log_db_daemon - Man Page

access_log directive

The path to the access log file is used to provide the database connection parameters.

  access_log daemon:/mysql_host:port/database/table/username/password squid

The 'daemon' prefix is mandatory and tells squid that the logfile_daemon helper is to be used instead of the normal file logging.

The last parameter tells squid which log format to use when writing lines to the log daemon. Presently squid format is supported.

mysql_host:port

Host where the mysql server is running. If left empty, 'localhost' is assumed.

database

Name of the database to connect to. If left empty, 'squid_log' is assumed.

table

Name of the database table where log lines are stored. If left empty, 'access_log' is assumed.

username

Username to use when connecting to the database. If left empty, 'squid' is assumed.

password

Password to use when connecting to the database. If left empty, no password is used.

To leave all fields to their default values, you can use a single slash:

  access_log daemon:/ squid

To specify only the database password, which by default is empty, you must leave unspecified all the other parameters by using null strings:

  access_log daemon://///password squid

logfile_daemon directive

This is the current way of telling squid where the logfile daemon resides.

  logfile_daemon /path/to/squid/libexec/logfile-daemon_mysql.pl

The script must be copied to the location specified in the directive.

Let's call the database 'squid_log' and the log table 'access_log'. The username and password for the db connection will be both 'squid'.

Database

Create the database:

  CREATE DATABASE squid_log;

User

Create the user:

  GRANT INSERT,SELECT,CREATE ON squid_log.* TO 'squid'@'localhost' IDENTIFIED BY 'squid';
  FLUSH PRIVILEGES;

Note that only CREATE, INSERT and SELECT privileges are granted to the 'squid' user. This ensures that the logfile daemon script cannot change or modify the log entries.

Table

The Daemon will attempt to initialize this table if none exists when it starts.

The table created should look like:

  CREATE TABLE access_log (
    id                   INTEGER NOT NULL AUTO_INCREMENT PRIMARY KEY,
    time_since_epoch     DECIMAL(15,3),
    time_response        INTEGER,
    ip_client            CHAR(15),
    ip_server            CHAR(15),
    http_status_code     VARCHAR(10),
    http_reply_size      INTEGER,
    http_method          VARCHAR(20),
    http_url             TEXT,
    http_username        VARCHAR(20),
    http_mime_type       VARCHAR(50),
    squid_hier_status    VARCHAR(20),
    squid_request_status VARCHAR(20)
  );

Clients accessing the cache

  SELECT DISTINCT ip_client FROM access_log;

Number of request per day

  SELECT
    DATE(FROM_UNIXTIME(time_since_epoch)) AS date_day,
    COUNT(*) AS num_of_requests
  FROM access_log
  GROUP BY 1
  ORDER BY 1;

Request status count

To obtain the raw count of each request status:

  SELECT squid_request_status, COUNT(*) AS n
  FROM access_log
  GROUP BY squid_request_status
  ORDER BY 2 DESC;

To calculate the percentage of each request status:

  SELECT
    squid_request_status,
    (COUNT(*)/(SELECT COUNT(*) FROM access_log)*100) AS percentage
  FROM access_log
  GROUP BY squid_request_status
  ORDER BY 2 DESC;

To distinguish only between HITs and MISSes:

  SELECT
    'hits',
    (SELECT COUNT(*)
    FROM access_log
    WHERE squid_request_status LIKE '%HIT%')
    /
    (SELECT COUNT(*) FROM access_log)*100
    AS percentage
  UNION
  SELECT
    'misses',
    (SELECT COUNT(*)
    FROM access_log
    WHERE squid_request_status LIKE '%MISS%')
    /
    (SELECT COUNT(*) FROM access_log)*100
    AS percentage;

Response time ranges

  SELECT
    '0..500',
    COUNT(*)/(SELECT COUNT(*) FROM access_log)*100 AS percentage
  FROM access_log
  WHERE time_response >= 0 AND time_response < 500
  UNION
  SELECT
    '500..1000',
    COUNT(*)/(SELECT COUNT(*) FROM access_log)*100 AS percentage
  FROM access_log
  WHERE time_response >= 500 AND time_response < 1000
  UNION
  SELECT
    '1000..2000',
    COUNT(*)/(SELECT COUNT(*) FROM access_log)*100 AS percentage
  FROM access_log
  WHERE time_response >= 1000 AND time_response < 2000
  UNION
  SELECT
    '>= 2000',
    COUNT(*)/(SELECT COUNT(*) FROM access_log)*100 AS percentage
  FROM access_log
  WHERE time_response >= 2000;

Traffic by mime type

  SELECT
    http_mime_type,
    SUM(http_reply_size) as total_bytes
  FROM access_log
  GROUP BY http_mime_type
  ORDER BY 2 DESC;

Traffic by client

  SELECT
    ip_client,
    SUM(http_reply_size) AS total_bytes
  FROM access_log
  GROUP BY 1
  ORDER BY 2 DESC;

The MyISAM storage engine is known to be faster than the InnoDB one, so although it doesn't support transactions and referential integrity, it might be more appropriate in this scenario. You might want to append "ENGINE=MYISAM" at the end of the table creation code in the above SQL script.

Indexes should be created according to the queries that are more frequently run. The DDL script only creates an implicit index for the primary key column.

This script currently implements only the L (i.e. "append a line to the log") command, therefore the log lines are never purged from the table. This approach has an obvious scalability problem.

One solution would be to implement e.g. the "rotate log" command in a way that would calculate some summary values, put them in a "summary table" and then delete the lines used to calculate those values.

Similar cleanup code could be implemented in an external script and run periodically independently from squid log commands.

This script has only been tested in low-volume scenarios (single client, less than 10 req/s). Tests in high volume environments could reveal performance bottlenecks and bugs.