Argo Workflow Helm Chart

Published
Length
3 min read

Luis Felipe Dias

All posts

You deploy the argo workflow helm chart to run container jobs on Kubernetes. This chart sets up the main controller, the web interface, and the cluster roles. Teams often install the chart with broad cluster permissions to make jobs run fast. When roles are too wide, any job or user can change live pods. A wide role also lets a bad task read secrets across the cluster. How the argo workflow helm chart sets up roles The argo workflow helm chart bundles all parts needed to run jobs in Kubernete

Your coding agents run with your access.

Claude Code reaches prod with a developer's credentials, and approval prompts get clicked through. hoop.dev checks every query, masks sensitive data, and holds risky changes for a human.

You deploy the argo workflow helm chart to run container jobs on Kubernetes. This chart sets up the main controller, the web interface, and the cluster roles. Teams often install the chart with broad cluster permissions to make jobs run fast. When roles are too wide, any job or user can change live pods. A wide role also lets a bad task read secrets across the cluster.

How the argo workflow helm chart sets up roles

The argo workflow helm chart bundles all parts needed to run jobs in Kubernetes. It installs two main deployments inside your target namespace. The first part is the workflow controller, which watches for new job definitions. When a new job arrives, the controller creates pods to run each step. The second part is the server, which provides a visual dashboard and an API.

To do this work, the chart creates service accounts and binds them to roles. You can set up the chart to work across the whole cluster. You can also lock it down to a single namespace for tighter control. The controller service account needs rights to create pods and read config maps. It also watches pod status so it knows when a task finishes.

Task pods do not use the controller service account to run their steps. Instead, each workflow uses a separate service account to talk to the Kubernetes API. The executor inside the pod uses this account to save logs and outputs. If a team skips setting up a specific role, pods use the default account. Teams often grant extra rights to that default account so jobs do not fail. That small choice gives every task pod the same high level of power.

Why native chart permissions need careful boundaries

Setting up namespaces and role bindings in the chart solves basic cluster isolation. You can stop a team in one namespace from seeing jobs in another namespace. You can also tie the dashboard to your identity provider for single sign-on. This maps user groups to specific roles when people log in to check jobs.

Still, chart roles only check who starts a session or which pod runs. They do not inspect the actual commands or database queries inside a running step. Once a human or a job has a valid role, the cluster trusts them. An engineer with edit rights on the cluster can open a shell into pods. They can also read raw database credentials stored in Kubernetes secrets.

Helm Chart Security + Agentic Workflow Security: architecture patterns and best practices

Continue reading? Get the full guide.

Free. No spam. Unsubscribe anytime.

At the same time, automated jobs often connect to production databases to move data. The chart has no built-in way to mask sensitive fields in query results. If a script pulls customer records, raw emails and card numbers flow into logs. Native Kubernetes audit logs record that a pod started or a secret was read. They do not show which SQL queries ran or what data left the database.

Managing secrets and storage in your chart setup

Workflows often need to pass files between steps using object storage like S3. The chart lets you set up an artifact store for all job outputs. To reach that store, the chart mounts storage keys through standard Kubernetes secrets. You should scope these secrets so only specific service accounts can read them.

When jobs need to reach external databases, teams face a similar secret problem. Storing static database passwords in chart values puts plain text keys in version control. A better path is to keep secrets out of the chart values file. You should create secret objects in the namespace before you run the chart install. Even then, any user who can run a pod with that service account can read those mounted keys. You must keep the controller role, the server role, and the task role separate.

Check Kubernetes API requests and responses on the wire

Kubernetes RBAC decides who may call the API server. RBAC does not check response payloads or split listing Secret names from reading Secret values when a role grants get access. With hoop.dev, you run an open-source proxy, a sidecar for your Kubernetes API, inline on the API connection. On every kubectl or API call an AI agent or engineer runs against production, hoop.dev decides to pass, mask, block, or hold it for human review.

On each call, hoop.dev guardrails read the method, normalized path, and headers to refuse bad deletes or full Secret reads with a 403 error, or hold risky calls for approval. On the way back, hoop.dev data masking walks JSON responses by key and masks matched values in Secrets, ConfigMaps, and pod logs in memory. View the project on GitHub.

Frequently asked questions

Should I install the chart in cluster mode or namespace mode?

Use namespace mode when different teams share the same Kubernetes cluster. This stops one controller or job from reading secrets in other namespaces. Use cluster mode only when one platform team runs all jobs for everyone.

Why do my workflow pods fail with permission errors on startup?

The pod executor needs rights to patch pod metadata and read task results. Make sure the service account assigned to the workflow has a valid role binding. Do not fix this by giving cluster admin rights to the default account.

How do I keep plain text passwords out of my chart values?

Never type database passwords or storage keys directly into your custom values file. Create the Kubernetes secret in your namespace first using your secret management tool. Then tell the chart to reference that existing secret by its name.

The query runs. The emails stay hidden.