub, queue, message.json, 1)
-        thread = threading.Thread(target=bot.polling)
-        thread.start()
-        return thread
+            queues.set_task(queue, message.json, 1)
+        bot.polling()

requirements.txt

@@ -1,14 +1,10 @@
 annotated-types==0.7.0
-certifi==2024.8.30
+certifi==2024.12.14
 charset-normalizer==3.4.0
-grpcio==1.68.1
-grpcio-tools==1.68.1
 idna==3.10
-protobuf==5.29.1
-pydantic==2.10.2
-pydantic_core==2.27.1
+pydantic==2.10.4
+pydantic_core==2.27.2
 pyTelegramBotAPI==4.1.1
 requests==2.32.3
-setuptools==75.6.0
 typing_extensions==4.12.2
-urllib3==2.2.3
+urllib3==2.3.0

schemas/tasks.proto

@@ -1,40 +0,0 @@
-syntax = "proto3";
-
-package queues;
-
-import "google/protobuf/struct.proto";
-
-service Tasks {
-    rpc Put (PutRequest) returns (EmptyResponse) {}
-
-    rpc Take (TakeRequest) returns (TakeResponse) {}
-
-    rpc Finish (FinishRequest) returns (EmptyResponse) {}
-}
-
-message Task {
-    string id = 1;
-    int64 attempt = 2;
-    google.protobuf.Struct payload = 3;
-}
-
-message PutRequest {
-    string queue = 1;
-    int64 seconds_to_execute = 2;
-    optional int64 delay = 3;
-    google.protobuf.Struct payload = 4;
-}
-
-message TakeRequest {
-    string queue = 1;
-}
-
-message FinishRequest {
-    string id = 1;
-}
-
-message EmptyResponse {}
-
-message TakeResponse {
-    optional Task task = 1;
-}

utils/queues.py

@@ -1,6 +1,19 @@
+from concurrent.futures import ThreadPoolExecutor
+import datetime
+import json
+import os
+import traceback
+import uuid
+import zoneinfo
+import requests
 import time
-import tasks_pb2_grpc
-import tasks_pb2
+
+
+stage = os.getenv("STAGE", 'local')
+if stage == 'local':
+    QUEUES_URL = 'http://localhost:1239'
+else:
+    QUEUES_URL = 'http://queues:1239'
 
 
 class QueuesException(Exception):
@@ -8,22 +21,55 @@ class QueuesException(Exception):
 
 
 class TasksHandlerMixin:
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.executor = ThreadPoolExecutor(max_workers=1)
+
+    def _send_metric(self, start: datetime.datetime, end: datetime.datetime, success: bool):
+        def send():
+            requests.post(f'{QUEUES_URL}/api/v1/metric', json={
+                'service': 'botalka',
+                'queue': self.queue_name,
+                'success': success,
+                'timestamp': start.strftime("%Y-%m-%dT%H:%M:%S") + "Z",
+                "success": success,
+                "execution_time_ms": (end - start).microseconds // 1000,
+                "environment": stage,
+            })
+        
+        self.executor.submit(send)
+
     def poll(self):
         while True:
-            response: tasks_pb2.TakeResponse = self.stub.Take(tasks_pb2.TakeRequest(self.queue_name))
-            task: tasks_pb2.Task = response.task
+            try:
+                response = requests.get(f'{QUEUES_URL}/api/v1/take', headers={'queue': self.queue_name}).json()
+            except requests.JSONDecodeError:
+                print('Unable to decode json')
+                time.sleep(3)
+                continue
+            task = response.get('task')
             if not task:
                 time.sleep(0.2)
                 continue
+            start = datetime.datetime.now(zoneinfo.ZoneInfo("Europe/Moscow"))
             try:
-                self.process(task.payload)
+                print(f'process task with id {task["id"]}, attempt {task["attempt"]}')
+                self.process(task['payload'])
+                success = True
             except Exception as exc:
                 print(f'Error processing message id={task["id"]}, payload={task["payload"]}, exc={exc}')
-                continue
+                traceback.print_stack()
+                success = False
+            end = datetime.datetime.now(zoneinfo.ZoneInfo("Europe/Moscow"))
+            if success:
                 try:
-                self.stub.Finish(tasks_pb2.FinishRequest(task.id))
+                    resp = requests.post(f'{QUEUES_URL}/api/v1/finish', json={'id': task['id']})
+                    if resp.status_code != 202:
+                        raise QueuesException
+                    print(f'finish task with id {task["id"]}')
                 except:
-                print(f'Failed to finish task id={task.id}')
+                    print(f'Failed to finish task id={task["id"]}')
+            self._send_metric(start, end, success)
 
     @property
     def queue_name(self):
@@ -32,12 +78,12 @@ class TasksHandlerMixin:
     def process(self, payload):
         raise NotImplemented
 
-def set_task(stub: tasks_pb2_grpc.TasksStub, queue_name: str, payload: dict, seconds_to_execute: int, delay: int|None = None):
-    stub.Put(
-        tasks_pb2.PutRequest(
-            queue=queue_name, 
-            seconds_to_execute=seconds_to_execute, 
-            delay=delay, 
-            payload=payload
-        )
-    )
+
+def set_task(queue_name: str, payload: dict, seconds_to_execute: int, delay: int|None = None):
+    resp = requests.post(f'{QUEUES_URL}/api/v1/put', headers={'queue': queue_name}, json={
+        'payload': payload,
+        'seconds_to_execute': seconds_to_execute,
+        'delay': delay,
+    })
+    if resp.status_code != 202:
+        raise QueuesException