Skip to content

Latest commit

 

History

History
154 lines (122 loc) · 6.07 KB

USAGE.md

File metadata and controls

154 lines (122 loc) · 6.07 KB

Usage

Configuring

Within the rating-operator deployment is included a base configuration, a RatingRule.

RatingRules

$ kubectl get ratingrules.rating.smile.fr
NAME                          AGE
rating-default-rules          46m

$ kubectl describe ratingrules.rating.smile.fr rating-rating-default-rules
[...]
Spec:
# The "Metrics" part describe the location of the metric in the metering-operator
# This part is about to be removed, as the metering-operator is now deprecated.
  Metrics:
    request_cpu:
      presto_column:  pod_request_cpu_core_seconds
      presto_table:   report_metering_pod_rating_cpu_request_hourly
      report_name:    pod-rating-cpu-request-hourly
      Unit:           core-seconds
    request_memory:
      presto_column:  pod_request_memory_byte_seconds
      presto_table:   report_metering_pod_rating_memory_request_hourly
      report_name:    pod-rating-memory-request-hourly
      Unit:           byte-seconds
    usage_cpu:
      presto_column:  pod_usage_cpu_core_seconds
      presto_table:   report_metering_pod_rating_cpu_usage_hourly
      report_name:    pod-rating-cpu-usage-hourly
      Unit:           core-seconds
    usage_memory:
      presto_column:  pod_usage_memory_byte_seconds
      presto_table:   report_metering_pod_rating_memory_usage_hourly
      report_name:    pod-rating-memory-usage-hourly
      Unit:           byte-seconds


# The "Rules" part holds the value with which to multiply metrics
  Rules:
    Name:  rules_default
    Ruleset:
      Metric:  request_cpu
      Value:   0.5
      Unit:    core-hours
      Metric:  usage_cpu
      Value:   0.8
      Unit:    core-hours
      Metric:  request_memory
      Value:   0.4
      Unit:    GiB-hours
      Metric:  usage_memory
      Value:   0.8
      Unit:    GiB-hours
[...]

This configuration will be used until a new one is created. You can edit the above with:

$ kubectl edit ratingrules.rating.smile.fr <yourconfig>

For more advanced use cases, we strongly advice to create a new configuration. The rating-operator does not allow modification on the base configuration, and it is provided to help the application run as default.

DISCLAIMER The part below describe features that are only compatible with the metering-operator based rating, which is deprecated. The RatingRules also provide assistance through metrics declarations and configuration versionning for the metering-operator version. To understand why, read the configuration versionning part of this document document.

More information on RatingRules can be found in the custom resources documentation.

RatedRules

The other part of the configuration is composed of RatingRuleInstances, describing metrics using promQL and of a base RatingRule that holds values to be used in queries.

$ kubectl get ratingruleinstances.rating.smile.fr
NAME                                                           AGE
rating-rule-instance-pod-request-cpu                              2m
rating-rule-instance-pod-request-memory                           2m
rating-rule-instance-pod-usage-cpu                                2m
rating-rule-instance-pod-usage-memory                             2m

$ kubectl get ratingruleinstances.rating.smile.fr rating-rule-instance-pod-usage-cpu -o yaml
kind: RatingRuleInstance
metadata:
  name: rating-rule-instance-pod-usage-cpu
spec:
  metric: usage_cpu * on() group_right sum(rate(container_cpu_usage_seconds_total[1m]))
    BY (pod, namespace) + on (pod, namespace) group_left(node) (sum(kube_pod_info{pod_ip!="",node!="",host_ip!=""})
    by (pod, namespace, node) * 0)
  name: pods_usage_cpu
  timeframe: 60s

More informations on RatingRuleInstances can be found in the custom resources documentation.

Adding new metrics

To rate a new metric, you need to create a RatingRuleInstance:

$ kubectl get ratingRuleInstances.rating.smile.fr                             
NAME                            AGE
rating-rule-instance-rated-usage-cpu   21s

$ kubectl describe ratingRuleInstances.rating.smile.fr rating-rule-instance-rated-usage-cpu
Name:         rating-rule-instance-rated-usage-cpu
Namespace:    rating
Labels:       <none>
Annotations:  <none>
API Version:  rating.smile.fr/v1
Kind:         RatingRuleinstance
[...]
Spec:
  # Metric have to be a valid prometheus query
  # In this example, we are using the value of "usage_memory" described in the RatingRules above to rate our frames
  Metric: -|
    (sum(rate(container_memory_usage_bytes[1h])) BY (pod, namespace) + on (pod, namespace) group_left(node) (sum(kube_pod_info{pod_ip!="",node!="",host_ip!=""}) by (pod, namespace, node) * 0)) * on () group_left() usage_memory

  # This is the name of the metric once rated
  Name:       test_metric

  # Only second is supported
  Timeframe:  10s
[...]

As soon as the RatingRuleInstances is validated, the rating-operator-engine will start processing frames. More information on RatingRuleInstances can be found in the custom resources documentation.

Templates and instances historization

In Rating operator, we keep a history of the templates and instances modification in the Postgres database.

  1. Templates historization:
id template_name template_group template_var template_query
2021-10-26 10:28:27.266831 name gname cpu-memory-price ceil(ceil...
  1. Instances historization:
start_time end_time instance_name instance_promql instance_values
2021-10-26 10:25:53.735273 2021-10-26 10:25:54.193911 umber_instance_simulation_aws_a1_large ceil(ceil... {'cpu': '2', 'price': '0.05
1', 'memory': '4'}