Plugin Generator
Plugins allow you to provide your own generator.
- You can write in any language
- Simple: a plugin just responds to RPC HTTP requests.
- You can use it in a sidecar, or standalone deployment.
- You can get your plugin running today, no need to wait 3-5 months for review, approval, merge and an Argo software release.
- You can combine it with Matrix or Merge.
To start working on your own plugin, you can generate a new repository based on the example applicationset-hello-plugin.
Simple example
Using a generator plugin without combining it with Matrix or Merge.
apiVersion: argoproj.io/v1alpha1
kind: ApplicationSet
metadata:
name: myplugin
spec:
generators:
- plugin:
# Specify the configMap where the plugin configuration is located.
configMapRef:
name: my-plugin
# You can pass arbitrary parameters to the plugin. `input.parameters` is a map, but values may be any type.
# These parameters will also be available on the generator's output under the `generator.input.parameters` key.
input:
parameters:
key1: "value1"
key2: "value2"
list: ["list", "of", "values"]
boolean: true
map:
key1: "value1"
key2: "value2"
key3: "value3"
# You can also attach arbitrary values to the generator's output under the `values` key. These values will be
# available in templates under the `values` key.
values:
value1: something
# When using a Plugin generator, the ApplicationSet controller polls every `requeueAfterSeconds` interval (defaulting to every 30 minutes) to detect changes.
requeueAfterSeconds: 30
template:
metadata:
name: myplugin
annotations:
example.from.input.parameters: "{{ generator.input.parameters.map.key1 }}"
example.from.values: "{{ values.value1 }}"
# The plugin determines what else it produces.
example.from.plugin.output: "{{ something.from.the.plugin }}"
configMapRef.name
: AConfigMap
name containing the plugin configuration to use for RPC call.input.parameters
: Input parameters included in the RPC call to the plugin. (Optional)
Note
The concept of the plugin should not undermine the spirit of GitOps by externalizing data outside of Git. The goal is to be complementary in specific contexts. For example, when using one of the PullRequest generators, it’s impossible to retrieve parameters related to the CI (only the commit hash is available), which limits the possibilities. By using a plugin, it’s possible to retrieve the necessary parameters from a separate data source and use them to extend the functionality of the generator.
Add a ConfigMap to configure the access of the plugin
apiVersion: v1
kind: ConfigMap
metadata:
name: my-plugin
namespace: argocd
data:
token: "$plugin.myplugin.token" # Alternatively $<some_K8S_secret>:plugin.myplugin.token
baseUrl: "http://myplugin.plugin-ns.svc.cluster.local."
token
: Pre-shared token used to authenticate HTTP request (points to the right key you created in theargocd-secret
Secret)baseUrl
: BaseUrl of the k8s service exposing your plugin in the cluster.
Store credentials
apiVersion: v1
kind: Secret
metadata:
name: argocd-secret
namespace: argocd
labels:
app.kubernetes.io/name: argocd-secret
app.kubernetes.io/part-of: argocd
type: Opaque
data:
# ...
# The secret value must be base64 encoded **once**
# this value corresponds to: `printf "strong-password" | base64`
plugin.myplugin.token: "c3Ryb25nLXBhc3N3b3Jk"
# ...
Alternative
If you want to store sensitive data in another Kubernetes Secret
, instead of argocd-secret
, ArgoCD knows how to check the keys under data
in your Kubernetes Secret
for a corresponding key whenever a value in a configmap starts with $
, then your Kubernetes Secret
name and :
(colon) followed by the key name.
Syntax: $<k8s_secret_name>:<a_key_in_that_k8s_secret>
NOTE: Secret must have label
app.kubernetes.io/part-of: argocd
Example
another-secret
:
apiVersion: v1
kind: Secret
metadata:
name: another-secret
namespace: argocd
labels:
app.kubernetes.io/part-of: argocd
type: Opaque
data:
# ...
# Store client secret like below.
# Ensure the secret is base64 encoded
plugin.myplugin.token: <client-secret-base64-encoded>
# ...
HTTP server
A Simple Python Plugin
You can deploy it either as a sidecar or as a standalone deployment (the latter is recommended).
In the example, the token is stored in a file at this location : /var/run/argo/token
string-password
import json
from http.server import BaseHTTPRequestHandler, HTTPServer
with open("/var/run/argo/token") as f:
plugin_token = f.read().strip()
class Plugin(BaseHTTPRequestHandler):
def args(self):
return json.loads(self.rfile.read(int(self.headers.get('Content-Length'))))
def reply(self, reply):
self.send_response(200)
self.end_headers()
self.wfile.write(json.dumps(reply).encode("UTF-8"))
def forbidden(self):
self.send_response(403)
self.end_headers()
def unsupported(self):
self.send_response(404)
self.end_headers()
def do_POST(self):
if self.headers.get("Authorization") != "Bearer " + plugin_token:
self.forbidden()
if self.path == '/api/v1/getparams.execute':
args = self.args()
self.reply({
"output": {
"parameters": [
{
"key1": "val1",
"key2": "val2"
},
{
"key1": "val2",
"key2": "val2"
}
]
}
})
else:
self.unsupported()
if __name__ == '__main__':
httpd = HTTPServer(('', 4355), Plugin)
httpd.serve_forever()
Execute getparams with curl :
curl http://localhost:4355/api/v1/getparams.execute -H "Authorization: Bearer string-password" -d \
'{
"applicationSetName": "fake-appset",
"input": {
"parameters": {
"param1": "value1"
}
}
}'
Some things to note here:
- You only need to implement the calls
/api/v1/getparams.execute
- You should check that the
Authorization
header contains the same bearer value as/var/run/argo/token
. Return 403 if not - The input parameters are included in the request body and can be accessed using the
input.parameters
variable. - The output must always be a list of object maps nested under the
output.parameters
key in a map. generator.input.parameters
andvalues
are reserved keys. If present in the plugin output, these keys will be overwritten by the contents of theinput.parameters
andvalues
keys in the ApplicationSet’s plugin generator spec.
With matrix and pull request example
In the following example, the plugin implementation is returning a set of image digests for the given branch. The returned list contains only one item corresponding to the latest built image for the branch.
apiVersion: argoproj.io/v1alpha1
kind: ApplicationSet
metadata:
name: fb-matrix
spec:
goTemplate: true
generators:
- matrix:
generators:
- pullRequest:
github: ...
requeueAfterSeconds: 30
- plugin:
configMapRef:
name: cm-plugin
input:
parameters:
branch: "{{.branch}}" # provided by generator pull request
values:
branchLink: "https://git.example.com/org/repo/tree/{{.branch}}"
template:
metadata:
name: "fb-matrix-{{.branch}}"
spec:
source:
repoURL: "https://github.com/myorg/myrepo.git"
targetRevision: "HEAD"
path: charts/my-chart
helm:
releaseName: fb-matrix-{{.branch}}
valueFiles:
- values.yaml
values: |
front:
image: myregistry:{{.branch}}@{{ .digestFront }} # digestFront is generated by the plugin
back:
image: myregistry:{{.branch}}@{{ .digestBack }} # digestBack is generated by the plugin
project: default
syncPolicy:
automated:
prune: true
selfHeal: true
syncOptions:
- CreateNamespace=true
destination:
server: https://kubernetes.default.svc
namespace: "{{.branch}}"
info:
- name: Link to the Application's branch
value: "{{values.branchLink}}"
To illustrate :
The generator pullRequest would return, for example, 2 branches:
feature-branch-1
andfeature-branch-2
.The generator plugin would then perform 2 requests as follows :
curl http://localhost:4355/api/v1/getparams.execute -H "Authorization: Bearer string-password" -d \
'{
"applicationSetName": "fb-matrix",
"input": {
"parameters": {
"branch": "feature-branch-1"
}
}
}'
Then,
curl http://localhost:4355/api/v1/getparams.execute -H "Authorization: Bearer string-password" -d \
'{
"applicationSetName": "fb-matrix",
"input": {
"parameters": {
"branch": "feature-branch-2"
}
}
}'
For each call, it would return a unique result such as :
{
"output": {
"parameters": [
{
"digestFront": "sha256:a3f18c17771cc1051b790b453a0217b585723b37f14b413ad7c5b12d4534d411",
"digestBack": "sha256:4411417d614d5b1b479933b7420079671facd434fd42db196dc1f4cc55ba13ce"
}
]
}
}
Then,
{
"output": {
"parameters": [
{
"digestFront": "sha256:7c20b927946805124f67a0cb8848a8fb1344d16b4d0425d63aaa3f2427c20497",
"digestBack": "sha256:e55e7e40700bbab9e542aba56c593cb87d680cefdfba3dd2ab9cfcb27ec384c2"
}
]
}
}
In this example, by combining the two, you ensure that one or more pull requests are available and that the generated tag has been properly generated. This wouldn’t have been possible with just a commit hash because a hash alone does not certify the success of the build.