Skip to main content

Kubernetes Resource Sizing for Ignition

Before continuing
  • charts.ia.io is the authoritative reference for the chart value keys this page mentions.

For the gateway's underlying CPU and memory sizing (tags, history rate, project complexity, module load), follow Inductive Automation's Server Sizing and Architecture Guide. This page covers what is specific to running Ignition on Kubernetes: how those targets map to Helm chart values, sizing the data volume, and configuring probes.

Mapping CPU and Memory to Chart Values​

Once you have target CPU and memory numbers from the IA sizing guide, they go into the chart's gateway.resources field:

gateway:
  resourcesEnabled: true
  resources:
    requests:
      cpu: "2"
      memory: 4Gi
    limits:
      cpu: "4"
      memory: 8Gi
  maxRAMPercentage: 75

The chart sets the JVM heap automatically from the container memory limit via gateway.maxRAMPercentage (default 75). The JVM also needs space for metaspace, native libraries, code cache, and direct buffers; keeping ~25% of the memory limit off-heap prevents the JVM from being OOM-killed by the kernel under load. Override gateway.maxRAMPercentage only if you have measured a reason to, and never set it above ~85% on a container with a hard memory limit.

See charts.ia.io for the full set of related values.

Data Volume Sizing​

The chart's gateway.dataVolumeStorageSize sets the PVC size for /usr/local/bin/ignition/data. The default is 3Gi, which is only suitable for short-lived lab clusters.

Profilegateway.dataVolumeStorageSizeWhat's stored
Small (dev/test)10GiInternal DB, modules, certs. No significant tag history.
Medium50GiAbove plus alarm journal and short-term tag history.
Large with history200Gi+Above plus tag history database if not externalized.
Production with external history DB50GiTag history offloaded to a dedicated database; PVC holds only internal state.

Tag history is best stored in an external database (PostgreSQL, MSSQL) for any production deployment, not in the gateway's internal database.

Data volume cannot be resized in place

gateway.dataVolumeStorageSize is baked into the StatefulSet's volumeClaimTemplates at install time. helm upgrade cannot change it - the API server rejects the StatefulSet update. To grow storage later you either use a CSI driver that supports volume expansion (and patch the PVC directly) or reinstall. Pick a generous size up front. The Helm Lab demonstrates this limitation.

Probe Configuration​

The chart configures readiness, liveness, and startup behavior using an exec-based command (the gateway image's bundled health-check.sh), not a raw HTTP probe. The defaults are sensible for Ignition's 60-180 second startup and you typically should not override them.

gateway:
  readinessProbe:
    initialDelaySeconds: 10
    periodSeconds: 5
    failureThreshold: 2
    timeoutSeconds: 3

Override the exec command itself with gateway.readinessProbe.commandOverride only if you have a specific reason; raw HTTP probes against the gateway can produce false negatives during slow startup or when the gateway is mid-restart. See charts.ia.io for the current default command.