Log Rotation and Retention

Logging is a nearly indispensable part of any networked service, but especially with high traffic installations care needs to be taken to ensure that log files don’t exhaust storage space and cause maintenance or outage nightmares.

Traffic Server provides a two-pronged solution: log rotation (also called log rolling) to keep individual logs as manageable in size as possible for easier ingestion and analysis by humans and other programs, and log retention to keep logs from using more space than available and necessary.

This section covers both features.

Rotation Options

Traffic Server provides automatic log file rolling. At specific intervals during the day or when log files reach a certain size, Traffic Server closes its current set of log files and opens new log files. Depending on the amount of traffic your servers are exposed to, you may find that increasing the frequency of log rolling is beneficial, or even necessary, to maintain manageable log file sets. Traffic Server nodes processing moderately high levels of traffic may want to start by rolling logs every six hours, and adjusting from there.

Log file rolling offers the following benefits:

• It defines a consistent interval over which log analysis can be performed.
• It keeps any single log file from becoming too large and helps to keep the logging system within the specified space limits.
• It provides an easy way to identify files that are no longer being used so that an automated script can clean the logging directory and run log analysis programs.

Rolled Log Filename Format

Traffic Server provides a consistent naming scheme for rolled log files that enables you to easily identify log files. When Traffic Server rolls a log file, it saves and closes the old file before it starts a new file. Traffic Server renames the old file to include the following information:

• The original log file’s name (such as access.log).
• The hostname of the Traffic Server node that generated the log file (useful in Traffic Server log collation configurations).
• Two timestamps separated by a hyphen (-). The first timestamp is a lower bound for the timestamp of the first record in the log file. The lower bound is the time when the new buffer for log records is created. Under low load, the first timestamp in the filename can be different from the timestamp of the first entry. Under normal load, the first timestamp in the filename and the timestamp of the first entry are similar. The second timestamp is an upper bound for the timestamp of the last record in the log file (this is normally the rolling time).
• The suffix .old, which makes it easy for automated scripts to find rolled log files.

Timestamps have the following format:

%Y%M%D.%Hh%Mm%Ss-%Y%M%D.%Hh%Mm%Ss

The following table describes the format:

The following is an example of a rolled log filename:

access.log.mymachine.20110912.12h00m00s-20000913.12h00m00s.old

The logging system buffers log records before writing them to disk. When a log file is rolled, the log buffer might be partially full. If it is, then the first entry in the new log file will have a timestamp earlier than the time of rolling. When the new log file is rolled, its first timestamp will be a lower bound for the timestamp of the first entry.

For example, suppose logs are rolled every three hours, and the first rolled log file is:

access.log.mymachine.20110912.12h00m00s-19980912.03h00m00s.old

If the lower bound for the first entry in the log buffer at 3:00:00 is 2:59:47, then the next log file will have the following timestamp when rolled:

access.log.mymachine.20110912.02h59m47s-19980912.06h00m00s.old

The contents of a log file are always between the two timestamps. Log files do not contain overlapping entries, even if successive timestamps appear to overlap.

Rolling Intervals

Log files are rolled at specific intervals relative to a given hour of the day. Three options may be used to control when log files are rolled:

• A file size threshold, which will prevent any individual log from growing too large.
• The offset hour, which is an hour between 0 (midnight) and 23.
• The rolling interval.

Both the offset hour and the rolling interval determine when log file rolling starts. Rolling occurs every rolling interval and at the offset hour. For example, if the rolling interval is six hours and the offset hour is 0 (midnight), then the logs will roll at midnight (00:00), 06:00, 12:00, and 18:00 each day. If the rolling interval is 12 hours and the offset hour is 3, then logs will roll at 03:00 and 15:00 each day.

To set log file rolling options and/or configure Traffic Server to roll log files when they reach a certain size, adjust the following settings in records.config:

1. Enable log rolling with proxy.config.log.rolling_enabled.

   CONFIG proxy.config.log.rolling_enabled INT 1

2. Configure the upper limit on log file size with proxy.config.log.rolling_size_mb.

   CONFIG proxy.config.log.rolling_size_mb INT 1024

3. Set the offset hour with proxy.config.log.rolling_offset_hr.

   CONFIG proxy.config.log.rolling_offset_hr INT 0

4. Set the interval (in seconds) with proxy.config.log.rolling_interval_sec.

   CONFIG proxy.config.log.rolling_interval_sec INT 21600

5. Run the command traffic_ctl config reload to apply the configuration changes.

You can fine-tune log file rolling settings for individual log files in the log.* specification in logging.yaml. The custom log file uses the rolling settings provided in the relevant log function call, which override the default settings you specify in Traffic Manager or records.config described above.

Retention Options

Traffic Server enables you to control the amount of disk space that the logging directory can consume. This allows the system to operate smoothly within a specified space window for a long period of time. After you establish a space limit, Traffic Server continues to monitor the space in the logging directory. When the free space dwindles to the headroom limit, it enters a low space state and takes the following actions:

• If the autodelete option is enabled, then Traffic Server identifies previously-rolled log files (log files with the .old extension). It starts deleting files one by one, beginning with the oldest file, until it emerges from the low state. Traffic Server logs a record of all deleted files in the system error log.
• If the autodelete option is disabled or there are not enough old log files to delete for the system to emerge from its low space state, then Traffic Server issues a warning and continues logging until the allocated log space is exhausted (which, if configured appropriately, will be well before your actual filesystem space is fully consumed and causes additional problems). At this point, event logging stops even though proxy traffic is still served without client-visible interruption. Traffic Server resumes event logging when enough space becomes available for it to exit the low space state. To make space available, either explicitly increase the logging space limit or remove files from the logging directory manually.

You can run a cron(8) script in conjunction with Traffic Server to automatically remove old log files from the logging directory before Traffic Server enters the low space state. Relocate the old log files to a temporary partition, where you can run a variety of log analysis scripts. Following analysis, either compress the logs and move to an archive location, or simply delete them.

Traffic Server periodically checks the amount of log space used against both the log space allocation configured by proxy.config.log.max_space_mb_for_logs and the actual amount of space available on the disk partition. The used log space is calculated by summing the size of all files present in the logging directory and is published in the proxy.process.log.log_files_space_used metric. The proxy.config.log.max_space_mb_headroom configuration variable specifies an amount of headroom that is subtracted from the log space allocation. This can be tuned to reduce the risk of completely filling the disk partition.

Setting Log File Management Options

To set log management options, follow the steps below:

1. In the records.config file, edit the following variables
   • proxy.config.log.max_space_mb_for_logs
   • proxy.config.log.max_space_mb_headroom
2. Run the command traffic_ctl config reload to apply the configuration changes.

Retaining Logs For No More Than a Specified Period

If for security reasons logs need to be purged to make sure no log entry remains on the box for more then a specified period of time, we could achieve this by setting the rolling interval, the maximum number of rolled log files, and forcing Traffic Server to roll even when there is no traffic.

Let us say we wanted the oldest log entry to be kept on the box to be no older than 2-hour old.

Set proxy.config.output.logfile.rolling_interval_sec (yaml: rolling_interval_sec) to 3600 (1h) which will lead to rolling every 1h.

Set proxy.config.output.logfile.rolling_max_count (yaml: rolling_max_count) to 1 which will lead to keeping only one rolled log file at any moment (rolled will be trimmed on every roll).

Set proxy.config.output.logfile.rolling_allow_empty (yaml: rolling_allow_empty) to 1 (default: 0) which will allow logs to be open and rolled even if there was nothing to be logged during the previous period (i.e. no requests to Traffic Server).

The above will ensure logs are rolled every 1h hour, only 1 rolled log file to be kept (rest will be trimmed/removed) and logs will be rolling (“moving”) even if nothing is logged (i.e. no traffic to Traffic Server).