Autoscaling adjusts application or infrastructure capacity as workload demand changes. In serverless platforms, the platform handles scaling while developers deploy containerized services rather than managing underlying compute. section 9.1 section 9.2.1
Implementation choices include:
- Scale containers from incoming requests, including scaling resources to zero when no workload is active. section 9.1 section 9.2.1
- Tune Knative services with concurrency limits and minimum or maximum container instances. section 9.2.1
- Choose horizontal or vertical infrastructure arrangements when scaling Kubernetes environments across teams, stages, or locations. section 11.1.1
Cluster autoscaling can help control costs and enforce minimum resource requirements. Treat deletion as an operational dependency: remove node pools before deleting the cluster, and protect SSH keys with a system such as HashiCorp Vault for retrieval and rotation. section 5.6.2
Request-driven service scaling
Deploy an HTTPS-serving container to Cloud Run. The platform scales its container instances according to requests, while the service can use any language capable of answering HTTPS calls and does not require Anthos. section 9.1
Table 9.1 Knative vs. PaaS, CaaS, and FaaS (view table figure)
Compare autoscaling, scale-to-zero, load balancing, execution limits, and language support across Knative, PaaS, CaaS, and FaaS. section 9.2.1
