Skip to content

Commit

Permalink
📖 WSL/Windows setup docs, rework setup docs (#2080)
Browse files Browse the repository at this point in the history
- Move minikube environment setup docs to its own document.

- Document what works for setting up minikube, OLM and the Konveyor
  operator on WSL / Windows

- Move some other docs to `docs/`

- Rework the main README.md:
  - have a quick start section
  - move some details to `development.md`
  - move environment install docs to `local-minikube-setup.md`

---------

Signed-off-by: Scott J Dickerson <sdickers@redhat.com>
  • Loading branch information
sjd78 committed Sep 9, 2024
1 parent 7cd3ebb commit 55daf94
Show file tree
Hide file tree
Showing 8 changed files with 549 additions and 145 deletions.
195 changes: 58 additions & 137 deletions README.md

Large diffs are not rendered by default.

12 changes: 9 additions & 3 deletions development.md → docs/development.md
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,12 @@ This document describes the process for running and developing tackle2-ui on you

Tackle2-ui can be developed using macOS(x86_64 only) or Linux environments.

## File Naming Conventions

- Use kebab-case for file names.
- The root page/parent level components are placed directly in their respective directories.
- Presentation layer components are placed within the `components/` subdirectory of the parent component.

## React hooks

Our project utilizes [Hooks](https://reactjs.org/docs/hooks-intro.html) wherever possible as this pattern has been proven and settled on in the react community. The hooks pattern is highly testable and provides a clear way to reuse stateful logic without overcomplicating components with complex lifecycle methods. Overall, we have largely left class components behind and have adopted functional components / hooks as the way forward.
Expand All @@ -17,7 +23,7 @@ For handling spacing/layout requirements that do not fit the standard PF mold, w

## Form development

We are using [react-hook-form](https://react-hook-form.com) in tandem with [patternfly](https://www.patternfly.org). Custom wrapper components have been developed to aid with this integration and their usage can be referenced in the [proxy-form](./client/src/app/pages/proxies/proxy-form.tsx) component.
We are using [react-hook-form](https://react-hook-form.com) in tandem with [patternfly](https://www.patternfly.org). Custom wrapper components have been developed to aid with this integration and their usage can be referenced in the [proxy-form](/client/src/app/pages/proxies/proxy-form.tsx) component.

### Steps to create a form

Expand Down Expand Up @@ -62,5 +68,5 @@ export interface FormValues {

For more info about working on the Tackle 2.x UI, check out these READMEs:

- [tests.md](./tests.md)
- [internationalization.md](./INTERNATIONALIZATION.md)
- [tests.md](/docs/tests.md)
- [internationalization.md](/INTERNATIONALIZATION.md)
237 changes: 237 additions & 0 deletions docs/local-minikube-setup.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,237 @@
# Setting up Konveyor on a local minikube instance

The preferred local development for the tackle2-ui is to install the Konveyor
operator on a minikube instance. When the operator is installed and running,
the UI development server can attach to it and use the operator's hub api. There
will also be a running UI instance to work with.

The setup process follows the general steps:

- install minikube
- start minikube with the **dashboard** and **ingress** addons
- install the operator framework (OLM)
- install the Konveyor operator
- optionally add test data

## Install and start minikube

[Minikube](https://github.com/kubernetes/minikube) implements a local Kubernetes cluster
and is available on macOS, Linux, and Windows. All you need to run minikube is a container
runtime ([Docker](https://docs.docker.com/engine/install/) or [podman](https://podman.io/docs/installation))
or a Virtual Machine environment.

See the minikube [getting started guide](https://minikube.sigs.k8s.io/docs/start/) for install
details for your platform.

By default, Minikube uses a [driver](https://minikube.sigs.k8s.io/docs/drivers/) with 6 GB
of memory and 2 CPUs. This may not be enough to run all the services, so we need to increase
the allocated memory. In our experience, 10 GB of memory and 4 CPUs is fine:

```sh
minikube config set memory 10240
minikube config set cpus 4
```

> [!IMPORTANT]
> Depending on your driver, administrator access may be required. Common driver choices are
> Docker for container-based virtualization and KVM for hardware-assisted virtualization on Linux
> systems. If you're not sure which driver is best for you or if you're encountering compatibility
> issues, minikube supports auto-selecting a driver based on your system's capabilities and
> installed software.
Start minikube with the command:

```sh
minikube start --addons=dashboard --addons=ingress
```

> [!NOTE]
> We need to enable the **dashboard** and **ingress** addons. The dashboard addon installs
> the dashboard service that provides a web UI for the Kubernetes objects. The ingress addon
> allows us to create Ingress CRs to expose the Tackle UI and Tackle Hub API.
## Install OLM

Since Konveyor is deployed as an operator, the [operator framework](https://operatorframework.io/)'s
[operator lifecycle manager (OLM)](https://olm.operatorframework.io/) needs to be installed
on our new minikube instance before installing the Konveyor operator.

Since the minikube olm addon is disabled until OLM issue
[2534](https://github.com/operator-framework/operator-lifecycle-manager/issues/2534)
is resolved we need to install the [OLM manually](https://github.com/operator-framework/operator-lifecycle-manager/releases).

For version `v0.28.0` (latest version as of 17-May-2024) use the scripts:

```sh
curl -L https://github.com/operator-framework/operator-lifecycle-manager/releases/download/v0.28.0/install.sh -o install.sh
chmod +x install.sh
./install.sh v0.28.0
```

## Install the Konveyor Operator

There are a few good ways to install the Konveyor operator:

- Use the script [`hack/setup-operator.sh`](/hack/setup-operator.sh). It is a local variation of
the script from the operator that applies all CRs needed to install the Konveyor operator and
setup the Tackle instance. Use of this script is described in the next section.

- Follow the official instructions for [Installing Konveyor Operator](https://konveyor.github.io/konveyor/installation/#installing-konveyor-operator)

- Use one of the `hack/install-*` scripts in the [konveyor/operator](https://github.com/konveyor/operator)
repository. These scripts require `kubectl` and the `operator-sdk`.

### Running the setup script

To run the setup script without cloning the repo, make sure the `kubectl` command (not alias)
is working and configured for minikube instance, and that the bash shell is available:

```sh
curl https://raw.githubusercontent.com/konveyor/tackle2-ui/main/hack/setup-operator.sh -o setup-operator.sh
chmod +x setup-operator.sh
./setup-operator.sh
```

You may also run the script directly from you tackle2-ui clone:

```sh
cd tackle2-ui/hack
./setup-operator.sh
```

> [!WARNING]
> While CRDs are being established, you may see the script output `NotFound` errors.
> You can safely ignore these. The script will wait and recheck for the CRD again before
> proceeding.
The installation is complete when the script outputs "condition met" messages and terminates.

### Customizing the setup script

The setup script provides optional environment variables that may be used to customize the
images used, settings, and features enabled during the install.

Configuration environment variable include:
| variable | default | description |
| --- | --- | --- |
| TACKLE*CR | *(empty)\_ | Allows specifying the full Tackle CR |
| ADDON_ANALYZER_IMAGE | quay.io/konveyor/tackle2-addon-analyzer:latest | image for the ADDON_ANALYZER pod |
| ANALYZER_CONTAINER_REQUESTS_CPU | 0 | cpu count for the analyzer (0 is no restriction) |
| ANALYZER_CONTAINER_REQUESTS_MEMORY | 0 | memory size for the analyzer (0 is no restriction) |
| FEATURE_AUTH_REQUIRED | false | include keycloak SSO in operator deployment |
| HUB_BUCKET_VOLUME_SIZE | 100Gi | PV claim size for the HUB buckets |
| HUB_DATABASE_VOLUME_SIZE | 10Gi | PV claim size for the HUB database |
| HUB_IMAGE | quay.io/konveyor/tackle2-hub:lates | image for the HUB pod |
| IMAGE_PULL_POLICY | Always | When should images needed by the operator be pulled **link to docs for the pull policy** |
| NAMESPACE | konveyor-tackle | kubernetes namespace used |
| OPERATOR_INDEX_IMAGE | quay.io/konveyor/tackle2-operator-index:latest | base operator container image |
| UI_IMAGE | quay.io/konveyor/tackle2-ui:latest | image for the UI pod |
| UI_INGRESS_CLASS_NAME | nginx | nginx is the used as the ingress for minikube to access the UI |

For example, if you wish to run tackle with keycloak authentication enabled, export the following variable before running the install script:

```sh
export FEATURE_AUTH_REQUIRED=true
./setup-operator.sh
```

If the `TACKLE_CR` variable is defined, its contents will be used to create the `Tackle`
instance in lieu of one built in the script based on the other config variables. For
example, this will create a Tackle instance with authentication turned on and the UI
using a specially built test image:

```sh
export TACKLE_CR=$(cat <<EOF
kind: Tackle
apiVersion: tackle.konveyor.io/v1alpha1
metadata:
name: tackle
spec:
feature_auth_required: true
ui_image_fqin: quay.io/sdickers/tackle2-ui:test-new-feature
EOF
)
./setup-operator.sh
```

## Optional Steps...

### Considerations for MacOS

Some core utils used in install and setup scripts may not be available by default on
a Mac. They are available to install via [Homebrew](https://brew.sh/) as the `coreutils`
package:

```sh
brew install coreutils
```

### Sample data from tackle2-hub

The [tackle2-hub repo](https://github.com/konveyor/tackle2-hub) has some scripts to add
basic entities to a Konveyor instance. Using the scripts is a reasonable way to get a
base setup. Importing that data is [described here](/hack/README.md#adding-a-base-set-of-data-to-an-empty-instance).

### Configuring kubectl for minikube

You will need `kubectl` on your PATH and configured to control minikube in order to proceed.
There are a few ways to set this up:

1. **Install kubectl yourself**

If you already [have the `kubectl` CLI tool installed](https://kubernetes.io/docs/tasks/tools/#kubectl)
and available on your PATH, the `minikube start` command should configure it to control the minikube
cluster. You should see the following message when minikube starts if this worked correctly:

```
🏄 Done! kubectl is now configured to use "minikube" cluster and "default" namespace by default
```

2. **Use a shell alias for minikube's built-in kubectl**

Minikube provides its own internal `kubectl` which you can use by running `minikube kubectl --`
followed by your CLI arguments. If you want to use the built-in `minikube kubectl` as the `kubectl`
on your shell, you can setup a shell alias.

The following example shows how to setup an alias for bash on Fedora:

```sh
mkdir -p ~/.bashrc.d
cat << EOF > ~/.bashrc.d/minikube
alias kubectl="minikube kubectl --"
EOF
source ~/.bashrc
```
> [!WARNING]
> The alias will only work for an interactive shell. If you want to use `minikube kubectl --`
> as the command in scripts, you'll need to use a shim shell script.
3. **Use a shim shell script to route kubectl to minikube**
To be able to use `minikube kubectl --` as the command `kubectl` in general use (interactive
prompt and in other scripts), a simple shell script can be used. Any location that is in
the `$PATH` can be used. Installing for all system users can typically be done to `/usr/local/bin`:
```sh
cat << EOF > kubectl.sh
#! /usr/bin/env bash
minikube kubectl -- $@
EOF
sudo install -m 777 kubectl.sh /usr/local/bin/kubectl
```
## Starting Over
If at any point your minikube Konveyor operator instance stops working properly, it
is simple to destroy the environment and rebuild.
To destroy an existing minikube instance (with Konveyor installed):
```sh
minikube stop
minikube delete
```
Rebuild by starting a new instance of minikube and following the rest of the
install steps in this document.
File renamed without changes.
10 changes: 5 additions & 5 deletions hack/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,13 +7,13 @@ creating/importing various pieces of data.

For creating a base set of data in a tackle-hub instance, a good place to start
is [tackle-hub's hack folder](https://github.com/konveyor/tackle2-hub/tree/main/hack).
If you have hub running or port forwarded on port :9002, a basic set of data can be
added to the instance by:
If you have the Konveyor operator running on minikube without authentication, a basic
set of data can be added to the instance by:

```sh
> git clone https://github.com/konveyor/tackle2-hub.git
> cd tackle2-hub/hack
> HOST=localhost:9002 ./add/all.sh
git clone https://github.com/konveyor/tackle2-hub.git
cd tackle2-hub/hack
HOST=http://$(minikube ip)/hub ./add/all.sh
```

## Useful commands
Expand Down
Loading

0 comments on commit 55daf94

Please sign in to comment.