Connecting a local application to a Kubernetes cluster (Local Mode)

Glossary

• Local app - a locally running application that can be used instead of a Deploy action with a local mode configuration.

• Target kubernetes workload - a Kubernetes Workload produced from a Deploy action with a local mode configuration that's deployed to the target kubernetes cluster.

Development status

This feature is still experimental. We're still working on some open tasks to improve the feature stability and usability. It means that:

• some incompatible changes can be made until the first non-experimental release

• there are some functional limitations, see the Current limitations section below

Introduction

A local application can be used instead of a kubernetes workload by adding a local mode configuration to the Deploy action

Local mode feature is only supported by certain action types and providers.

Supported action types

• container

• kubernetes

• helm

Supported providers

• kubernetes
• local kubernetes

Pre-requisites

Local mode uses kubectl port-forwarding and plain SSH port forwarding under the hood.

Requirements for the local machine environment:

• OpenSSH 7.6 or higher

• Kubectl

Current limitations

There is a number of functional limitations in the current version.

Reachability of the underlying workloads

The best matching use-case for local mode is to locally run an "isolated" application, i.e. an application that does not make any calls to other resources in the kubernetes cluster (i.e. databases or other applications).

If your application makes HTTP calls to some other kubernetes resources using kubernetes DNS names, then such calls will fail because the local DNS configuration is not aware about any DNS names configured in the kubernetes cluster.

A concrete example can be found in the local-mode project.

Compatibility with sync mode

A Deploy action cannot be running in local and sync modes simultaneously. Local mode always takes precedence over sync mode if both are configured in the relevant garden.yml configuration file and if both --local and --sync flags are enabled.

Windows compatibility

The local mode is not supported natively for Windows OS. It should be used with WSL in Windows environments.

Number of the containers running in local mode

Only one container can be run in local mode for each kubernetes or helm Deploy action.

Cluster state on exit

The local mode leaves the proxy container deployed in the target kubernetes cluster after exit. The affected Deploys must be re-deployed manually by using garden deploy.

How it works

Usually, a Garden Deploy action declares a configuration and a deployment policy of a kubernetes workload. A typical deployment flow looks like this:

1. build a container image if necessary

2. configure a docker container for the image

3. prepare kubernetes workloads to be deployed

4. deploy the configured kubernetes workloads to the target kubernetes cluster

The local mode changes the usual deployment flow by changing the manifest configuration step (item 3 from the list above) and the deployment step (item 4).

Changes in workload configuration

Local mode does the following modifications in the target kubernetes workload configuration before the actual deployment:

1. Replaces target kubernetes workload's container with a special proxy container that is based on openssh-server. This container exposes its SSH port and the same HTTP ports as the Deploy action configured in local mode.
2. Sets the number of replicas of the target kubernetes workload is always set to 1.

3. Disables the basic health-checks (startup, readiness and liveness probes). See the section below for details.

Health-checks

The startup, readiness and liveness probes are disabled for all Deploy axtions running in local mode. This has been done because of some technical reasons.

The lifecycle of a local app can be completely controlled by a user. Thus, the health checks may be unwanted and obstructing.

The kubernetes cluster readiness checks are applied to a proxy container which sends the traffic to the local app. When a readiness probe happens, the target local app and the relevant port forward are not ready yet. Thus, the readiness probe can cause the failure of the local mode startup.

The liveness checks can cause unnecessary re-deployment of the proxy container in the target cluster. Also, those checks create some extra traffic to the local app. That might be noisy and unnecessary if the local app is running in the debugger.

Changes in deployment execution

Once the kubernetes workloads are configured, local mode executes the deployment step in a specific way:

1. The local app is started by Garden if localMode.command field is specified in the Deploy action configuration. Otherwise, the local app should be started manually.

2. The SSH port forwarding from a randomly assigned local port to the proxy container SSH port is initialized by means of kubectl port-forward command.

3. The reverse port forwarding (on top of the previous SSH port forwarding) between the remote proxy container's HTTP port and the local application HTTP port is established by means of ssh command.

As a result, the original target kubernetes workload is replaced by a workload that runs a proxy container, let's call it a proxy-workload. The proxy-workload is connected with the local app via the 2-layered port-forwarding described above. This connection schema allows to route the target kubernetes workload's traffic to the local app and back. For the rest entities in the kubernetes cluster, the local app acts as an original kubernetes workload.

In order to maintain secure connections, Garden generates a new SSH key pair for each Deploy action running in local mode on every CLI execution.

Note! Garden automates the SSH key acceptance with option -oStrictHostKeyChecking=accept-new, this is the reason why you need OpenSSH 7.6 or higher.

Configuration

To configure a Deploy action for local mode, add localMode configuration entry to the spec field of your Deploy action configuration. See the examples below for details.

Configuring local mode for container action type

kind: Build
name: node-app
type: container

---

kind: Deploy
name: node-app
type: container
dependencies:
  - build.node-app
...
spec:
  image: ${actions.build.node-app.outputs.deploymentImageId}
  localMode:
    ports:
      - local: 8090 # The port of the local app, will be used for port-forward setup.
        remote: 8080 # The port of the remote app, will be used for port-forward setup.
    # Starts the local app which will replace the target one in the kubernetes cluster.
    # Optional. If not specified, then the local app should be started manually.
    command: [ npm, run, serve ]
    # Defines how to restart the local app on failure/exit.
    # Optional. If not specified, then the default values will be applied.
    restart:
      delayMsec: 2000 # 2 sec delay between local app restarts
      max: 100 # limit restart attempts to 100
    ...
...

An example can be found in the local-mode project.

Configuring local mode for kubernetes and helm action types

kind: Build
name: backend
type: container

---

kind: Deploy
name: backend
type: kubernetes # this example looks the same for helm actions (i.e. with `type: helm`)
build: backend
dependencies:
  - build.backend
localMode:
  ports:
    - local: 8090
      remote: 8080
  command: [ "../backend-local/main" ]
  # Target resource selector is necessary for `kubernetes` and `helm` action types
  target:
    kind: Deployment
    name: backend-deployment
    containerName: backend
...
# manifests or files

A kubernetes example can be found in the local-mode-k8s project. A helm example can be found in the local-mode-helm project.

Deploying with local mode

To run your Deploy actions in local mode, you can use deploy command with the special flag:

# Deploy specific Deploys in local mode:
garden deploy --local=app-1
garden deploy --local=app-1 app-2

# Deploy all applicable Deploys in local mode:
garden deploy --local
garden deploy --local=*

Local mode always runs in persistent mode, it means that the Garden process won't exit until it's terminated explicitly. All port-forwards established by local mode will be stopped on the process exit. The local application will be stopped if it was started via the localMode.command configuration option. Otherwise, if the local application was started manually, it will continue running.

Watching the local application's logs

If you run your local application with the localMode.command configuration option, then you can easily watch the local application's logs in real-time by running a garden command with verbose log level:

garden deploy --local -l 3

Otherwise, you can find the logs in .garden/deploy.debug.*.log files.