Dockerfile for building netdata
Shell Makefile
Latest commit e2f623c Jan 8, 2017 @titpetric committed on GitHub Merge pull request #33 from Djamy/master
Added support for Telegram & Pushbullet alerts with env vars

README.md

netdata

Dockerfile for building and running a netdata deamon for your host instance.

Netdata monitors your server with thoughts of performance and memory usage, providing detailed insight into very recent server metrics. It's nice, and now it's also dockerized.

More info about project: https://github.com/firehol/netdata

Using

docker run -d --cap-add SYS_PTRACE \
           -v /proc:/host/proc:ro \
           -v /sys:/host/sys:ro \
           -p 19999:19999 titpetric/netdata

Open a browser on http://server:19999/ and watch how your server is doing.

Getting emails on alarms

Netdata supports forwarding alarms to an email address. You can set up sSMTP by setting the following ENV variables:

  • SSMTP_TO - This is the address alarms will be delivered to.
  • SSMTP_SERVER - This is your SMTP server. Defaults to smtp.gmail.com.
  • SSMTP_PORT - This is the SMTP server port. Defaults to 587.
  • SSMTP_USER - This is your username for the SMTP server.
  • SSMTP_PASS - This is your password for the SMTP server. Use an app password if using Gmail.
  • SSMTP_TLS - Use TLS for the connection. Defaults to YES.
  • SSMTP_HOSTNAME - The hostname mail will come from. Defaults to localhost.

For example, using gmail:

-e SSMTP_TO=user@gmail.com -e SSMTP_USER=user -e SSMTP_PASS=password

Alternatively, if you already have s sSMTP config, you can use that config with:

-v /path/to/config:/etc/ssmtp

See the following link for details on setting up sSMTP: SSMTP - ArchWiki

Getting alarms in slack

Netdata supports sending alerts to slack via webhooks. You can set that up by setting the following ENV variables:

  • SLACK_WEBHOOK_URL - This is your incoming slack webhook
  • SLACK_CHANNEL - This is the default channel that alerts will get sent to

For example:

-e SLACK_WEBHOOK_URL=https://hooks.slack.com/services/XXXX -e SLACK_CHANNEL=alerts

Getting alarms in Telegram

Netdata supports sending alerts to Telegram via token and chat ID. You can set that up by setting the following ENV variables: For more details about Telegram alerts, see this page - GitHub

  • TELEGRAM_BOT_TOKEN - This is your bot token
  • TELEGRAM_CHAT_ID - This is the chat ID

For example:

-e TELEGRAM_BOT_TOKEN=22624413:AAGy12TkSMBYVBTe4lQt3BfUYvUs5h7I1jn -e TELEGRAM_CHAT_ID=137165138

Getting alarms in Pushbullet

Netdata supports sending alerts to Pushbullet via API token. You can set that up by setting the following ENV variables: More details about Pushbullet alerts are provided here - GitHub

  • PUSHBULLET_ACCESS_TOKEN - This is your API token
  • PUSHBULLET_DEFAULT_EMAIL - This is the default email that alerts will get sent to if there is not a Pushbullet account attached to it

For example:

-e PUSHBULLET_ACCESS_TOKEN=o.l8VuizWhXgbERf2Q78ghtzb1LDCYvbSD -e PUSHBULLET_DEFAULT_EMAIL=your.email@gmail.com

Monitoring docker container metrics

Netdata supports fetching container data from docker.sock. You can forward it to the netdata container with:

-v /var/run/docker.sock:/var/run/docker.sock

This will allow netdata to resolve container names.

Note: forwarding docker.sock exposes the administrative docker API. If due to some security issue access has been obtained to the container, it will expose full docker API, allowing to stop, create or delete containers, as well as download new images in the host.

TL;DR If you care about security, consider forwarding a secure docker socket with docker-proxy-acl

Monitoring docker notes on some systems (Debian jessie)

On debian jessie only 'cpu' and 'disk' metrics show up under individual docker containers. To get the memory metric, you will have to add cgroup_enable=memory swapaccount=1 to /etc/default/grub, appending the GRUB_CMDLINE_LINUX_DEFAULT variable:

$ cat /etc/default/grub  | grep GRUB_CMDLINE_LINUX_DEFAULT
GRUB_CMDLINE_LINUX_DEFAULT="quiet cgroup_enable=memory swapaccount=1"

After rebooting your linux instance, the memory accounting subsystem of the kernel will be enabled. Netdata will pick up additional metrics for the containers when it starts.

Environment variables

It's possible to pass a NETDATA_PORT environment variable with -e, to start up netdata on a different port.

docker run -e NETDATA_PORT=80 [...]

Some explanation is in order

Docker needs to run with the SYS_PTRACE capability. Without it, the mapped host/proc filesystem is not fully readable to the netdata deamon, more specifically the "apps" plugin:

16-01-12 07:58:16: ERROR: apps.plugin: Cannot process /host/proc/1/io (errno 13, Permission denied)

See the following link for more details: /proc/1/environ is unavailable in a container that is not priviledged

Limitations

In addition to the above requirements and limitations, monitoring the complete network interface list of the host is not possible from within the Docker container. If you're running netdata and want to graph all the interfaces available on the host, you will have to use --net=host mode.

See the following link for more details: network interfaces missing when mounting proc inside a container

Work-around

I provided a script called fakenet.sh which provides a copy of the /proc/net filesystem. You should start this script before you start the netdata container. You can do it like this:

wget https://raw.githubusercontent.com/titpetric/netdata/master/fakenet.sh
chmod a+x fakenet.sh
nohup ./fakenet.sh >/dev/null 2>&1 &

Using the above command, the fakenet script will start in the background and will keep running there. You can use other tools like screen or tmux to provide similar capability.

The script fills out the /dev/shm/fakenet location, which you must mount into the container. You must mount it into /fakenet/proc/net exactly with the option like this:

-v /dev/shm/fakenet:/fakenet/proc/net

The script refreshes network information about every 250ms (four times per second). The interval may be increased to give better accuracy of netdata, but CPU usage will also increase. Because of this, the data is not very accurate and some spikes and valleys will occur because of a shifting window between when the reading was taken (fakeproc) and between when the reading was read by netdata. This means the margin for error is whatever data can be collected in ~250ms.

While the solution might not fit everybody, it's security-positive because the netdata container can only inspect the fake proc/net location, and can't actually access any of the networks because it runs on a private LAN / custom network which is managed and firewalled by docker. You may even open access via application, like a nginx reverse proxy where you can add authentication etc.

Pro/con list:

  • + network isolation stays in tact
  • + all network device metrics are available
  • - one more service to provide fakenet
  • - accuracy vs. cpu use is a trade-off

Additional notes

Netdata provides monitoring via a plugin architecture. This plugin supports many projects that don't provide data over the /proc filesystem. When you're running netdata in the container, you will have difficulty providing many of these paths to the netdata container.

What you do get (even with the docker version) is:

  • Host CPU statististics
  • Host Network I/O, QoS
  • Host Disk I/O
  • Applications monitoring
  • Container surface metrics (cpu/disk per name)

You will not get detailed application metrics (mysql, etc.) from other containers or from the host if running netdata in a container. It may be possible to get some of those metrics, but it might not be easy, and most likely not worth it. For most detailed metrics, netdata needs to share the same environment as the application server it monitors. This means it would need to run either in the same container (not even remotely practical), or in the same virtual machine (no containers).

What I can tell you is that it's very stable, and snappy. Godspeed!