SYNOPSIS

vnstatd [ -Ddnpsv? ] [ --config file ] [ --daemon ] [ --debug ] [ -g group ] [ --group group ] [ --help ] [ --noadd ] [ --nodaemon ] [ --pidfile file ] [ --sync ] [ --u user ] [ --user user ] [ --version ]

DESCRIPTION

The purpose of vnstatd is to provide a more flexible and robust way for updating vnstat(1) databases than what using cron for updating can provide. The daemon makes possible updating databases more often but at the same time causes less disk access since data can be cached and written only later to disk at a user configurable interval. The availability of each interface is automatically tracked which removes the need for additional scripts to be implemented and called when an interface comes online or goes offline.

vnstatd is the command for starting the daemon. The daemon can either fork itself to run as a background process or stay attached to the terminal. It supports logging to a user selectable file or using syslog.

Once started, the daemon will read vnstat.conf(5) if available and then check if there are any databases available in the database directory that has been specified in the configuration file. New databases will be created for all available interfaces excluding pseudo interfaces lo, lo0 and sit0 if no databases are found during startup. This behaviour can be disabled if needed. The daemon will then proceed to track the availability of monitored interfaces, process the interface traffic statistics and write new values to databases at a configured interval. As a result, the daemon ends up spending most of the time sleeping between updates.

OPTIONS

--config file

Use file as configuration file instead of using normal configuration file search functionality.

-d, --daemon

Fork process to background and run as a daemon.

-D, --debug

Provide additional output for debug purposes. The process will stay attached to the terminal for output.

-g, --group group

Set daemon process group to group during startup. group can be either the name of the group or a numerical group id. This option can only be used when the process is started as root.

--noadd

Disable automatic creation of new databases for all available interfaces if the daemon is started with zero database found. Pseudo interfaces lo, lo0 and sit0 are excluded from getting added.

-n, --nodaemon

Stay in foreground attached to the current terminal and start the update process.

-p, --pidfile file

Write the process id to file and use it for locking so that another instance of the daemon cannot be started if the same file is specified.

-s, --sync

Synchronize internal counters in the database with interface counters for all available interfaces before starting traffic monitoring. Use this option if the traffic between the previous shutdown and the current startup of the daemon needs to be ignored. This option isn't required in normal use because the daemon will automatically synchronize the internal counters after a system reboot, if enought time has passed since the daemon was previously running or if the internal counters are clearly out of sync.

-u, --user user

Set daemon process user to user during startup. user can be either the login of the user or a numerical user id. This option can only be used when the process is started as root.

-v, --version

Show current version of the daemon executable.

-?, --help

Show a command option summary.

CONFIGURATION

The behaviour of the daemon is configured mainly using the configuration keywords UpdateInterval, PollInterval and SaveInterval in the configuration file.

UpdateInterval defines in seconds how often the interface data is fetched and updated. This is similar to the run interval for alternative cron based updating. However, the difference is that the data doesn't directly get written to disk during updates.

PollInterval defines in seconds how often the list of available interfaces is checked for possible changes. The minimum value is 2 seconds and the maximum 60 seconds. PollInterval also defines the resolution for other intervals.

SaveInterval defines in minutes how often cached interface data is written to disk. A write can only occur during the updating of interface data. Therefore, the value should be a multiple of UpdateInterval with a maximum value of 60 minutes.

The default values of UpdateInterval 30, SaveInterval 5 and PollInterval 5 are usually suitable for most systems and provide a similar behaviour as cron based updating does but with a better resolution for interface changes and fast interfaces.

For embedded and/or low power systems more tuned configurations are possible. In such cases if the interfaces are mostly static the PollInterval can be increased to around 10-30 seconds and UpdateInterval set to 60 seconds. Higher values up to 300 seconds are possible if the interface speed is 10 Mbit or less. SaveInterval can be increased for example to 15, 30 or even 60 minutes depending on how often the data needs to be viewed.

SIGNALS

The daemon is listening to signals SIGHUP, SIGINT and SIGTERM. Sending the SIGHUP signal to the daemon will cause cached data to be written to disk, a rescan of the database directory and a reload of settings from the configuration file. However, the pid file location will not be changed even if it's configuration setting has been modified.

SIGTERM and SIGINT signals will cause the daemon to write all cached data to disk and then exit.

FILES

/var/lib/vnstat/

Default database directory. Files are named according to the monitored interfaces.

/etc/vnstat.conf

Config file that will be used unless $HOME/.vnstatrc exists. See the configuration chapter and vnstat.conf(5) for more information.

/var/log/vnstat.log

Log file that will be used if logging to file is enable and no other file is specified in the config file.

/var/run/vnstat.pid

File used for storing the process id if no other file is specified in the configuration file or using the command line parameter.

RESTRICTIONS

Updates needs to be executed at least as often as it is possible for the interface to generate enough traffic to overflow the kernel interface traffic counter. Otherwise, it is possible that some traffic won't be seen. This isn't an issue for 64 bit kernels but at least one update every hour is always required in order to provide proper input. With 32 bit kernels, the maximum time between two updates depends on how fast the interface can transfer 4 GiB. Calculated theoretical times are:

10 Mbit: 54 minutes
100 Mbit: 5 minutes
1000 Mbit: 30 seconds

However, for 1000 Mbit interfaces updating once every minute is usually a useable solution if faster updates can't be used.

Virtual and aliased interfaces cannot be monitored because the kernel doesn't provide traffic information for that type of interfaces. Such interfaces are usually named eth0:0, eth0:1, eth0:2 etc. where eth0 is the actual interface being aliased.

AUTHOR

Teemu Toivola <tst at iki dot fi>

RELATED TO vnstatd…