Kubernetes with Helm

import { Aside, Steps } from ‘@astrojs/starlight/components’;

The PulsusDB repository includes a Helm chart at deploy/charts/pulsusdb. The chart supports two independent topology choices:

PulsusDB as one all Deployment or split writer and reader Deployments;
bundled ClickHouse as one node or a sharded cluster.

Install from the OCI chart

helm install pulsusdb oci://ghcr.io/digitalis-io/charts/pulsusdb \
  --version CHART_VERSION \
  --set initJob.retentionDays=7 \
  --set image.tag=APPLICATION_IMAGE_TAG

initJob.retentionDays is required while the schema Job is enabled. The application image tag must refer to an image you can pull.

Evaluate from a checkout

git clone https://github.com/digitalis-io/pulsusdb.git
cd pulsusdb
helm upgrade --install pulsusdb deploy/charts/pulsusdb \
  --set initJob.retentionDays=7 \
  --set image.tag=APPLICATION_IMAGE_TAG

Split writer and reader tiers

pulsusdb:
  split:
    enabled: true
  writer:
    replicaCount: 2
  reader:
    replicaCount: 2

initJob:
  retentionDays: 30

helm upgrade --install pulsusdb deploy/charts/pulsusdb -f values-split.yaml

Use external ClickHouse

clickhouse:
  enabled: false
  auth:
    existingSecret: pulsusdb-clickhouse
    existingSecretPasswordKey: password

pulsusdb:
  config:
    clickhouse:
      server: clickhouse.example.internal
      http_port: 8123
      database: pulsus

initJob:
  retentionDays: 30

Create the referenced Secret without committing credentials:

kubectl create secret generic pulsusdb-clickhouse \
  --from-literal=password='replace-me'

Probe contract

Probe	Handler	Meaning
Liveness	TCP on the configured PulsusDB port	Restart only if the process stops listening
Readiness	`GET /ready`	Remove the pod from service when ClickHouse or reader warm-up is unavailable
Startup	TCP on the configured PulsusDB port	Delay liveness until the listener binds

Do not replace liveness with /ready. A ClickHouse outage should make PulsusDB unready, not restart every pod.

Schema Job warning

The chart’s schema Job runs the destructive schema/schema.sh script. Its name changes when the Job spec changes, which can cause a Helm upgrade to create a new Job and rebuild the database. Review rendered manifests before upgrading:

helm template pulsusdb deploy/charts/pulsusdb \
  --set initJob.retentionDays=30 > rendered.yaml

For a controlled data lifecycle, disable the Job and operate schema changes yourself:

initJob:
  enabled: false