使用 CloudWatch for Python (Boto3) 的 SDK Logs 範例 - AWS SDK 程式碼範例

文件 AWS SDK AWS 範例 SDK 儲存庫中有更多可用的 GitHub 範例。

本文為英文版的機器翻譯版本,如內容有任何歧義或不一致之處,概以英文版為準。

使用 CloudWatch for Python (Boto3) 的 SDK Logs 範例

下列程式碼範例示範如何使用 AWS SDK for Python (Boto3) 搭配 CloudWatch Logs 來執行動作和實作常見案例。

Actions 是大型程式的程式碼摘錄,必須在內容中執行。雖然動作會示範如何呼叫個別服務函數,但您可以在相關案例中查看內容中的動作。

案例是程式碼範例,示範如何透過呼叫服務內的多個函數或與其他函數結合,來完成特定任務 AWS 服務。

每個範例都包含完整原始程式碼的連結,您可以在其中找到如何在內容中設定和執行程式碼的指示。

動作

下列程式碼範例示範如何使用 GetQueryResults

SDK for Python (Boto3)
注意

還有更多 on GitHub。尋找完整範例,並了解如何在 AWS 程式碼範例儲存庫中設定和執行。

def _wait_for_query_results(self, client, query_id): """ Waits for the query to complete and retrieves the results. :param query_id: The ID of the initiated query. :type query_id: str :return: A list containing the results of the query. :rtype: list """ while True: time.sleep(1) results = client.get_query_results(queryId=query_id) if results["status"] in [ "Complete", "Failed", "Cancelled", "Timeout", "Unknown", ]: return results.get("results", [])
  • 如需 API 詳細資訊,請參閱 GetQueryResults AWS SDK for Python (Boto3) Word 參考中的 API

下列程式碼範例示範如何使用 StartLiveTail

SDK for Python (Boto3)

包括必需的檔案。

import boto3 import time from datetime import datetime

啟動 Live Tail 工作階段。

# Initialize the client client = boto3.client('logs') start_time = time.time() try: response = client.start_live_tail( logGroupIdentifiers=log_group_identifiers, logStreamNames=log_streams, logEventFilterPattern=filter_pattern ) event_stream = response['responseStream'] # Handle the events streamed back in the response for event in event_stream: # Set a timeout to close the stream. # This will end the Live Tail session. if (time.time() - start_time >= 10): event_stream.close() break # Handle when session is started if 'sessionStart' in event: session_start_event = event['sessionStart'] print(session_start_event) # Handle when log event is given in a session update elif 'sessionUpdate' in event: log_events = event['sessionUpdate']['sessionResults'] for log_event in log_events: print('[{date}] {log}'.format(date=datetime.fromtimestamp(log_event['timestamp']/1000),log=log_event['message'])) else: # On-stream exceptions are captured here raise RuntimeError(str(event)) except Exception as e: print(e)
  • 如需 API 詳細資訊,請參閱 StartLiveTail AWS SDK for Python (Boto3) Word 參考中的 API

下列程式碼範例示範如何使用 StartQuery

SDK for Python (Boto3)
注意

還有更多 on GitHub。尋找完整範例,並了解如何在 AWS 程式碼範例儲存庫中設定和執行。

def perform_query(self, date_range): """ Performs the actual CloudWatch log query. :param date_range: A tuple representing the start and end datetime for the query. :type date_range: tuple :return: A list containing the query results. :rtype: list """ client = boto3.client("logs") try: try: start_time = round( self.date_utilities.convert_iso8601_to_unix_timestamp(date_range[0]) ) end_time = round( self.date_utilities.convert_iso8601_to_unix_timestamp(date_range[1]) ) response = client.start_query( logGroupName=self.log_groups, startTime=start_time, endTime=end_time, queryString="fields @timestamp, @message | sort @timestamp asc", limit=self.limit, ) query_id = response["queryId"] except client.exceptions.ResourceNotFoundException as e: raise DateOutOfBoundsError(f"Resource not found: {e}") while True: time.sleep(1) results = client.get_query_results(queryId=query_id) if results["status"] in [ "Complete", "Failed", "Cancelled", "Timeout", "Unknown", ]: return results.get("results", []) except DateOutOfBoundsError: return [] def _initiate_query(self, client, date_range, max_logs): """ Initiates the CloudWatch logs query. :param date_range: A tuple representing the start and end datetime for the query. :type date_range: tuple :param max_logs: The maximum number of logs to retrieve. :type max_logs: int :return: The query ID as a string. :rtype: str """ try: start_time = round( self.date_utilities.convert_iso8601_to_unix_timestamp(date_range[0]) ) end_time = round( self.date_utilities.convert_iso8601_to_unix_timestamp(date_range[1]) ) response = client.start_query( logGroupName=self.log_groups, startTime=start_time, endTime=end_time, queryString="fields @timestamp, @message | sort @timestamp asc", limit=max_logs, ) return response["queryId"] except client.exceptions.ResourceNotFoundException as e: raise DateOutOfBoundsError(f"Resource not found: {e}")
  • 如需 API 詳細資訊,請參閱 StartQuery AWS SDK for Python (Boto3) Word 參考中的 API

案例

下列程式碼範例示範如何使用 CloudWatch Logs 查詢超過 10,000 筆記錄。

SDK for Python (Boto3)
注意

還有更多 on GitHub。尋找完整範例,並了解如何在 AWS 程式碼範例儲存庫中設定和執行。

此檔案會叫用範例模組,以管理超過 10,000 個結果的 CloudWatch 查詢。

import logging import os import sys import boto3 from botocore.config import Config from cloudwatch_query import CloudWatchQuery from date_utilities import DateUtilities # Configure logging at the module level. logging.basicConfig( level=logging.INFO, format="%(asctime)s - %(levelname)s - %(filename)s:%(lineno)d - %(message)s", ) class CloudWatchLogsQueryRunner: def __init__(self): """ Initializes the CloudWatchLogsQueryRunner class by setting up date utilities and creating a CloudWatch Logs client with retry configuration. """ self.date_utilities = DateUtilities() self.cloudwatch_logs_client = self.create_cloudwatch_logs_client() def create_cloudwatch_logs_client(self): """ Creates and returns a CloudWatch Logs client with a specified retry configuration. :return: A CloudWatch Logs client instance. :rtype: boto3.client """ try: return boto3.client("logs", config=Config(retries={"max_attempts": 10})) except Exception as e: logging.error(f"Failed to create CloudWatch Logs client: {e}") sys.exit(1) def fetch_environment_variables(self): """ Fetches and validates required environment variables for query start and end dates. :return: Tuple of query start date and end date as integers. :rtype: tuple :raises SystemExit: If required environment variables are missing or invalid. """ try: query_start_date = int(os.environ["QUERY_START_DATE"]) query_end_date = int(os.environ["QUERY_END_DATE"]) except KeyError: logging.error( "Both QUERY_START_DATE and QUERY_END_DATE environment variables are required." ) sys.exit(1) except ValueError as e: logging.error(f"Error parsing date environment variables: {e}") sys.exit(1) return query_start_date, query_end_date def convert_dates_to_iso8601(self, start_date, end_date): """ Converts UNIX timestamp dates to ISO 8601 format using DateUtilities. :param start_date: The start date in UNIX timestamp. :type start_date: int :param end_date: The end date in UNIX timestamp. :type end_date: int :return: Start and end dates in ISO 8601 format. :rtype: tuple """ start_date_iso8601 = self.date_utilities.convert_unix_timestamp_to_iso8601( start_date ) end_date_iso8601 = self.date_utilities.convert_unix_timestamp_to_iso8601( end_date ) return start_date_iso8601, end_date_iso8601 def execute_query( self, start_date_iso8601, end_date_iso8601, log_group="/workflows/cloudwatch-logs/large-query", ): """ Creates a CloudWatchQuery instance and executes the query with provided date range. :param start_date_iso8601: The start date in ISO 8601 format. :type start_date_iso8601: str :param end_date_iso8601: The end date in ISO 8601 format. :type end_date_iso8601: str :param log_group: Log group to search: "/workflows/cloudwatch-logs/large-query" :type log_group: str """ cloudwatch_query = CloudWatchQuery( [start_date_iso8601, end_date_iso8601], ) cloudwatch_query.query_logs((start_date_iso8601, end_date_iso8601)) logging.info("Query executed successfully.") logging.info( f"Queries completed in {cloudwatch_query.query_duration} seconds. Total logs found: {len(cloudwatch_query.query_results)}" ) def main(): """ Main function to start a recursive CloudWatch logs query. Fetches required environment variables, converts dates, and executes the query. """ logging.info("Starting a recursive CloudWatch logs query...") runner = CloudWatchLogsQueryRunner() query_start_date, query_end_date = runner.fetch_environment_variables() start_date_iso8601 = DateUtilities.convert_unix_timestamp_to_iso8601( query_start_date ) end_date_iso8601 = DateUtilities.convert_unix_timestamp_to_iso8601(query_end_date) runner.execute_query(start_date_iso8601, end_date_iso8601) if __name__ == "__main__": main()

本單元處理超過 10,000 個結果的 CloudWatch 查詢。

import logging import time from datetime import datetime import threading import boto3 from date_utilities import DateUtilities class DateOutOfBoundsError(Exception): """Exception raised when the date range for a query is out of bounds.""" pass class CloudWatchQuery: """ A class to query AWS CloudWatch logs within a specified date range. :ivar date_range: Start and end datetime for the query. :vartype date_range: tuple :ivar limit: Maximum number of log entries to return. :vartype limit: int """ def __init__(self, date_range): self.lock = threading.Lock() self.log_groups = "/workflows/cloudwatch-logs/large-query" self.query_results = [] self.date_range = date_range self.query_duration = None self.datetime_format = "%Y-%m-%d %H:%M:%S.%f" self.date_utilities = DateUtilities() self.limit = 10000 def query_logs(self, date_range): """ Executes a CloudWatch logs query for a specified date range and calculates the execution time of the query. :return: A batch of logs retrieved from the CloudWatch logs query. :rtype: list """ start_time = datetime.now() start_date, end_date = self.date_utilities.normalize_date_range_format( date_range, from_format="unix_timestamp", to_format="datetime" ) logging.info( f"Original query:" f"\n START: {start_date}" f"\n END: {end_date}" ) self.recursive_query((start_date, end_date)) end_time = datetime.now() self.query_duration = (end_time - start_time).total_seconds() def recursive_query(self, date_range): """ Processes logs within a given date range, fetching batches of logs recursively if necessary. :param date_range: The date range to fetch logs for, specified as a tuple (start_timestamp, end_timestamp). :type date_range: tuple :return: None if the recursive fetching is continued or stops when the final batch of logs is processed. Although it doesn't explicitly return the query results, this method accumulates all fetched logs in the `self.query_results` attribute. :rtype: None """ batch_of_logs = self.perform_query(date_range) # Add the batch to the accumulated logs with self.lock: self.query_results.extend(batch_of_logs) if len(batch_of_logs) == self.limit: logging.info(f"Fetched {self.limit}, checking for more...") most_recent_log = self.find_most_recent_log(batch_of_logs) most_recent_log_timestamp = next( item["value"] for item in most_recent_log if item["field"] == "@timestamp" ) new_range = (most_recent_log_timestamp, date_range[1]) midpoint = self.date_utilities.find_middle_time(new_range) first_half_thread = threading.Thread( target=self.recursive_query, args=((most_recent_log_timestamp, midpoint),), ) second_half_thread = threading.Thread( target=self.recursive_query, args=((midpoint, date_range[1]),) ) first_half_thread.start() second_half_thread.start() first_half_thread.join() second_half_thread.join() def find_most_recent_log(self, logs): """ Search a list of log items and return most recent log entry. :param logs: A list of logs to analyze. :return: log :type :return List containing log item details """ most_recent_log = None most_recent_date = "1970-01-01 00:00:00.000" for log in logs: for item in log: if item["field"] == "@timestamp": logging.debug(f"Compared: {item['value']} to {most_recent_date}") if ( self.date_utilities.compare_dates( item["value"], most_recent_date ) == item["value"] ): logging.debug(f"New most recent: {item['value']}") most_recent_date = item["value"] most_recent_log = log logging.info(f"Most recent log date of batch: {most_recent_date}") return most_recent_log def perform_query(self, date_range): """ Performs the actual CloudWatch log query. :param date_range: A tuple representing the start and end datetime for the query. :type date_range: tuple :return: A list containing the query results. :rtype: list """ client = boto3.client("logs") try: try: start_time = round( self.date_utilities.convert_iso8601_to_unix_timestamp(date_range[0]) ) end_time = round( self.date_utilities.convert_iso8601_to_unix_timestamp(date_range[1]) ) response = client.start_query( logGroupName=self.log_groups, startTime=start_time, endTime=end_time, queryString="fields @timestamp, @message | sort @timestamp asc", limit=self.limit, ) query_id = response["queryId"] except client.exceptions.ResourceNotFoundException as e: raise DateOutOfBoundsError(f"Resource not found: {e}") while True: time.sleep(1) results = client.get_query_results(queryId=query_id) if results["status"] in [ "Complete", "Failed", "Cancelled", "Timeout", "Unknown", ]: return results.get("results", []) except DateOutOfBoundsError: return [] def _initiate_query(self, client, date_range, max_logs): """ Initiates the CloudWatch logs query. :param date_range: A tuple representing the start and end datetime for the query. :type date_range: tuple :param max_logs: The maximum number of logs to retrieve. :type max_logs: int :return: The query ID as a string. :rtype: str """ try: start_time = round( self.date_utilities.convert_iso8601_to_unix_timestamp(date_range[0]) ) end_time = round( self.date_utilities.convert_iso8601_to_unix_timestamp(date_range[1]) ) response = client.start_query( logGroupName=self.log_groups, startTime=start_time, endTime=end_time, queryString="fields @timestamp, @message | sort @timestamp asc", limit=max_logs, ) return response["queryId"] except client.exceptions.ResourceNotFoundException as e: raise DateOutOfBoundsError(f"Resource not found: {e}") def _wait_for_query_results(self, client, query_id): """ Waits for the query to complete and retrieves the results. :param query_id: The ID of the initiated query. :type query_id: str :return: A list containing the results of the query. :rtype: list """ while True: time.sleep(1) results = client.get_query_results(queryId=query_id) if results["status"] in [ "Complete", "Failed", "Cancelled", "Timeout", "Unknown", ]: return results.get("results", [])