|
|
- import logging
- import logging.handlers
- import requests
- import traceback
- import datetime
- import json
-
- from threading import Event, Thread, Condition, Lock, enumerate
- from time import sleep
-
-
- class LogzioHandler(logging.Handler):
-
- # Hold all logs buffered
- logs = []
-
- # Event for locking buffer additions while draining
- buffer_event = Event()
-
- # Condition to count log messages
- logs_counter_condition = Condition()
-
- # Lock to only drain logs once
- drain_lock = Lock()
-
- def __init__(self, token, logs_drain_count=100, logs_drain_timeout=10,
- logzio_type="python", url="https://listener.logz.io:8071/"):
-
- if token is "":
- raise Exception("Logz.io Token must be provided")
-
- logging.Handler.__init__(self)
- self.logs_drain_count = logs_drain_count
- self.logs_drain_timeout = logs_drain_timeout
- self.logzio_type = logzio_type
- self.url = "{0}?token={1}".format(url, token)
-
- self.is_main_thread_active = lambda: any((i.name == "MainThread") and i.is_alive() for i in enumerate())
-
- self.buffer_event.set()
-
- # Create threads
- timeout_thread = Thread(target=self.wait_to_timeout_and_drain)
- counter_thread = Thread(target=self.count_logs_and_drain)
-
- # And start them
- timeout_thread.start()
- counter_thread.start()
-
- def wait_to_timeout_and_drain(self):
-
- while True:
- sleep(self.logs_drain_timeout)
- if len(self.logs) > 0:
- self.drain_messages()
-
- if not self.is_main_thread_active():
- # Signal the counter thread so it would exit as well
- try:
- self.logs_counter_condition.acquire()
- self.logs_counter_condition.notify()
- finally:
- self.logs_counter_condition.release()
- break
-
- def count_logs_and_drain(self):
- try:
- # Acquire the condition
- self.logs_counter_condition.acquire()
-
- # Running indefinite
- while True:
-
- # Waiting for new log lines to come
- self.logs_counter_condition.wait()
-
- if not self.is_main_thread_active():
- break
-
- # Do we have enough logs to drain?
- if len(self.logs) >= self.logs_drain_count:
- self.drain_messages()
-
- finally:
- self.logs_counter_condition.release()
-
- def add_to_buffer(self, message):
-
- # Check if we are currently draining buffer so we wont loose logs
- self.buffer_event.wait()
-
- try:
- # Acquire the condition
- self.logs_counter_condition.acquire()
- self.logs.append(json.dumps(message))
-
- # Notify watcher for a new log coming in
- self.logs_counter_condition.notify()
-
- finally:
- # Release the condition
- self.logs_counter_condition.release()
-
- def handle_exceptions(self, message):
- if message.exc_info:
- return '\n'.join(traceback.format_exception(*message.exc_info))
- else:
- return message.getMessage()
-
- def format_message(self, message):
-
- message_field = self.handle_exceptions(message)
- now = datetime.datetime.utcnow()
- timestamp = now.strftime("%Y-%m-%dT%H:%M:%S") + ".%03d" % (now.microsecond / 1000) + "Z"
-
- return_json = {
- "logger": message.name,
- "line_number": message.lineno,
- "path_name": message.pathname,
- "log_level": message.levelname,
- "message": message_field,
- "type": self.logzio_type,
- "@timestamp": timestamp
- }
-
- return return_json
-
- def backup_logs(self, logs):
- timestamp = datetime.datetime.now().strftime("%d%m%Y-%H%M%S")
- print("Backing up your logs to logzio-failures-{0}.txt".format(timestamp))
- with open("logzio-failures-{0}.txt".format(timestamp), "a") as f:
- f.writelines('\n'.join(logs))
-
- def drain_messages(self):
- try:
- self.buffer_event.clear()
- self.drain_lock.acquire()
-
- # Copy buffer
- temp_logs = list(self.logs)
- self.logs = []
-
- # Release the event
- self.buffer_event.set()
-
- # Not configurable from the outside
- sleep_between_retries = 2000
- number_of_retries = 4
-
- success_in_send = False
- headers = {"Content-type": "text/plain"}
-
- for current_try in range(number_of_retries):
- response = requests.post(self.url, headers=headers, data='\n'.join(temp_logs))
-
- if response.status_code != 200: # 429 400, on 400 print stdout
- if response.status_code == 400:
-
- print("Got unexpected 400 code from logz.io, response: {0}".format(response.text))
- self.backup_logs(temp_logs)
-
- if response.status_code == 401:
- print("You are not authorized with logz.io! dropping..")
- break
-
- sleep(sleep_between_retries)
- sleep_between_retries *= 2
- else:
- success_in_send = True
- break
-
- if not success_in_send:
-
- # Write to file
- self.backup_logs(temp_logs)
-
- finally:
- self.buffer_event.set()
- self.drain_lock.release()
-
- def emit(self, record):
- self.add_to_buffer(self.format_message(record))
|