Skip to content

finally fixed: run postgresql on kubernetes without pvb & keep your sanity

introduction: why everyone panics about postgresql on k8s

search for “postgresql on kubernetes” and you’ll see the same gloomy props: hundreds of yaml lines, persistentvolumeclaims (pvcs) piling up, and horror stories of data loss. as a devops full-stack coder, you want time for coding and seo, not to babysit volumes. good news: you can drop pvc/hostpath entirely for many workloads, keep your sanity, and still run a rock-solid postgresql cluster.

what to expect from this guide

  • a brief locker-room tour of when state must live in k8s volumes
  • the volume-less (pvc-less) pattern that relies on managed services and init containers
  • copy-paste-ready yaml and commands for beginners
  • tips to tune performance, monitor, and backup without losing your weekend to devops fire drills

stage 1: plan your data philosophy

rule of thumb

use pvc when:

  • you can’t tolerate multi-minute latency (cold start of managed pg)
  • you must stay self-hosted at the edge (air-gapped environments)

skip pvc when:

  • you’re allowed to ride managed postgresql (cloud sql, rds, alloydb, aurora)
  • your staging / development environments accept an occasional 3-5 min spin-up
never put production customer data on a container’s ephemeral storage—unless your managed host’s sla says it’s their data, not yours.

stage 2: pick your cloud sql or serverless shape

below table assumes google cloud sql, but the concept maps 1-to-1 in aws rds or azure db.

workload machine type cloud sql tier cost (usd / month)
dev / test 1 vcpu, 1–2 gb ram db-f1-micro $7-$10
small saas 2 vcpu, 4 gb ram db-g1-small $25-$30
medium prod 4 vcpu, 16 gb ram db-standard-4 $100-$130
tip: turn on automatic storage increases (continuous disk-resize) only if your clients obsess about rois; otherwise use alerts.

stage 3: let kubernetes find the database through secrets & config


apiversion: v1
kind: secret
metadata:
  name: pg-credentials
type: opaque
stringdata:
  postgres_user: "your_app"
  postgres_password: "super-secure-pw"
  postgres_db: "app_db"
  postgres_host: "10.123.45.67"   # cloud sql private ip
  postgres_port: "5432"

if you prefer environment variables, mount the secret instead of hard-coding.


apiversion: apps/v1
kind: deployment
metadata:
  name: web-api
spec:
  template:
    spec:
      containers:
      - name: app
        image: mycompany/backend:latest
        envfrom:
        - secretref:
            name: pg-credentials
        ports:
        - containerport: 8080
notice: zero pvc mounted—nothing to expand or back up on the cluster itself.

stage 4: tame migrations on first pod spin-up

skipping pvc means initcontainer runs migrations fast.


apiversion: batch/v1
kind: job
metadata:
  name: db-migrate
spec:
  template:
    spec:
      restartpolicy: onfailure
      containers:
      - name: migrate
        image: mycompany/backend:latest
        command: ["npx", "prisma", "migrate", "deploy"]
        envfrom:
        - secretref:
            name: pg-credentials

hook the job inside your github actions or jenkins pipeline so your ci/cd flow triggers migrations automatically—classic devops.stage 5: wire your application—minimal tcpping health check


readinessprobe:
  exec:
    command:
    - /bin/sh
    - -c
    - "pg_isready -h ${postgres_host} -p ${postgres_port} -u ${postgres_user}"
  initialdelayseconds: 5
  periodseconds: 10

this lives in your deployment alongside the main container, no kyverno, istio, or pvc required!

stage 6: cost-optimized backups & monitoring without local disk

  • auto-backup: flip the daily snapshot switch in the cloud sql console. export once a week with pg_dump --no-owner if you need open-source usda archives paranoid mode.
  • piggy-back observability: connect pg_stat_statements exporter to grafana cloud. done in 15 minutes.

stage 7: one-liners you’ll use every sprint

scenario command
connect from local gcloud sql connect app-db --user=your_app --database=app_db
get credentials into shell kubectl get secret/pg-credentials -o jsonpath='{.data.postgres_password}' | base64 -d
list running jobs kubectl get jobs --field-selector=metadata.name=db-migrate

common hurdles & quick fixes

  • hurdle: “i thought once cloud name changed, all drivers times-out.”
    fix: store hostname in k8s secret and reload pods via kubectl rollout restart deployment/web-api.
  • hurdle: slow cold boot scares qa during nightly integration tests.
    fix: enable cloud sql serverless v2 (scale-to-zero kept off) or keep a small idle instance ($2/day guardrail).
  • hurdle: “we need read-replicas regionally.”
    fix: one-click replica tier in cloud console; add reader endpoints to your postgres_host_reader variable—no pvc, no kubelets mutated.

tl;dr cheat sheet

  1. provision managed postgresql, **skip pvc**.
  2. store connection string in a k8s secret; mount via envfrom.
  3. run migrations inside **k8s job** so no local volume folders.
  4. backup and monitor outside the cluster—cheaper, cleaner, simpler docs for coding.

with this setup, postgresql becomes a world-class kitchen sink that doesn't leak into your cluster’s storage plumbing. happy coding, even happier seo headlines: “finally fixed: run postgresql on kubernetes without pvc & keep your sanity.”

The views expressed in this post are those of its author and do not represent pilput. See our Terms of Service.

Comments

Share your thoughts and join the conversation.

Loading comments...

Join the Discussion

Please log in to share your thoughts and engage with the community.