Datadog offers native Docker container monitoring, either by running the Agent
on the host or running in a sidecar container. Which is the best way to run it?
It ultimately depends on the tooling you have in place to manage the Agent's
configuration. If you want to go Docker all the way, you can run the Agent as
a sidecar and control its configuration with custom Dockerfiles
.
Let's see what it looks like.
To build a meaningful setup, we start from the example that Docker put together to illustrate Compose. A simple Python web application that connects to Redis to store the number of hits.
Here is the docker-compose.yml
that powers the whole setup.
version: "4"
services:
web:
build: web
command: ddtrace-run python app.py
ports:
- "5000:5000"
volumes:
- ./web:/code # modified here to take into account the new app path
links:
- redis
environment:
- DATADOG_HOST=datadog # used by the web app to initialize the Datadog library
- DD_AGENT_HOST=dd-agent # points to dd-agent to send traces
redis:
image: redis
# agent section
datadog:
container_name: dd-agent
build: datadog
links:
- redis # ensures that redis is a host that the container can find
- web # ensures that the web app can send metrics
environment:
- DD_API_KEY=__your_datadog_api_key_here__
- DD_DOGSTATSD_NON_LOCAL_TRAFFIC=true # enables agent to receive custom metrics from other containers
- DD_APM_ENABLED=true # enables tracing
- DD_APM_NON_LOCAL_TRAFFIC=true # enables agent to receive traces from other containers
- DD_AGENT_HOST=dd-agent # allows web container to forward traces to agent
- DD_SITE=datadoghq.com # determines datadog instance to send data to (e.g change to datadoghq.eu for EU1)
volumes:
- /var/run/docker.sock:/var/run/docker.sock
- /proc/:/host/proc/:ro
- /sys/fs/cgroup:/host/sys/fs/cgroup:ro
Because the Agent needs to monitor Redis it needs:
- The proper
redisdb.yaml
in the container's/etc/datadog-agent/conf.d
to find the Redis node.
The Agent's Dockerfile
takes care of #1.
FROM datadog/agent:latest
ADD conf.d/redisdb.yaml /etc/datadog-agent/conf.d/redisdb.yaml
- And the Compose yaml files creates the link to Redis with:
links:
- redis
The Agent needs the following in order to collect Python traces from the web
container:
- The
ddtrace
library.
This can be achieved by adding the following to the requirements.txt
for our web
app. The datadog
library is used to collect our custom (dogstatsd
) metrics
flask
redis
datadog
ddtrace <--
- Tags that identify the app and its origin.
This can be achieved by simply adding a few environment variables to the web
container's Dockerfile
.
FROM python:2.7
ADD . /code
WORKDIR /code
RUN pip install -r requirements.txt
# This is where you set DD tags
ENV DD_SERVICE web <-- This sets the "service" name in Datadog
ENV DD_ENV sandbox <-- This sets the "env" name in Datadog
ENV DD_VERSION 1.0 <-- This sets the "version" number in Datadog
- Environment variables & container name set to ensure connection between your app container and the Agent.
You will want to make sure that the DD_AGENT_HOST
environment variables set in your docker-compose.yml
are the same for both the Agent and your app container.
Also set the container_name
for the datadog
container to be the same as this DD_AGENT_HOST
value.
How to test this?
- Install Docker Compose
- Clone this repository
- Update your
DD_API_KEY
indocker-compose.yml
- Run all containers with
docker-compose up
- Refresh
localhost:5000
a few times in your web browser to generate some traces. - Verify in Datadog that your container picks up the Docker & Redis metrics along with all the APM traces.
- An easy way to verify you are receiving these metrics is by visiting the
Metrics Summary
page. Navigate toMetrics -> Summary
, then search for eitherdocker
orredis
to see if any metrics appear. See example below.
- An easy way to verify you are receiving APM traces is by visiting the
Trace Explorer
page. Navigate toAPM -> Traces -> Explorer
and you should see a list of traces. If you click on a trace, your flame graph should look like this.