Skip to content

Observability#

This document describes the observability stack used at Open Food Facts to monitor applications.

Having a good observability stack is critical to spend less time when debugging failures, to have a comprehension of how applications behave over time, and to have the ability to compare a software version with the previously deployed one.

Munin#

See Munin a tool to monitor servers and services.

Influx / Prometheus / Grafana#

See openfoodfacts-monitoring project on github

The observability stack used in the OFF stack is comprised of the following applications:

  • Filebeat as a logs collection agent deployed on each QEMU VM with Docker containers.
  • ElasticSearch for centralized storage and indexing of logs collected from Docker.
  • Prometheus for scraping metrics from Prometheus exporters' /metrics endpoint, running as sidecar containers of the applications.
  • AlertManager to send alerts based on Prometheus metrics, integrated with dedicated Slack channels.
  • InfluxDB is the storage backend for data harvested by prometheus
  • Grafana for visualizing Prometheus metrics, InfluxDB and other metrics; and create dashboards. (official doc)
  • Prometheus exporters such as the Apache Prometheus Exporter, which collect metrics from applications and expose them on a port in the Prometheus metric format. Some applications natively export Prometheus metrics and do not need additional exporters.

The observability stack diagram is as follows:

Observability stack

Prometheus#

Some interesting pages of prometheus include:

  • https://prometheus.openfoodfacts.org/alerts to show currently triggered alerts
  • https://prometheus.openfoodfacts.org/targets where you can check the status of the targets (services observed by prometheus)
  • https://prometheus.openfoodfacts.org/rules for alerting rules

multi-target exporter pattern#

We use the multi-target exporter pattern on some configurations.

This pattern helps writting a lot of targets that follow same rules in the same way.

Some points that are not easy to understand:

  • targets will be in __address__ at the begining of the process
  • the last rule overwrite __address__ to put some static value, but it's ok as previous rules where already processed
  • simple params like instance, app are parameters for the job configuration, not for the target url
  • to add params to target url we use __param_xxxx (for exemple __param_target to add a target parameter to url)
  • instance is very important, as each job must have their separate instance name.

You can look at https://prometheus.openfoodfacts.org/targets to see if your targets are processed correctly. A mouse over in the "labels" column shows you parameter before processing.

Blackbox exporter#

Is a service that can be used by Prometheus to probe for websites.

Prometheus will call the service as if it was a metric exporter with the appropriate target (that you set through target on __param_target in the configuration (if you use multi-target exporter pattern)

Exposing metrics behind a proxy#

Prometheus server is in OVH datacenter. To expose metrics from Free datacenter through the internet, we use the nginx reverse proxy.

See free-exporters site configuration, and scraper configuration.

Testing it on monitoring container#

Blackbox is on port 9105 so you can test it, for example using: 

# http probe
curl "http://localhost:9115/probe?module=http_probe&target=https%3A%2F%2Fsearch.openfoodfacts.org%2F"
# icmp probe
curl "http://localhost:9115/probe?module=icmp&app=ovh1&target=ov1.openfoodfacts.org"