7. Pin Supabase to x86 Nodes#

Status: Accepted

Context#

Supabase container images have inconsistent ARM64 support. The community Helm chart uses suffix-based image tags for architecture, and Kong in particular lacks reliable ARM64 images.

Decision#

Set nodeSelector: kubernetes.io/arch: amd64 on all Supabase components. Clusters with only one x86 node will concentrate all Supabase pods there; adding a second dedicated x86 worker distributes the load.

Example: in the author’s cluster, Supabase initially ran on ws03 (a workstation) and later moved to nuc2 (a dedicated Intel NUC worker).

Consequences#

  • All Supabase pods require at least one x86/amd64 node in the cluster

  • Total Supabase footprint is ~2.5 GB RAM across all components

  • ARM migration possible but not prioritized (Kong lacks reliable ARM64 images)

  • If the only x86 node is a workstation, a toleration is needed temporarily; remove it once a dedicated x86 worker is available (see ADR 0005)