Prometheus Monitoring

Each Rook Ceph cluster has some built in metrics collectors/exporters for monitoring with Prometheus.

If you do not have Prometheus running, follow the steps below to enable monitoring of Rook. If your cluster already contains a Prometheus instance, it will automatically discover Rook's scrape endpoint using the standard prometheus.io/scrape and prometheus.io/port annotations.

Prometheus Operator¶

First the Prometheus operator needs to be started in the cluster so it can watch for our requests to start monitoring Rook and respond by deploying the correct Prometheus pods and configuration. A full explanation can be found in the Prometheus operator repository on GitHub, but the quick instructions can be found here:

1

kubectl apply -f https://raw.githubusercontent.com/coreos/prometheus-operator/v0.40.0/bundle.yaml

This will start the Prometheus operator, but before moving on, wait until the operator is in the Running state:

1

kubectl get pod

Once the Prometheus operator is in the Running state, proceed to the next section to create a Prometheus instance.

Prometheus Instances¶

With the Prometheus operator running, we can create service monitors that will watch the Rook cluster.

There are two sources for metrics collection:

• Prometheus manager module: It is responsible for exposing all metrics other than ceph daemons performance counters.
• Ceph exporter: It is responsible for exposing only ceph daemons performance counters as prometheus metrics.

From the root of your locally cloned Rook repo, go the monitoring directory:

1
2

$ git clone --single-branch --branch v1.11.11 https://github.com/rook/rook.git
cd rook/deploy/examples/monitoring

Create the service monitor as well as the Prometheus server pod and service:

1
2
3
4

kubectl create -f service-monitor.yaml
kubectl create -f exporter-service-monitor.yaml
kubectl create -f prometheus.yaml
kubectl create -f prometheus-service.yaml

Ensure that the Prometheus server pod gets created and advances to the Running state before moving on:

1

kubectl -n rook-ceph get pod prometheus-rook-prometheus-0

Prometheus Web Console¶

Once the Prometheus server is running, you can open a web browser and go to the URL that is output from this command:

1

echo "http://$(kubectl -n rook-ceph -o jsonpath={.status.hostIP} get pod prometheus-rook-prometheus-0):30900"

You should now see the Prometheus monitoring website.

Click on Graph in the top navigation bar.

In the dropdown that says insert metric at cursor, select any metric you would like to see, for example ceph_cluster_total_used_bytes

Click on the Execute button.

Below the Execute button, ensure the Graph tab is selected and you should now see a graph of your chosen metric over time.

Prometheus Consoles¶

You can find Prometheus Consoles for and from Ceph here: GitHub ceph/cephmetrics - dashboards/current directory.

A guide to how you can write your own Prometheus consoles can be found on the official Prometheus site here: Prometheus.io Documentation - Console Templates.

Prometheus Alerts¶

To enable the Ceph Prometheus alerts via the helm charts, set the following properties in values.yaml:

• rook-ceph chart: monitoring.enabled: true
• rook-ceph-cluster chart: monitoring.enabled: true monitoring.createPrometheusRules: true

Alternatively, to enable the Ceph Prometheus alerts with example manifests follow these steps:

1. Create the RBAC and prometheus rules:

   1 2	kubectl create -f deploy/examples/monitoring/rbac.yaml kubectl create -f deploy/examples/monitoring/localrules.yaml

2. Make following changes to your CephCluster object (e.g., cluster.yaml).

   1 2 3 4 5 6 7 8 9 10 11

   apiVersion: ceph.rook.io/v1 kind: CephCluster metadata: name: rook-ceph namespace: rook-ceph [...] spec: [...] monitoring: enabled: true [...]

3. Deploy or update the CephCluster object.

   1	kubectl apply -f cluster.yaml

Customize Alerts¶

The Prometheus alerts can be customized with a post-processor using tools such as Kustomize. For example, first extract the helm chart:

1

helm template -f values.yaml rook-release/rook-ceph-cluster > cluster-chart.yaml

Now create the desired customization configuration files. This simple example will show how to update the severity of a rule, add a label to a rule, and change the for time value.

Create a file named kustomization.yaml:

1
2
3
4
5
6
7
8
9

patches:
- path: modifications.yaml
  target:
    group: monitoring.coreos.com
    kind: PrometheusRule
    name: prometheus-ceph-rules
    version: v1
resources:
- cluster-chart.yaml

Create a file named modifications.yaml

1
2
3
4
5
6
7
8

- op: add
  path: /spec/groups/0/rules/0/labels
  value:
    my-label: foo
    severity: none
- op: add
  path: /spec/groups/0/rules/0/for
  value: 15m

Finally, run kustomize to update the desired prometheus rules:

1
2

kustomize build . > updated-chart.yaml
kubectl create -f updated-chart.yaml

Grafana Dashboards¶

The dashboards have been created by @galexrt. For feedback on the dashboards please reach out to him on the Rook.io Slack.

The following Grafana dashboards are available:

• Ceph - Cluster
• Ceph - OSD (Single)
• Ceph - Pools

Updates and Upgrades¶

When updating Rook, there may be updates to RBAC for monitoring. It is easy to apply the changes with each update or upgrade. This should be done at the same time you update Rook common resources like common.yaml.

1

kubectl apply -f deploy/examples/monitoring/rbac.yaml

Teardown¶

To clean up all the artifacts created by the monitoring walk-through, copy/paste the entire block below (note that errors about resources "not found" can be ignored):

1
2
3
4

kubectl delete -f service-monitor.yaml
kubectl delete -f prometheus.yaml
kubectl delete -f prometheus-service.yaml
kubectl delete -f https://raw.githubusercontent.com/coreos/prometheus-operator/v0.40.0/bundle.yaml

Then the rest of the instructions in the Prometheus Operator docs can be followed to finish cleaning up.

Special Cases¶

Tectonic Bare Metal¶

Tectonic strongly discourages the tectonic-system Prometheus instance to be used outside their intentions, so you need to create a new Prometheus Operator yourself. After this you only need to create the service monitor as stated above.

CSI Liveness¶

To integrate CSI liveness and grpc into ceph monitoring we will need to deploy a service and service monitor.

1

kubectl create -f csi-metrics-service-monitor.yaml

This will create the service monitor to have prometheus monitor CSI

Collecting RBD per-image IO statistics¶

RBD per-image IO statistics collection is disabled by default. This can be enabled by setting enableRBDStats: true in the CephBlockPool spec. Prometheus does not need to be restarted after enabling it.

Using custom label selectors in Prometheus¶

If Prometheus needs to select specific resources, we can do so by injecting labels into these objects and using it as label selector.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12

apiVersion: ceph.rook.io/v1
kind: CephCluster
metadata:
  name: rook-ceph
  namespace: rook-ceph
  [...]
spec:
  [...]
  labels:
    monitoring:
      prometheus: k8s
  [...]

Horizontal Pod Scaling using Kubernetes Event-driven Autoscaling (KEDA)¶

Using metrics exported from the Prometheus service, the horizontal pod scaling can use the custom metrics other than CPU and memory consumption. It can be done with help of Prometheus Scaler provided by the KEDA. See the KEDA deployment guide for details.

Following is an example to autoscale RGW:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
 name: rgw-scale
 namespace: rook-ceph
spec:
 scaleTargetRef:
   kind: Deployment
   name: rook-ceph-rgw-my-store-a # deployment for the autoscaling
 minReplicaCount: 1
 maxReplicaCount: 5
 triggers:
 - type: prometheus
   metadata:
     serverAddress: http://rook-prometheus.rook-ceph.svc:9090
     metricName: collecting_ceph_rgw_put
     query: |
       sum(rate(ceph_rgw_put[2m])) # prometheus query used for autoscaling
     threshold: "90"