Scaling Catwalks in GCP
How is the scaling configured in GCP? How does the scaling of Catwalks work? What are the KPIs (CPU usage, load, ...?) used to determine if there needs to be a scaling? What are the reaction-times we can expect?
In general we use Google's autoscaling group feature.
Currently, we use the CPU usage as a scaling factor as we made the best experience with that context in mind but we are aware that this might differ in the future, so our rolling-update script allows to parameterize those settings per customer.
As threshold, so that enough instances are available in load scenarios, we identified a value between 70-80%.
At the moment we use high CPU, normal RAM instances (4 vCPUs, 3,6 GB - n1-highcpu-4) because in the current scenarios Frontastic needs more computing power than memory. So we also allow this to be customized per customer.