By multi-target exporter pattern we refer to a specific design, in which:

This pattern is only used for certain exporters, such as the blackbox and the SNMP exporter.

The reason is that we either can’t run an exporter on the targets, e.g. network gear speaking SNMP, or that we are explicitly interested in the distance, e.g. latency and reachability of a website from a specific point outside of our network, a common use case for the blackbox exporter.

Multi-target exporters are flexible regarding their environment and can be run in many ways. As regular programs, in containers, as background services, on baremetal, on virtual machines. Because they are queried and do query over network they do need appropriate open ports. Otherwise they are frugal.

Now let’s try it out for yourself!

Use Docker to start a blackbox exporter container by running this in a terminal. Depending on your system configuration you might need to prepend the command with a sudo:

You should see a few log lines and if everything went well the last one should report msg="Listening on address" as seen here:

There are two ways of querying:

You can manually try the first query type with curl in another terminal or use this link:

The response should be something like this:

Those are metrics in the Prometheus format. They come from the exporter’s instrumentation and tell us about the state of the exporter itself while it is running. This is called whitebox monitoring and very useful in daily ops practice. If you are curious, try out our guide on how to instrument your own applications.

For the second type of querying we need to provide a target and module as parameters in the HTTP GET Request. The target is a URI or IP and the module must defined in the exporter’s configuration. The blackbox exporter container comes with a meaningful default configuration.
We will use the target prometheus.io and the predefined module http_2xx. It tells the exporter to make a GET request like a browser would if you go to prometheus.io and to expect a 200 OK response.

You can now tell your blackbox exporter to query prometheus.io in the terminal with curl:

This will return a lot of metrics:

Notice that almost all metrics have a value of 0. The last one reads probe_success 0. This means the prober could not successfully reach prometheus.io. The reason is hidden in the metric probe_ip_protocol with the value 6. By default the prober uses IPv6 until told otherwise. But the Docker daemon blocks IPv6 until told otherwise. Hence our blackbox exporter running in a Docker container can’t connect via IPv6.

We could now either tell Docker to allow IPv6 or the blackbox exporter to use IPv4. In the real world both can make sense and as so often the answer to the question “what is to be done?” is “it depends”. Because this is an exporter guide we will change the exporter and take the opportunity to configure a custom module.

The modules are predefined in a file inside the docker container called config.yml which is a copy of blackbox.yml in the github repo.

We will copy this file, adapt it to our own needs and tell the exporter to use our config file instead of the one included in the container.

First download the file using curl or your browser:

Open it in an editor. The first few lines look like this:

YAML uses whitespace indentation to express hierarchy, so you can recognise that two modules named http_2xx and http_post_2xx are defined, and that they both have a prober http and for one the method value is specifically set to POST.
You will now change the module http_2xx by setting the preferred_ip_protocol of the prober http explicitly to the string ip4.

If you want to know more about the available probers and options check out the documentation.

Now we need to tell the blackbox exporter to use our freshly changed file. You can do that with the flag --config.file="blackbox.yml". But because we are using Docker, we first must make this file available inside the container using the --mount command.

First you stop the old container by changing into its terminal and press ctrl+c. Make sure you are in the directory containing your blackbox.yml. Then you run this command. It is long, but we will explain it:

With this command, you told docker to:

If everything is correct, you should see something like this:

Now you can try our new IPv4-using module http_2xx in a terminal:

Which should return Prometheus metrics like this:

You can see that the probe was successful and get many useful metrics, like latency by phase, status code, ssl status or certificate expiry in Unix time.
The blackbox exporter also offers a tiny web interface at localhost:9115 for you to check out the last few probes, the loaded config and debug information. It even offers a direct link to probe prometheus.io. Handy if you are wondering why something does not work.

So far, so good. Congratulate yourself. The blackbox exporter works and you can manually tell it to query a remote target. You are almost there. Now you need to tell Prometheus to do the queries for us.

Below you find a minimal prometheus config. It is telling Prometheus to scrape the exporter itself as we did before using curl 'localhost:9115/metrics':

prometheus.yml for Linux:

prometheus.yml for macOS and Windows:

Now run a Prometheus container and tell it to mount our config file from above. Because of the way networking on the host is addressable from the container you need to use a slightly different command on Linux than on MacOS and Windows.:

Run Prometheus on Linux (don’t use --network="host" in production):

Run Prometheus on MacOS and Windows:

This command works similarly to running the blackbox exporter using a config file.

If everything worked, you should be able to go to localhost:9090/targets and see under blackbox an endpoint with the state UP in green. If you get a red DOWN make sure that the blackbox exporter you started above is still running. If you see nothing or a yellow UNKNOWN you are really fast and need to give it a few more seconds before reloading your browser’s tab.

To tell Prometheus to query "localhost:9115/probe?target=prometheus.io&module=http_2xx" you add another scrape job blackbox-http where you set the metrics_path to /probe and the parameters under params: in the Prometheus config file prometheus.yml:

After saving the config file switch to the terminal with your Prometheus docker container and stop it by pressing ctrl+C and start it again to reload the configuration by using the existing command.

The terminal should return the message "Server is ready to receive web requests." and after a few seconds you should start to see colourful graphs in your Prometheus.

This works, but it has a few disadvantages:

To fix this, we will use relabeling. Relabeling is useful here because behind the scenes many things in Prometheus are configured with internal labels. The details are complicated and out of scope for this guide. Hence we will limit ourselves to the necessary. But if you want to know more check out this talk. For now it suffices if you understand this:

Here is the config you will use to do that. Don’t worry if this is a bit much at once, we will go through it step by step:

So what is new compared to the last config?

params does not include target anymore. Instead we add the actual targets under static configs: targets. We also use several because we can do that now:

relabel_configs contains the new relabeling rules:

Before applying the relabeling rules, the URI of a request Prometheus would make would look like this: "http://prometheus.io/probe?module=http_2xx". After relabeling it will look like this "http://localhost:9115/probe?target=http://prometheus.io&module=http_2xx".

Now let us explore how each rule does that:

First we take the values from the label __address__ (which contain the values from targets) and write them to a new label __param_target which will add a parameter target to the Prometheus scrape requests:

After this our imagined Prometheus request URI has now a target parameter: "http://prometheus.io/probe?target=http://prometheus.io&module=http_2xx".

Then we take the values from the label __param_target and create a label instance with the values.

Our request will not change, but the metrics that come back from our request will now bear a label instance="http://prometheus.io".

After that we write the value localhost:9115 (the URI of our exporter) to the label __address__. This will be used as the hostname and port for the Prometheus scrape requests. So that it queries the exporter and not the target URI directly.

Our request is now "localhost:9115/probe?target=http://prometheus.io&module=http_2xx". This way we can have the actual targets there, get them as instance label values while letting Prometheus make a request against the blackbox exporter.

Often people combine these with a specific service discovery. Check out the configuration documentation for more information. Using them is no problem, as these write into the __address__ label just like targets defined under static_configs.

That is it. Restart the Prometheus docker container and look at your metrics. Pay attention that you selected the period of time when the metrics were actually collected.