Psstat (a Telegraf exec)

Psstat gathers resource usage information of specified processes. The output of psstat can be fed into Telegraf as influx format. psstat is an alternative to the build in procstat of Telegraf itself. The main differences are: (1.) only a few fields are outputted, greatly reducing the data pushed to Influxdb; (2.) resource usage of child processes is calculated as well giving you a better insight in the resource usage of a process.

psstat was build for speed and a low impact on the host machine. So data is gathered from /proc only. As written above, only a few fields are gathered and shown in the output. These fields are CPU usage (percentage), memory usage (percentage) and number of processes. More detailed explanation of the output can be found below at measurements.

A cache to calculate the difference between measurements is kept in a directory of your choice, however, I advise to make this a tmpfs disk to keep it fast. See cache on tmpfs for instructions.

If you run multiple exec's for psstat you should give each exec its own cache-name to prevent data corruption in the cache. If this becomes corrupt the output will be empty most likely.

Processes to monitor can be specified in different ways:

• By PID number
• By PID file
• By pattern (matched in /proc/<id>/stat)
• By Systemd Unit pattern

The first run of the psstat won't return any result as that run is required to build the first cache. Consecutive runs will output data.

Example usage in Telegraf

[[inputs.exec]]
  commands = ["/usr/bin/psstat --tag env=production --pid-file /var/run/nginx.pid --pattern mysql:mysqld_safe"]
  data_format = "influx"

The above configuration would result in output like:

> psstat,env=production,process_name=nginx p_cpu=0.42,p_mem=0.001,n_proc=5i 1562471764000000000
> psstat,env=production,process_name=mysql p_cpu=3.14,p_mem=1.592,n_proc=65i 1562471764000000000

Cache on tmpfs

To keep the cache fast you can store it on a tmpfs disk. By default psstat tries to write cache data to /tmp which should be a tmpfs already.

If you want a dedicated directory for the cache of psstat (for example in a multi tenant environment) you can do the following:

Assumption: Telegraf is used which runs as the user telegraf. If not, adjust the user and group in the commands below.

Add the following line to /etc/fstab:

tmpfs /mnt/psstat tmpfs auto,rw,mode=755,size=5m,uid=telegraf,gid=telegraf 0 2

This creates a disk of 5 MB to store caches.

Then you need to create the directory specified above and mount all devices:

mkdir /mnt/psstat
chown telegraf: /mnt/psstat
mount -a

Measurements

CPU usage is measured as a percentage where 100% is written as 1.000. If a value greater than 1.000 is shown, it means more than 1 core is used for that process (and its children).

• psstat.p_cpu value=3.141

Memory usage is measured as a percentage between 0.000 and 1.000.

• psstat.p_mem value=0.042

Number of processes, integer. This is the count of the process and its children.

• psstat.n_proc value=3

Package building

To build a package, the tool fpm is used, please follow the installation guide to be able to build a package.

The following OSes are supported:

• Debian, Ubuntu: make deb
• CentOS, RHEL: make rpm