Reference: PrometheusRule CRD¶
A PrometheusRule holds Prometheus alerting and recording rules. Alerting rules
evaluate a PromQL expression and fire an alert while the expression returns
results. Recording rules evaluate an expression and store the result as a new
time series, which makes expensive queries cheap to reuse in dashboards and
alerts. Rules are organized in named groups, and each group is evaluated on its
own interval.
- API version:
monitoring.coreos.com/v1 - Kind:
PrometheusRule - Scope: Namespaced
- Short name:
promrule
This page documents Prometheus Operator v0.95.1.
Example¶
The following example defines one recording rule and two alerting rules for the
your-application service. The first alert fires when no instance has been
reachable for five minutes. The second alert fires when more than 5% of HTTP
requests have failed over the last ten minutes. It assumes the application
exposes an http_requests_total counter with a code label.
apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
name: your-application
namespace: your-application-namespace
spec:
groups:
- name: your-application.rules
interval: 1m
rules:
- record: job:http_requests:rate5m
expr: sum by (job) (rate(http_requests_total{job="your-application"}[5m]))
- alert: YourApplicationDown
expr: sum(up{job="your-application"}) == 0
for: 5m
labels:
severity: critical
annotations:
summary: "your-application is down"
description: "No instance of your-application has been scraped successfully for 5 minutes."
- alert: YourApplicationHighErrorRate
expr: |
sum(rate(http_requests_total{job="your-application", code=~"5.."}[10m]))
/
sum(rate(http_requests_total{job="your-application"}[10m]))
> 0.05
for: 10m
labels:
severity: warning
annotations:
summary: "your-application has a high error rate"
description: "{{ $value | humanizePercentage }} of requests failed over the last 10 minutes."
Specification (spec)¶
| Field | Type | Description |
|---|---|---|
groups |
[]object |
Optional. The rule groups in this resource. See spec.groups[]. |
spec.groups[]¶
| Field | Type | Description |
|---|---|---|
name |
string |
Required. The name of the rule group. Must not be empty. |
rules |
[]object |
Optional. The alerting and recording rules in the group. See spec.groups[].rules[]. |
interval |
string |
Optional. How often the rules in the group are evaluated, as a duration such as 1m. Defaults to the global evaluation interval. |
labels |
map[string]string |
Optional. Labels to add or overwrite on the results of every rule in the group. Labels set on a rule take precedence. Requires Prometheus 3.0 or later. |
limit |
integer |
Optional. The maximum number of alerts an alerting rule, or series a recording rule, can produce. |
query_offset |
string |
Optional. Shifts the evaluation timestamp of the group into the past by this duration. Requires Prometheus 2.53 or later. |
partial_response_strategy |
string |
Optional. Only used by Thanos Ruler and ignored by Prometheus. Allowed values: abort, warn. |
spec.groups[].rules[]¶
Each rule is either a recording rule or an alerting rule. Set exactly one of
record and alert.
| Field | Type | Description |
|---|---|---|
expr |
int-or-string |
Required. The PromQL expression to evaluate. |
record |
string |
Optional. For recording rules, the name of the time series to write. Must be a valid metric name. |
alert |
string |
Optional. For alerting rules, the name of the alert. Must be a valid label value. |
for |
string |
Optional. For alerting rules, how long the expression must return results before the alert fires, as a duration such as 5m. Until then the alert is pending. |
keep_firing_for |
string |
Optional. For alerting rules, how long the alert keeps firing after the expression stops returning results. Must not be empty if set. |
labels |
map[string]string |
Optional. Labels to add or overwrite on the result. For alerts, labels such as severity are commonly used for routing. |
annotations |
map[string]string |
Optional. For alerting rules, annotations to add to each alert, such as summary and description. Values can use Prometheus template variables such as {{ $value }} and {{ $labels.<name> }}. |
Durations use the Prometheus format, a sequence of number and unit pairs such as
1h30m. Valid units are ms, s, m, h, d, w and y.
Status (status)¶
The Prometheus Operator writes status only when it runs with the
StatusForConfigurationResources feature gate enabled, which is off by
default upstream. The field can be empty.
| Field | Type | Description |
|---|---|---|
bindings[].resource |
string |
The type of workload that selected this resource, for example prometheuses or thanosrulers. |
bindings[].name |
string |
The name of the workload that selected this resource. |
bindings[].namespace |
string |
The namespace of that workload. |
bindings[].conditions[].type |
string |
The condition type. Only Accepted is supported. |
bindings[].conditions[].status |
string |
The condition status, for example True or False. |
bindings[].conditions[].reason |
string |
A short reason for the last transition. |
bindings[].conditions[].message |
string |
A human-readable explanation, for example why the operator rejected the resource. |
bindings[].conditions[].observedGeneration |
integer |
The metadata.generation the condition is based on. If it is lower than the current generation, the condition is out of date. |
Platform Notes¶
- You can create
PrometheusRuleresources in your own namespaces. - Rules can only use metrics that are already collected. To collect metrics,
see
ServiceMonitor,PodMonitorand the getting started guide.