Guide for how to use keda to auto scaling

[hongbo-miao]

It would be great to provide a guide for how to use Keda to auto scaling.

Currently I have Keda in my Kubernetes cluster. Here is my Trino values.yaml file:

# ...
server:
  workers: 10
  keda:
    enabled: true
    pollingInterval: 30
    cooldownPeriod: 300
    initialCooldownPeriod: 0
    minReplicaCount: 0
    maxReplicaCount: 20
    triggers:
      - type: prometheus
        metricType: Value
        metadata:
          serverAddress: http://hm-mimir-query-frontend.production-hm-mimir.svc:8080/prometheus
          threshold: "1"
          metricName: required_workers
          query: >-
            sum by (service)
            (avg_over_time(trino_execution_ClusterSizeMonitor_RequiredWorkers{service={{ include "trino.fullname" . | quote }}}[5s]))

If I have

jmx:
  enabled: true
  exporter:
    enabled: true

and let Prometheus scrape it and write to Mimir.

Do I still need?

serviceMonitor:
  enabled: true

Sorry I am quite lost. 😞

[nineinchnick]

Are you using the Prometheus Operator? Then you need the service monitor. I'm not sure if we need a complete guide, or just reference Prometheus docs.

[hongbo-miao]

Thanks, @nineinchnick! I believe the ServiceMonitor approach and direct scraping are two different ways to write data to Prometheus, so we might only need one of them (?) Please correct me if I am wrong. Thanks!

jmx:
  enabled: true
  exporter:
    enabled: true
serviceMonitor:
  enabled: false

or

jmx:
  enabled: true
  exporter:
    enabled: false
serviceMonitor:
  enabled: true

[nineinchnick]

You haven't answered if you're using Prometheus Operator, so I can't recommend you any specific setup.

[hongbo-miao]

Oh I am, here is my Prometheus deployment code with Prometheus Operator enabled: https://github.com/hongbo-miao/hongbomiao.com/blob/adae8e52f54670f76dd9499be080f328fd218592/kubernetes/argo-cd/applications/production-hm/prometheus/argo-cd-manifests/hm-prometheus-application.yaml#L31-L32

[nineinchnick]

ServiceMonitor is used by the Operator to configure Prometheus to scrape metrics from Trino. You need the JMX exporter, so Prometheus can scrape metrics. Trino also exposes JMX metrics as OpenMetrics, so you might not need the exporter, but it might require some extra configuration: https://trino.io/docs/current/admin/openmetrics.html

[luismacosta]

It would be great to provide a guide for how to use Keda to auto scaling.

Currently I have Keda in my Kubernetes cluster. Here is my Trino values.yaml file:

...

server:
workers: 10
keda:
enabled: true
pollingInterval: 30
cooldownPeriod: 300
initialCooldownPeriod: 0
minReplicaCount: 0
maxReplicaCount: 20
triggers:
- type: prometheus
metricType: Value
metadata:
serverAddress: http://hm-mimir-query-frontend.production-hm-mimir.svc:8080/prometheus
threshold: "1"
metricName: required_workers
query: >-
sum by (service)
(avg_over_time(trino_execution_ClusterSizeMonitor_RequiredWorkers{service={{ include "trino.fullname" . | quote }}}[5s]))
If I have

jmx:
  enabled: true
  exporter:
    enabled: true

and let Prometheus scrape it and write to Mimir.

Do I still need?

serviceMonitor:
  enabled: true

Sorry I am quite lost. 😞

I've implemented workers autoscaling based in this https://medium.com/bestsecret-tech/maximize-performance-the-bestsecret-to-scaling-trino-clusters-with-keda-c209efe4a081