Horizontal Pod Autoscaler

Seven applications have HPA enabled for automatic scaling based on CPU utilization.

HPA-Managed Applications

Application	Min Replicas	Max Replicas	CPU Target
bookmarked	1	10	80%
educationelly	1	10	80%
educationelly-graphql	1	10	80%
firebook	1	10	80%
intervalai	1	10	80%
portfolio-api	1	10	80%
tenantflow	1	10	80%

The HPA vs ArgoCD Flapping Problem

When HPA is managing replica counts, ArgoCD can enter an infinite sync loop:

Root Cause

The portfolio-common deployment template always rendered spec.replicas, even when HPA was enabled. ArgoCD saw the Git-declared replica count as the desired state, while HPA continuously adjusted it.

The Fix (Two-Part)

1. Conditional replicas in the library chart:

# portfolio-common/templates/_deployment.tpl
{{- if not $.Values.autoscaling.enabled }}
replicas: {{ $.Values.replicaCount }}
{{- end }}

When autoscaling.enabled: true, the spec.replicas field is omitted entirely, letting HPA be the sole owner.

2. ArgoCD ignoreDifferences:

As a safety net, ArgoCD Application manifests include:

spec:
  ignoreDifferences:
    - group: apps
      kind: Deployment
      jsonPointers:
        - /spec/replicas

This tells ArgoCD to ignore the spec.replicas field when computing drift, preventing sync loops even if the field is present.

Symptoms

• ArgoCD shows "OutOfSync" status with "successfully synced" message
• Pods constantly recycling (terminate → create → terminate)
• ArgoCD sync history shows rapid successive syncs

PodDisruptionBudgets

HPA-managed and multi-replica applications have PodDisruptionBudgets (PDBs) to protect availability during voluntary disruptions like node drains, cluster upgrades, or rolling updates.

The library chart includes a _pdb.tpl template, with each app including a thin pdb.yaml wrapper:

# bookmarked/templates/pdb.yaml
{{- include "portfolio-common.pdb" . }}

Applications with PDBs

Application	Why
bookmarked	HPA-enabled, multi-replica
educationelly	HPA-enabled, multi-replica
educationelly-graphql	HPA-enabled, multi-replica
intervalai	HPA-enabled, multi-replica
portfolio-gatsby	Multi-replica static site

PDBs ensure the cluster maintains minimum available pods when a node is drained for maintenance. On a 3-node cluster, this is critical — draining one node moves ~33% of workloads, and without PDBs, all replicas of a service could be evicted simultaneously.