Skip to main content

Kubernetes Deployment

This section covers deployment and tuning for Kubernetes administrators. The iom-maintenance service is always deployed, so there is no feature flag to enable.

Resource Defaults​

Tune CPU and memory limits for the iom-maintenance service via Helm:

services:
maintenance:
resources:
requests:
memory: "1000Mi"
cpu: "2000m"
limits:
memory: "1000Mi"
cpu: "2000m"
Requests and limits must match

Set requests and limits to the same values. Mismatched values can cause maintenance operations to run in loops.

Archival Configuration​

Completed maintenance runs are archived to Iceberg tables for long-term retention. Configure via Helm:

services:
maintenance:
archival:
enabled: true
retentionDays: 30

How archival works:

  • Schedule: archival runs every hour.
  • What moves: finished runs older than retentionDays are moved in batches of 500.
  • Where they go: three Iceberg tables in spark_catalog.iomete_system_db:
    • maintenance_evaluation_runs
    • maintenance_execution_runs
    • maintenance_sql_runs
note

Archived runs no longer appear in the console's run history. To see runs older than retentionDays, query the archive tables in the SQL Editor.