使用 CloudWatch Logs 執行大型查詢

這是進入點。

// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
import { CloudWatchLogsClient } from "@aws-sdk/client-cloudwatch-logs";
import { CloudWatchQuery } from "./cloud-watch-query.js";

console.log("Starting a recursive query...");

if (!process.env.QUERY_START_DATE || !process.env.QUERY_END_DATE) {
  throw new Error(
    "QUERY_START_DATE and QUERY_END_DATE environment variables are required.",
  );
}

const cloudWatchQuery = new CloudWatchQuery(new CloudWatchLogsClient({}), {
  logGroupNames: ["/workflows/cloudwatch-logs/large-query"],
  dateRange: [
    new Date(Number.parseInt(process.env.QUERY_START_DATE)),
    new Date(Number.parseInt(process.env.QUERY_END_DATE)),
  ],
});

await cloudWatchQuery.run();

console.log(
  `Queries finished in ${cloudWatchQuery.secondsElapsed} seconds.\nTotal logs found: ${cloudWatchQuery.results.length}`,
);

這是類別，可視需要將查詢分割為多個步驟。

// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
import {
  StartQueryCommand,
  GetQueryResultsCommand,
} from "@aws-sdk/client-cloudwatch-logs";
import { splitDateRange } from "@aws-doc-sdk-examples/lib/utils/util-date.js";
import { retry } from "@aws-doc-sdk-examples/lib/utils/util-timers.js";

class DateOutOfBoundsError extends Error {}

export class CloudWatchQuery {
  /**
   * Run a query for all CloudWatch Logs within a certain date range.
   * CloudWatch logs return a max of 10,000 results. This class
   * performs a binary search across all of the logs in the provided
   * date range if a query returns the maximum number of results.
   *
   * @param {import('@aws-sdk/client-cloudwatch-logs').CloudWatchLogsClient} client
   * @param {{ logGroupNames: string[], dateRange: [Date, Date], queryConfig: { limit: number } }} config
   */
  constructor(client, { logGroupNames, dateRange, queryConfig }) {
    this.client = client;
    /**
     * All log groups are queried.
     */
    this.logGroupNames = logGroupNames;

    /**
     * The inclusive date range that is queried.
     */
    this.dateRange = dateRange;

    /**
     * CloudWatch Logs never returns more than 10,000 logs.
     */
    this.limit = queryConfig?.limit ?? 10000;

    /**
     * @type {import("@aws-sdk/client-cloudwatch-logs").ResultField[][]}
     */
    this.results = [];
  }

  /**
   * Run the query.
   */
  async run() {
    this.secondsElapsed = 0;
    const start = new Date();
    this.results = await this._largeQuery(this.dateRange);
    const end = new Date();
    this.secondsElapsed = (end - start) / 1000;
    return this.results;
  }

  /**
   * Recursively query for logs.
   * @param {[Date, Date]} dateRange
   * @returns {Promise<import("@aws-sdk/client-cloudwatch-logs").ResultField[][]>}
   */
  async _largeQuery(dateRange) {
    const logs = await this._query(dateRange, this.limit);

    console.log(
      `Query date range: ${dateRange
        .map((d) => d.toISOString())
        .join(" to ")}. Found ${logs.length} logs.`,
    );

    if (logs.length < this.limit) {
      return logs;
    }

    const lastLogDate = this._getLastLogDate(logs);
    const offsetLastLogDate = new Date(lastLogDate);
    offsetLastLogDate.setMilliseconds(lastLogDate.getMilliseconds() + 1);
    const subDateRange = [offsetLastLogDate, dateRange[1]];
    const [r1, r2] = splitDateRange(subDateRange);
    const results = await Promise.all([
      this._largeQuery(r1),
      this._largeQuery(r2),
    ]);
    return [logs, ...results].flat();
  }

  /**
   * Find the most recent log in a list of logs.
   * @param {import("@aws-sdk/client-cloudwatch-logs").ResultField[][]} logs
   */
  _getLastLogDate(logs) {
    const timestamps = logs
      .map(
        (log) =>
          log.find((fieldMeta) => fieldMeta.field === "@timestamp")?.value,
      )
      .filter((t) => !!t)
      .map((t) => `${t}Z`)
      .sort();

    if (!timestamps.length) {
      throw new Error("No timestamp found in logs.");
    }

    return new Date(timestamps[timestamps.length - 1]);
  }

  /**
   * Simple wrapper for the GetQueryResultsCommand.
   * @param {string} queryId
   */
  _getQueryResults(queryId) {
    return this.client.send(new GetQueryResultsCommand({ queryId }));
  }

  /**
   * Starts a query and waits for it to complete.
   * @param {[Date, Date]} dateRange
   * @param {number} maxLogs
   */
  async _query(dateRange, maxLogs) {
    try {
      const { queryId } = await this._startQuery(dateRange, maxLogs);
      const { results } = await this._waitUntilQueryDone(queryId);
      return results ?? [];
    } catch (err) {
      /**
       * This error is thrown when StartQuery returns an error indicating
       * that the query's start or end date occur before the log group was
       * created.
       */
      if (err instanceof DateOutOfBoundsError) {
        return [];
      }
      throw err;
    }
  }

  /**
   * Wrapper for the StartQueryCommand. Uses a static query string
   * for consistency.
   * @param {[Date, Date]} dateRange
   * @param {number} maxLogs
   * @returns {Promise<{ queryId: string }>}
   */
  async _startQuery([startDate, endDate], maxLogs = 10000) {
    try {
      return await this.client.send(
        new StartQueryCommand({
          logGroupNames: this.logGroupNames,
          queryString: "fields @timestamp, @message | sort @timestamp asc",
          startTime: startDate.valueOf(),
          endTime: endDate.valueOf(),
          limit: maxLogs,
        }),
      );
    } catch (err) {
      /** @type {string} */
      const message = err.message;
      if (message.startsWith("Query's end date and time")) {
        // This error indicates that the query's start or end date occur
        // before the log group was created.
        throw new DateOutOfBoundsError(message);
      }

      throw err;
    }
  }

  /**
   * Call GetQueryResultsCommand until the query is done.
   * @param {string} queryId
   */
  _waitUntilQueryDone(queryId) {
    const getResults = async () => {
      const results = await this._getQueryResults(queryId);
      const queryDone = [
        "Complete",
        "Failed",
        "Cancelled",
        "Timeout",
        "Unknown",
      ].includes(results.status);

      return { queryDone, results };
    };

    return retry(
      { intervalInMs: 1000, maxRetries: 60, quiet: true },
      async () => {
        const { queryDone, results } = await getResults();
        if (!queryDone) {
          throw new Error("Query not done.");
        }

        return results;
      },
    );
  }
}

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", [])