Prometheus
Prometheus Operator differences

Differences between the Prometheus Operator and the Collector

Summary

RequirementPrometheus OperatorCollector
DeploymentStatefulSets managed by Prometheus OperatorSidecar, Deployment, or DaemonSet
AlertingAlertManagerConfig, AlertManager, PrometheusRuleAlerts and Monitors
High availability / Long-term storageProvided by ThanosNatively supported
Service discoveryProbe, PodMonitor, ServiceMonitorsAnnotations, ServiceMonitors, Prometheus Discovery

Alerting

Chronosphere supports Prometheus alerts, but not AlertManagerConfig (opens in a new tab) or AlertManager custom resource definitions (CRDs) (opens in a new tab). Chronosphere alerting (called monitors) has concepts that don't apply to Prometheus alerting rules, has its own concepts and models, and doesn't support complex routing trees. For more information, refer to the monitors documentation.

Due to Chronosphere being a single data store, you can merge alerts, so an alert queries all metrics and not only metrics local to a Prometheus instance.

You manage alerting configuration separate of any cluster or Collector configuration with the Chronosphere app, Chronoctl, or Terraform. This approach brings more flexibility for managing configuration and means you can spread configuration responsibility between teams.

Scaling

Thanos support

Chronosphere is a scalable backend for Prometheus and doesn't require Thanos (opens in a new tab) or the ThanosRuler CRD (opens in a new tab).

Sharding across instances

The Prometheus Operator supports automatically sharding ServiceMonitors across multiple Prometheus instances. However, you still need to setup a remote write destination such as Thanos, or a single large instance.

The Collector handles scale by using a DaemonSet and scoping each instance of the Collector to a particular node. Using a DaemonSet is the recommended way to deploy the Collector, but there are other methods available you can read about in the Collector documentation.

There are advantages and disadvantages to deploying the Collector as a DaemonSet:

  • Advantages:
    • Using a DaemonSet means you don't need large or powerful instances to run the Collector.
    • The DaemonSet implementation reduces any impact of a single collector instance experiencing issues.
  • Disadvantages:
    • All instances created with a DaemonSet must have uniform resources.

Configure Prometheus

Prometheus Operator has a Prometheus CRD (opens in a new tab) for configuring global settings on the instances it creates. Chronosphere supports many of these settings, but instead you set them in Collector configuration. For more information, visit the configuration documentation.

Some Prometheus Operator settings that don't transfer to the context of the Collector, such as volumeMounts and priorityClass.

The Collector doesn't support the following fields from the ServiceMonitor CRD:

  • targetLabels
  • podTargetLabels

Instead, use Prometheus relabel_config (opens in a new tab) which allows advanced modifications to any target and its labels before ingesting the metrics.

Service discovery

Static scrape targets

To replicate the capabilities of the Prometheus Operator Probe CRD (opens in a new tab), Chronosphere recommends running a single instance of the Collector as a sidecar (if possible), or a one instance Deployment. If you run the Collector as a DaemonSet, all instances of the Collector attempt to scrape the same targets, resulting in multiple copies of the same metrics. For more information about available options, refer to the Collector documentation.