Handy Kubernetes sidecar for Prometheus 2.x. It watches for configmaps in all namespaces that have the annotation specified with the -annotation flag. When it finds one it pulls each value out and assumes it's a prometheus rule. It then validates them and adds them to a the rules file specified with -rulespath. If that file changes it hits the reload endpoint specified by the -endpoint flag.

• New 5.0 version, no breaking changes. Some work to reduce churn, added an event recorder, configmaps will get events if keys are successfully loaded or not. Brought up to spec with the current kubernetes sample controller. If it finds two RuleGroup(s) with the same name it will automatically salt the name of the duplicate in the final configmap.
• New 3.0 version, breaking changes! This works with the new prometheus 2 rules only, if you want to use the rule loader with prometheus 1.x please use the prom1.x-stable tag.

• -annotation - Used to customize the annotation label you'd like the rule loader to look like on your configmaps.
• -rulespath - The location you would like your rules to be written to. Should correspond to a rule_files path in your prometheus config.
• -endpoint - Endpoint to make a bodyless POST request to (Prometheus uses /-/reload)
• -batchtime - Configure how long you want it to sleep between reload attempts in seconds, if your configmaps churn a lot it can cause excessive reloads on prometheus.
• -kubeconfig - Use a kubeconfig to configure the connection to the api server, off cluster use only.
• master - Address of the master server, overrides server in kubeconfig. For off cluster use only.

for example:

./PrometheusRuleLoader -rulespath ./rules.rules -annotation 'prometheus.io/v2/rules' -endpoint http://127.0.0.1:9090/-/reload

Assuming a launch with the above commandline here is an example of a configmap.

kind: ConfigMap
apiVersion: v1
metadata:
  name: 'test-rules'
  namespace: "kube-system"
  labels:
    app: "prometheus"
  annotations: {
    "prometheus.io/v2/rules": "true"
  }
data:
  mygroupname: |
    - record: job:http_inprogress_requests:sum
      expr: sum(http_inprogress_requests) by (job)
    - alert: HighErrorRate
      expr: job:request_latency_seconds:mean5m{job="myjob"} > 0.5
      for: 10m
      labels:
        severity: page
      annotations:
        summary: High request latency

The value of the configmap that contains rules can either be in the format of []Rules, RuleGroup, or RuleGroups as detailed in github.com/prometheus/prometheus/pkg/rulefmt. If the values are in the []Rules format a group will be created around them and named configmapnamespace-configmapname-key.

Once all the appropriate configmaps are processed all the groups will be assembled into a single rule file named -rulespath.

The PrometheusRuleLoaders docker container should be deployed in the same pod as prometheus. They should both share a volume mount (and emptydir works fine here). PrometheusRuleLoader will use this shared space to write it's rule file to, meanwhile Prometheus should be configured to look for it's rule file at this path.