The Prometheus monitoring system and time series database.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 

203 lines
5.5 KiB

rule_files:
- rules.yml
evaluation_interval: 1m
tests:
# Basic tests for promql_expr_test, not dependent on rules.
- interval: 1m
input_series:
- series: test_full
values: "0 0"
- series: test_repeat
values: "1x2"
- series: test_increase
values: "1+1x2"
- series: test_histogram
values: "{{schema:1 sum:-0.3 count:32.1 z_bucket:7.1 z_bucket_w:0.05 buckets:[5.1 10 7] offset:-3 n_buckets:[4.1 5] n_offset:-5}}"
- series: test_histogram_repeat
values: "{{sum:3 count:2 buckets:[2]}}x2"
- series: test_histogram_increase
values: "{{sum:3 count:2 buckets:[2]}}+{{sum:1.3 count:1 buckets:[1]}}x2"
- series: test_stale
values: "0 stale"
- series: test_missing
values: "0 _ _ _ _ _ _ 0"
promql_expr_test:
# Ensure the sample is evaluated at the time we expect it to be.
- expr: timestamp(test_full)
eval_time: 0m
exp_samples:
- value: 0
- expr: timestamp(test_full)
eval_time: 1m
exp_samples:
- value: 60
- expr: timestamp(test_full)
eval_time: 2m
exp_samples:
- value: 60
# Repeat & increase
- expr: test_repeat
eval_time: 2m
exp_samples:
- value: 1
labels: "test_repeat"
- expr: test_increase
eval_time: 2m
exp_samples:
- value: 3
labels: "test_increase"
# Histograms
- expr: test_histogram
eval_time: 1m
exp_samples:
- labels: "test_histogram"
histogram: "{{schema:1 sum:-0.3 count:32.1 z_bucket:7.1 z_bucket_w:0.05 buckets:[5.1 10 7] offset:-3 n_buckets:[4.1 5] n_offset:-5}}"
- expr: test_histogram_repeat
eval_time: 2m
exp_samples:
- labels: "test_histogram_repeat"
histogram: "{{count:2 sum:3 counter_reset_hint:not_reset buckets:[2]}}"
- expr: test_histogram_increase
eval_time: 2m
exp_samples:
- labels: "test_histogram_increase"
histogram: "{{count:4 sum:5.6 counter_reset_hint:not_reset buckets:[4]}}"
# Ensure a value is stale as soon as it is marked as such.
- expr: test_stale
eval_time: 59s
exp_samples:
- value: 0
labels: "test_stale"
- expr: test_stale
eval_time: 1m
exp_samples: []
# Ensure lookback delta is respected, when a value is missing.
- expr: timestamp(test_missing)
eval_time: 4m59s
exp_samples:
- value: 0
- expr: timestamp(test_missing)
eval_time: 5m
exp_samples: []
# Minimal test case to check edge case of a single sample.
- input_series:
- series: test
values: 1
promql_expr_test:
- expr: test
eval_time: 0
exp_samples:
- value: 1
labels: test
# Test recording rules run even if input_series isn't provided.
- promql_expr_test:
- expr: count_over_time(fixed_data[1h])
eval_time: 1h
exp_samples:
- value: 60
- expr: timestamp(fixed_data)
eval_time: 1h
exp_samples:
- value: 3600
# Tests for alerting rules.
- interval: 1m
input_series:
- series: 'up{job="prometheus", instance="localhost:9090"}'
values: "0+0x1440"
promql_expr_test:
- expr: count(ALERTS) by (alertname, alertstate)
eval_time: 4m
exp_samples:
- labels: '{alertname="AlwaysFiring",alertstate="firing"}'
value: 1
- labels: '{alertname="InstanceDown",alertstate="pending"}'
value: 1
alert_rule_test:
- eval_time: 1d
alertname: AlwaysFiring
exp_alerts:
- {}
- eval_time: 1d
alertname: InstanceDown
exp_alerts:
- exp_labels:
severity: page
instance: localhost:9090
job: prometheus
exp_annotations:
summary: "Instance localhost:9090 down"
description: "localhost:9090 of job prometheus has been down for more than 5 minutes."
- eval_time: 0
alertname: AlwaysFiring
exp_alerts:
- {}
- eval_time: 0
alertname: InstanceDown
exp_alerts: []
# Tests for interval vs evaluation_interval.
- interval: 1s
input_series:
- series: 'test{job="test", instance="x:0"}'
# 2 minutes + 1 second of input data, recording rules should only run
# once a minute.
values: "0+1x120"
promql_expr_test:
- expr: job:test:count_over_time1m
eval_time: 0m
exp_samples:
- value: 1
labels: 'job:test:count_over_time1m{job="test"}'
- expr: timestamp(job:test:count_over_time1m)
eval_time: 10s
exp_samples:
- value: 0
labels: '{job="test"}'
- expr: job:test:count_over_time1m
eval_time: 1m
exp_samples:
- value: 60
labels: 'job:test:count_over_time1m{job="test"}'
- expr: timestamp(job:test:count_over_time1m)
eval_time: 1m10s
exp_samples:
- value: 60
labels: '{job="test"}'
- expr: job:test:count_over_time1m
eval_time: 2m
exp_samples:
- value: 60
labels: 'job:test:count_over_time1m{job="test"}'
- expr: timestamp(job:test:count_over_time1m)
eval_time: 2m59s999ms
exp_samples:
- value: 120
labels: '{job="test"}'