Kubernetes (Helm)

This page documents the Helm chart in charts/dagu as implemented in this repository.

Chart Path

text

charts/dagu

Rendered Objects

For a release named <release>, the chart renders:

ConfigMap/<fullname>-config
PersistentVolumeClaim/<fullname>-data
Deployment/<fullname>-coordinator
Deployment/<fullname>-scheduler
Deployment/<fullname>-ui
Deployment/<fullname>-worker-<poolName> for each entry in workerPools
Service/<fullname>-coordinator (ClusterIP, port 50055)
Service/<fullname>-scheduler (ClusterIP, port 8090)
Service/<fullname>-ui (ClusterIP, port ui.service.port)

The chart does not render:

Ingress
NetworkPolicy
ServiceAccount, Role, or RoleBinding
HorizontalPodAutoscaler
PodDisruptionBudget
Services for worker Deployments

<fullname> is resolved by the helper in charts/dagu/templates/_helpers.tpl:

fullnameOverride if set
otherwise let <name> be nameOverride if set, or dagu if not
if the release name already contains <name>, use the release name unchanged
otherwise use <release>-<name>

nameOverride and fullnameOverride are supported by the templates even though they are not listed in charts/dagu/values.yaml.

Install

Official Helm repository URL:

text

https://dagu-org.github.io/dagu

Add the repository and install the chart:

bash

helm repo add dagu https://dagu-org.github.io/dagu
helm repo update
helm install dagu dagu/dagu --set persistence.storageClass=<your-rwx-storage-class>

Render manifests without installing:

bash

helm template dagu dagu/dagu --set persistence.storageClass=<your-rwx-storage-class>

To upgrade an existing release:

bash

helm repo update
helm upgrade dagu dagu/dagu --set persistence.storageClass=<your-rwx-storage-class>

From the repository root, the local chart path remains available:

bash

helm install dagu ./charts/dagu --set persistence.storageClass=<your-rwx-storage-class>

Replace <your-rwx-storage-class> with a StorageClass in your cluster that supports ReadWriteMany. If your cluster default storage class already supports ReadWriteMany, you can omit the flag.

Version Fields

charts/dagu/Chart.yaml defines the chart version, which is the version published in the Helm repository.

The deployed container image comes from values.yaml -> image.repository and values.yaml -> image.tag. With the current defaults, the chart deploys ghcr.io/dagucloud/dagu:latest.

Pod Configuration

All Deployments created by the chart:

mount the shared PVC at /data
mount the ConfigMap at /etc/dagu
set enableServiceLinks: false
set DAGU_HOME=/data

Additional per-component environment variables:

UI: DAGU_PORT=<ui.service.port>
Coordinator: DAGU_COORDINATOR_HOST=0.0.0.0DAGU_COORDINATOR_ADVERTISE=<fullname>-coordinator.<namespace>.svc.cluster.local
Worker: DAGU_WORKER_ID from metadata.name

The chart also supports a shared extraEnv value. Every entry in extraEnv is added to the env: list of the coordinator, scheduler, UI, and all worker Deployments.

Container commands are fixed by the templates:

text

dagu coordinator --config /etc/dagu/dagu.yaml
dagu scheduler --config /etc/dagu/dagu.yaml
dagu server --config /etc/dagu/dagu.yaml
dagu worker --config /etc/dagu/dagu.yaml [--worker.labels ...]

Dagu Configuration Written By The Chart

The chart writes the following dagu.yaml to <fullname>-config:

yaml

host: "0.0.0.0"
port: 8080
api_base_path: "/api/v1"
default_execution_mode: "distributed"

coordinator:
  host: "0.0.0.0"
  port: 50055

scheduler:
  port: 8090

paths:
  data_dir: /data
  dags_dir: /data/dags
  log_dir: /data/logs
  base_config: /data/base.yaml
  suspend_flags_dir: /data/suspend
  admin_logs_dir: /data/admin/logs
  dag_runs_dir: /data/dag-runs
  queue_dir: /data/queue
  proc_dir: /data/proc
  service_registry_dir: /data/services
  api_keys_dir: /data/api-keys
  webhooks_dir: /data/webhooks
  users_dir: /data/users

auth:
  mode: "builtin"
  builtin:
    token:
      secret: ""
      ttl: "24h"

peer:
  insecure: true

The auth.mode, auth.builtin.token.secret, and auth.builtin.token.ttl fields come from values.yaml. The other fields shown above are fixed by the chart templates.

When config.envPassthrough is non-empty, the chart also writes top-level env_passthrough into dagu.yaml.

When config.envPassthroughPrefixes is non-empty, the chart also writes top-level env_passthrough_prefixes into dagu.yaml.

The UI Deployment also sets DAGU_PORT=<ui.service.port>. That environment variable overrides port from dagu.yaml at runtime when ui.service.port is changed from the default.

The following paths are not set explicitly in the chart, but Dagu derives them from paths.data_dir and paths.dags_dir at runtime:

yaml

paths:
  docs_dir: /data/dags/docs
  sessions_dir: /data/agent/sessions
  remote_nodes_dir: /data/remote-nodes
  workspaces_dir: /data/workspaces

That path layout is consistent with DAGU_HOME=/data and with the explicit base_config: /data/base.yaml.

The chart does not render DAG definition files. paths.dags_dir points to /data/dags on the shared PVC. Populate that directory through the UI/API or by writing files into the shared volume.

Values Exposed By `values.yaml`

The chart currently defines these top-level values in charts/dagu/values.yaml:

yaml

image:
  repository: ghcr.io/dagucloud/dagu
  tag: latest
  pullPolicy: IfNotPresent

scheduler:
  replicas: 1
  resources: ...

coordinator:
  replicas: 1
  resources: ...

workerPools:
  general:
    replicas: 2
    labels: {}
    resources: ...
    nodeSelector: {}
    tolerations: []
    affinity: {}

ui:
  replicas: 1
  service:
    port: 8080
  resources: ...

auth:
  mode: "builtin"
  builtin:
    token:
      secret: ""
      ttl: "24h"

persistence:
  enabled: true
  accessMode: ReadWriteMany
  size: 10Gi
  storageClass: ""
  skipValidation: false

config:
  envPassthrough: []
  envPassthroughPrefixes: []

extraEnv: []

image.pullPolicy is the actual key used by the chart. pull_policy is not used.

Step Environment Passthrough

The chart exposes Dagu's environment passthrough settings through values.yaml -> config:

yaml

config:
  envPassthrough:
    - SSL_CERT_FILE
    - HTTP_PROXY
    - HTTPS_PROXY
    - NO_PROXY
  envPassthroughPrefixes:
    - AWS_

Those values only change Dagu's filter. They do not create container environment variables.

If the variable must come from the pod environment, inject it with extraEnv:

yaml

extraEnv:
  - name: HTTP_PROXY
    value: http://proxy.example.com:8080
  - name: HTTPS_PROXY
    value: http://proxy.example.com:8080
  - name: NO_PROXY
    value: 127.0.0.1,localhost,.svc,.cluster.local
  - name: SSL_CERT_FILE
    value: /etc/ssl/certs/custom-ca.pem

Combined example:

yaml

config:
  envPassthrough:
    - SSL_CERT_FILE
    - HTTP_PROXY
    - HTTPS_PROXY
    - NO_PROXY

extraEnv:
  - name: HTTP_PROXY
    value: http://proxy.example.com:8080
  - name: HTTPS_PROXY
    value: http://proxy.example.com:8080
  - name: NO_PROXY
    value: 127.0.0.1,localhost,.svc,.cluster.local
  - name: SSL_CERT_FILE
    value: /etc/ssl/certs/custom-ca.pem

With that configuration:

the container processes receive the variables from extraEnv
Dagu forwards the listed variables into step execution
variables not present in the pod environment are not created by config.envPassthrough

Persistence Rules

The PVC template enforces two conditions:

persistence.enabled must be true
persistence.accessMode must be ReadWriteMany unless persistence.skipValidation=true

Examples:

bash

helm install dagu dagu/dagu \
  --set persistence.storageClass=<your-rwx-storage-class>

bash

helm install dagu dagu/dagu \
  --set persistence.accessMode=ReadWriteOnce \
  --set persistence.skipValidation=true \
  --set workerPools.general.replicas=1

persistence.skipValidation=true only disables the chart's RWX check. It does not change how Kubernetes storage works.

If persistence.storageClass is the empty string, the rendered PVC omits storageClassName and Kubernetes uses the cluster default behavior.

Worker Pools

Each entry in workerPools creates one Deployment:

text

<fullname>-worker-<poolName>

Pool names are validated by the template and must match:

text

^[a-z][a-z0-9-]*$

Examples:

valid: general, gpu-workers, us-east-1
invalid: GPU, 1-workers, my_pool

Worker Labels

workerPools.<poolName>.labels are Dagu worker capability labels. They are not copied into Kubernetes pod metadata.

The worker pod metadata always includes daguit.dev/worker-pool: <poolName>, but the contents of workerPools.<poolName>.labels are only used to build the --worker.labels argument.

When labels is non-empty, the worker Deployment adds:

text

--worker.labels <comma-separated key=value pairs>

When labels: {} is used, the chart does not add a --worker.labels argument.

Example values:

yaml

workerPools:
  gpu:
    replicas: 2
    labels:
      gpu: "true"
      cuda: "12.0"
    resources:
      requests:
        memory: "1Gi"
        cpu: "500m"
        nvidia.com/gpu: "1"
      limits:
        memory: "4Gi"
        cpu: "2"
        nvidia.com/gpu: "1"
    nodeSelector:
      nvidia.com/gpu.present: "true"
    tolerations:
      - key: nvidia.com/gpu
        operator: Exists
        effect: NoSchedule

The rendered worker command includes a --worker.labels flag built from that map. The exact pair order should not be relied on.

Kubernetes Scheduling Fields

These pool fields are copied directly into the worker pod spec:

nodeSelector
tolerations
affinity
resources

Example:

yaml

workerPools:
  gpu:
    replicas: 1
    labels:
      gpu: "true"
    nodeSelector:
      nvidia.com/gpu.present: "true"
    tolerations:
      - key: nvidia.com/gpu
        operator: Exists
        effect: NoSchedule
    affinity:
      nodeAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          nodeSelectorTerms:
            - matchExpressions:
                - key: nvidia.com/gpu.product
                  operator: In
                  values:
                    - NVIDIA-A100-SXM4-40GB

Minimal Values Example

yaml

image:
  tag: latest

auth:
  mode: builtin
  builtin:
    token:
      secret: ""
      ttl: 24h

persistence:
  enabled: true
  accessMode: ReadWriteMany
  size: 10Gi
  storageClass: "<your-rwx-storage-class>"

workerPools:
  general:
    replicas: 2
    labels: {}
    resources:
      requests:
        memory: 128Mi
        cpu: 100m
      limits:
        memory: 256Mi
        cpu: 200m
    nodeSelector: {}
    tolerations: []
    affinity: {}

Install with that file:

bash

helm install dagu dagu/dagu -f values.yaml

To force a different image tag:

yaml

image:
  tag: 2.2.4

Access The UI

bash

kubectl port-forward svc/dagu-ui 8080:8080

If you used a different release name or fullnameOverride, replace dagu-ui with the rendered UI service name.

Installation

Secrets

Workers

Kubernetes (Helm)

Chart Path

Rendered Objects

Install

Version Fields

Pod Configuration

Dagu Configuration Written By The Chart

Values Exposed By `values.yaml`

Step Environment Passthrough

Persistence Rules

Worker Pools

Worker Labels

Kubernetes Scheduling Fields

Minimal Values Example

Access The UI

Kubernetes (Helm) ​

Chart Path ​

Rendered Objects ​

Install ​

Version Fields ​

Pod Configuration ​

Dagu Configuration Written By The Chart ​

Values Exposed By values.yaml ​

Step Environment Passthrough ​

Persistence Rules ​

Worker Pools ​

Worker Labels ​

Kubernetes Scheduling Fields ​

Minimal Values Example ​

Access The UI ​

Kubernetes (Helm)

Chart Path

Rendered Objects

Install

Version Fields

Pod Configuration

Dagu Configuration Written By The Chart

Values Exposed By `values.yaml`

Step Environment Passthrough

Persistence Rules

Worker Pools

Worker Labels

Kubernetes Scheduling Fields

Minimal Values Example

Access The UI