Tactic move to Kubernetes
At Tactic we had to redesign the entire system for horizontal and vertical scalability. One of the things to do was move from hosting the pipelines on EC2 servers, to EKS Kubernetes clusters. This allowed the following things:
- Fault tolerance - when a process crashes it would recreate itself automatically
- Monitoring - we deployed Datadog within the cluster as well as having a nice overview of the system using the k9s tool (https://github.com/derailed/k9s)
- Service management - tools such as Datadog and Nginx were able to be installed and managed easily
- Automatic deployment - Helm and Docker allowed for seamless deployment and rollback on push with CICD
Inside our Kubernetes clusters, we also set up pod and node autoscaling. All processes in the system were run on SQS queues and Keda allowed us to scale our pods up and down as necessary based upon the queue size. We then used the node autoscaler which would interact with EKS and alter the number of nodes based upon the pods that were deployed. This way we had effectively unlimited scalability and a cost profile that scaled with usage.
Explore Others
PhishingLine
PhishingLine
A Linear regression model for detection of phishing pages.
Lyst data lake migration
Lyst data lake migration
A project transforming the data pipeline at
a fashion scale up
This website
This website
How this website is provisioned and deployed on
AWS