Download as PDF, PPTX
![Real life example for event driven scaling
15/22
● During game events we see huge spike in incoming requests during very short
period of time [~5minutes]
○ If we knew in advance that game events starts at 12:00 can we prepare our
application/infrastructure for it ?
● Sometimes we have to kick players from game and force them to relogin which
is a DB heavy operation
○ We want to do it in batches while monitoring DB load. To make sure that we do not overload it.
● Scale Message queue consumers based on queue messages](https://image.slidesharecdn.com/cncf-eventdrivenautoscalingwithkeda-210218172137/75/Cncf-event-driven-autoscaling-with-keda-15-2048.jpg)
More Related Content
Similar to Cncf event driven autoscaling with keda
Cncf event driven autoscaling with keda
- 1.
- 2. Labyrinth Labs Rock-solid infrastructureand DevOps ● Building rock-solid and secure foundations for all your digital operations. Our mission is to let you focus on your business without ever needing to worry about technical issues again. ● Making you ready for growing traffic, safe against new security vulnerabilities and data-loss. 2
- 3. TL;DR ● What isevent driven architecture and why does it look like serverless ? ● How KEDA works ○ When does it make sense to use it ? ○ What scalers KEDA supports ? ● Do you have any real life example for KEDA and event driven autoscaling ? 3
- 4. Event Driven Architecture ●Architecture design where systems/services communicate with events/messages. ● Event broker/message bus is used/shared for communication between services ● Services can scale based on events not resource load ○ Lambda, knative, functions ● Can process huge number of events very quickly ● Can make sure time ordering of events 4
- 5.
- 6. Simple explanation ● StandardK8s HPA is monitoring particular metrics and scale deployment based on them. ○ Imagine you are inviting friends to a BD party and you want to order a pizza for snack. In HPA terms you will order one pizza and when they will be eaten for 60% your order new batch. ● KEDA helps us to change HPA behaviour to Event driven ○ You order new pizza every time new guest comes to a party. 6
- 7. Kubernetes Event-driven Autoscaling ● OpenSource project (MIT license) ○ Originally written by Microsoft/Redhat engineers ○ Current version is v2.1.0 released in January 2021 ○ Open and built with community ○ Hosted at: https://github.com/kedacore/keda ● Part of CNCF currently in a Sandbox 7
- 8. Kubernetes Event-driven Autoscaling ● Monitorsrate of events and to proactively scale application resources / workload. ● Can upscale/downscale HPA/Deployment from/to zero containers. ○ Deployment can have 0 replicas at beginning (e.g. if there are no messages in a queue) ● Keda gathers info from scalers and feeds it to Kubernetes API ○ This will reuse already existing Metrics server and HPA implementations. ● Support large number of scalers, can be extended ○ AWS SQS, AWS cloudwatch, AWS kinesis, Apache Kafka, RabbitMQ, ActiveMQ, Google Cloud Platform Pub/Sub, Azure Storage Queue, Azure Service Bus and many more. 8
- 9.
- 10. Core Principles ● Workswith any Kubernetes cluster ● Can scale any container, deployment or workload. ● Don’t rebuild anything Kubernetes already has ○ HPA, Metrics Server ○ Use existing technologies to extend kubernetes features CRD 10
- 11.
- 12. Long running jobs Problem:If we scale scale our workload for message processing and it takes a while to finish single message we risk HPA will downscale our pod. ● Generic Kubernetes solution: is to postpone pod termination indefinitely (very long gracefulTerminationPeriod + lifecycle preStop hook. ● Use KEDA ScaleJob CRD to create k8s jobs per event. 12
- 13.
- 14. 14 ● RabbitMQ messageconsumer deployment ● Keda ScaleOut object monitoring number of messages in a given queue and scaling our deployment.
- 15. Real life examplefor event driven scaling 15/22 ● During game events we see huge spike in incoming requests during very short period of time [~5minutes] ○ If we knew in advance that game events starts at 12:00 can we prepare our application/infrastructure for it ? ● Sometimes we have to kick players from game and force them to relogin which is a DB heavy operation ○ We want to do it in batches while monitoring DB load. To make sure that we do not overload it. ● Scale Message queue consumers based on queue messages
- 16. Real life examplefor event driven scaling 16/22
- 17. Real life examplefor event driven scaling 17/22
- 18.
- 19. Thank You. We arehiring, remote working DevOps/Kubernetes engineers. [email protected] www.lablabs.io 19