This guide will teach you how to deploy Briefer on your Kubernetes cluster so that your whole team can access it and collaborate.

We’ll use Briefer’s Helm Chart in this guide. Still, you can feel free to take that chart and customize it to your liking, including using tools like Kustomize.

​
A Kubernetes deployment overview

Once complete, your Kubernetes deployment should include Briefer’s web and api services, a Jupyter server, and a Postgres database to store data. It can also include the AI service in case you want to enable code generation features.

All of Briefer’s components are included in its helm chart, except for the Ingress resources, which you should create yourself taking into account which ingress controller you’re using.

We recommend that you use an external Postgres instance like Supabase or AWS RDS.

​
Deploying Briefer using its Helm chart

Using Briefer’s Helm chart is the easiest way of setting up Briefer on your Kubernetes cluster.

  1. Start by adding Briefer’s chart repository to your Helm repository list. 
    helm repo add briefercloud https://charts.briefer.cloud
  2. Create a values.yaml file in which you’ll configure the URLs you want Briefer to use. 
    api:
  env:
    # update these two URLs with your future ingress URLs
    frontendUrl: https://app.briefer.yourcompany.com
    apiUrl: http://api.briefer.yourcompany.com
  3. Now, install the briefer chart from the newly added chart repository, passing your values.yaml file. 
    helm install briefer briefercloud/briefer \
-f ./values.yaml \
--namespace briefer \
--create-namespace
    Feel free to change the namespace if you want to. If you’re using an existing namespace, please remove the --create-namespace flag.
  4. After installing Briefer’s Helm chart, you can confirm that the manifests have been applied by listing all the pods in the chart’s target namespace (usually kubectl get pods -n briefer).

    If you’re running a Postgres instance within your cluster, it might be that the API will be ready before the Postgres instance is up. In that case, you’ll see that the API pod will keep restarting for a few minutes. That’s normal. It happens because the API needs to connect to the database to start, so it’ll just keep trying to start until Postgres is ready.

  5. Create the necessary Ingress resources so that you can access Briefer’s API and Web containers. Make sure to use the same URLs you’ve configured in step 2.

Now, see the section below to learn how to configure your Briefer deployment on Kubernetes.

​
Configuring your Briefer deployment on Kubernetes

You can configure your Briefer deployment through a values.yaml file that you’ll pass to Helm when installing Briefer’s chart.

In this section we’ll describe the most common configurations you might want to make.

To see a full values.yaml example, check out our GitHub repo.

​
Setting up Briefer’s URLs (mandatory)

When installing Briefer, you must tell it what are the URLs for the web and api services.

You can do that by setting api.env.frontendUrl and api.env.apiUrl in your values.yaml file.

api:
  env:
    # update these two URLs with your future ingress URLs
    frontendUrl: https://app.briefer.yourcompany.com
    apiUrl: http://api.briefer.yourcompany.com

Make sure that these values match the address you’re using in your Ingress resources.

​
Using an external Postgres instance (RDS, Supabase, etc)

We strongly recommend that you use an external Postgres instance to store Briefer’s data.

For that, you’ll need to disable enableLocalPostgres and set api.env.postgresHostname, api.env.postgresPort, api.env.postgresDatabase, api.secrets.postgresUsername, and api.secrets.postgresPassword, as shown in the example below.

enableLocalPostgres: false
api:
  env:
    postgresHostname: localhost
    postgresPort: 5432
    postgresDatabase: briefer
  secrets:
    postgresUsername: postgres
    postgresPassword: password

​
Using more than one replica for Briefers’ services

If you want to use more than one replica for Briefer’s services, you must set the replicas for each service in your values.yaml file.

After doing that, make sure to enable sticky sessions in your Ingress resources. Without sticky sessions (also known as Session Affinity), Briefer will not work correctly when using multiple replicas.

If you’re using the Nginx Ingress controller, for example, you can enable sticky sessions by adding the annotations below to your Ingress resources.

nginx.ingress.kubernetes.io/affinity: 'cookie'
nginx.ingress.kubernetes.io/session-cookie-name: 'http-cookie'
nginx.ingress.kubernetes.io/session-cookie-expires: '172800'
nginx.ingress.kubernetes.io/session-cookie-max-age: '172800'

It’s not necessary to use sticky sessions if you’re using only one replica.

​
Configuring pod sizes

We recommend that you explicitly configure the size of your jupyter pod so that it has enough memory and vCPUs to run your analyses.

Additionally, you can also set resources and limits for your api, web, and ai pods.

Below, you can see an example of how to configure your pod’s sizes.

# We recommend that you set these values
jupyter:
  resources:
    requests:
      cpu: 1
      memory: 4Gi
    limits:
      cpu: 1
      memory: 4Gi

# These ones usually aren't as important
api:
  resources:
    requests:
      cpu: 1
      memory: 4Gi
    limits:
      cpu: 1
      memory: 4Gi

web:
  resources:
    requests:
      cpu: 1
      memory: 2Gi
    limits:
      cpu: 1
      memory: 2Gi

ai:
  resources:
    requests:
      cpu: 0.5
      memory: 2Gi
    limits:
      cpu: 0.5
      memory: 2Gi

​
Running Jupyter pod on a specific node

Sometimes, it might be that you want to run your Jupyter pod on a specific node because it has more memory or CPUs available. In that case, you can use the affinity and tolerations options within jupyter in your values.yaml file, as show below.

jupyter:
  tolerations:
    - key: 'briefer.cloud/domain'
      operator: 'Equal'
      value: 'core'
      effect: 'NoSchedule'
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
          - matchExpressions:
              - key: briefer.cloud/domain
                operator: In
                values:
                  - jupyter

​
Disabling the AI service

Sometimes, you might not want to run the AI service to avoid sending requests to OpenAI. If that’s your case, you can disable the ai service by setting ai.enabled

ai:
  enabled: false

If you disable the AI service, you will not be able to use the “Edit with AI” and “Fix with AI” features.

​
Upgrading Briefer’s versions

The best way to upgrade your Briefer deployment is to install a newer Helm chart. All of Briefer’s Helm charts include image tags in their default values.yaml. Therefore, installing the latest published Helm chart should also cause your pods to use newer images. Here’s how to install Briefer’s most recent Helm chart.

  1. Pull the latest list of available charts from the briefercloud Helm repository. 
    helm repo update briefercloud
  2. List all the available versions 
    helm search repo briefercloud --versions
  3. Install the latest version or the desired version from the list. 
    helm install briefer briefercloud/briefer \
-f ./values.yaml \
--namespace briefer \
--create-namespace \
--version <YOUR_DESIRED_VERSION>

Alternatively, if you don’t want to install a newer Helm chart, you can just update the images tags’ within values.yaml, as shown below.

# We don't recommend upgrading images this way but it's possible
jupyter:
  image:
    repository: docker.io
    name: briefercloud/briefer-jupyter
    tag: <DESIRED_VERSION>

api:
  image:
    repository: docker.io
    name: briefercloud/briefer-web
    tag: <DESIRED_VERSION>

web:
  image:
    repository: docker.io
    name: briefercloud/briefer-api
    tag: <DESIRED_VERSION>

ai:
  image:
    repository: docker.io
    name: briefercloud/briefer-ai
    tag: <DESIRED_VERSION>

We don’t usually recommend users to change the image tag through values.yaml. That’s because newer versions of Briefer might require environment variables that are not passed to the pod’s in older versions of the chart. Still, this might be useful for quick changes sometimes.

GCE Virtual Machine (GCP)Troubleshooting