Run the orchestrator deploy as the third job of the lab workflow

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-10-01 17:02:07 +01:00
parent f59c51a823
commit ff767dbfdc
3 changed files with 125 additions and 167 deletions

View File

@@ -1,8 +1,10 @@
name: Build and Push to ACR (lab)
name: Build, Render and Deploy (lab)
# Dispatch only while the lab path is proven. Builds the image for the dispatched commit, then
# renders score-lab.yaml with the platform's provisioners and commits the result to deploy/lab,
# which Argo CD delivers.
# Dispatch only while the lab path is proven. Three jobs for the dispatched commit:
# build the image
# render score-lab.yaml with the platform's provisioners, committed to deploy/lab for Argo CD
# deploy score-lab.yaml to the orchestrator, which keeps the Argo CD Application and the
# public route in place
on:
workflow_dispatch: {}
@@ -163,3 +165,114 @@ jobs:
echo "::error::${{ gitea.ref_name }} moved while this job ran; run the workflow again"
exit 1
}
deploy:
name: Deploy to Orchestrator
needs: render
runs-on: ubuntu-latest
env:
PO_API_URL: https://api.dev.orchestrator.crucible.kyndemo.live
PO_ORG_ID: crucible
PO_AUTH_TOKEN: ${{ secrets.PO_AUTH_TOKEN }}
# One orchestrator project per application, bound to the lab HUB's runner: what the
# orchestrator creates is the Argo CD Application and the public route, and both live on the
# hub. Argo CD delivers the workload to the spoke from deploy/lab.
PROJECT_ID: demo-vivek-11-lab
PO_RUNNER_ID: crucible-orchestrator-dev-platform-lab-runner
OCTL_VERSION: 1.0.0
IMAGE: cruciblecoredevacr.azurecr.io/demo-vivek-11
steps:
- name: Checkout code
uses: actions/checkout@v4
- name: Install octl
run: |
set -euo pipefail
curl -fsSLo /tmp/octl.tar.gz \
"https://github.com/stellwerk-labs/platform-orchestrator-cli/releases/download/v${OCTL_VERSION}/platform-orchestrator-cli_${OCTL_VERSION}_linux_amd64.tar.gz"
tar xzf /tmp/octl.tar.gz -C /tmp
install -m 755 /tmp/octl /usr/local/bin/octl
octl --version
- name: Set environment
run: |
# One environment, of type `lab`. The type is what selects the Argo CD implementations:
# the workload becomes an Application and workload-ingress becomes the hub route.
echo "ENV_ID=lab" >> $GITHUB_ENV
- name: Ensure the project and its runner binding exist
run: |
set -euo pipefail
# Created on first deploy rather than by a pull request against config/projects.tf.
# None of this needs a repository change: project, runner rule and environment are
# all runtime objects.
#
# Neither create is idempotent, so both fall through to a read on the second run.
octl create project "$PROJECT_ID" \
--set display_name='demo-vivek-11 (lab)' || \
octl get project "$PROJECT_ID"
# The runner rule is what routes this project's deployments to the apps cluster.
# Creating it twice would leave two rules matching the same project, so it is
# created only when absent -- `create` would happily add a duplicate.
if octl get runner-rules -o json 2>/dev/null | grep -q "\"project_id\": *\"$PROJECT_ID\""; then
echo "Runner rule for '$PROJECT_ID' already exists."
else
octl create runner-rule \
--set project_id="$PROJECT_ID" \
--set runner_id="$PO_RUNNER_ID" \
--no-prompt
fi
- name: Ensure the environment exists
run: |
set -euo pipefail
octl create environment "$PROJECT_ID" "$ENV_ID" \
--set env_type_id=lab \
--set display_name="Lab" || \
octl get environment "$PROJECT_ID" "$ENV_ID"
- name: Deploy the Score workload
run: |
set -euo pipefail
# `octl score deploy` is ADDITIVE — it adds or updates a workload in the manifest and
# never removes one. That is the opposite of `octl deploy`, where omission is
# deletion. Do not substitute one for the other.
# No --show-logs: octl 1.0.0 has no such flag and exits 1 with `unknown flag`
# BEFORE contacting the orchestrator, so the whole deploy dies on an argument
# typo. Its nearest relatives are --runner-logs-level (default `info`, already
# what we want) and --skip-logs (which suppresses storage). Neither streams the
# runner's logs into this job, so there is nothing to substitute -- the runner
# logs are read from the orchestrator, not from here.
# The tag must be the commit the BUILD built, and it must be the WHOLE sha.
#
# build-push.yml tags with `` -- all 40 characters -- so the
# 7-character `${GITHUB_SHA:0:7}` this used to pass named a tag that has never
# existed in the registry.
#
# And on a workflow_run, GITHUB_SHA is the DEFAULT branch's head, while the build
# that produced the image ran on `dev`. They coincide only while the branches are
# level. `workflow_run.head_sha` is the triggering run's own commit, which is by
# definition the one that was built; `github.sha` covers the workflow_dispatch case,
# where there is no triggering run.
IMAGE_TAG="${{ gitea.sha }}"
echo "Deploying ${IMAGE}:${IMAGE_TAG}"
octl score deploy "$PROJECT_ID" "$ENV_ID" score-lab.yaml \
--default-image "${IMAGE}:${IMAGE_TAG}" \
--no-prompt
- name: Deployment summary
if: always()
run: |
# Same commit the deploy step resolved, abbreviated for reading only -- the
# deployed tag is the full sha.
DEPLOYED_SHA="${{ gitea.sha }}"
SHORT_SHA="${DEPLOYED_SHA:0:7}"
echo "## Deployment Result" >> $GITHUB_STEP_SUMMARY
echo "| Field | Value |" >> $GITHUB_STEP_SUMMARY
echo "|---|---|" >> $GITHUB_STEP_SUMMARY
echo "| Project | \`$PROJECT_ID\` |" >> $GITHUB_STEP_SUMMARY
echo "| Environment | \`$ENV_ID\` |" >> $GITHUB_STEP_SUMMARY
echo "| Commit | \`$SHORT_SHA\` |" >> $GITHUB_STEP_SUMMARY
echo "[View in Orchestrator Console](https://console.dev.orchestrator.crucible.kyndemo.live/orgs/$PO_ORG_ID/projects/$PROJECT_ID/environments/$ENV_ID)" >> $GITHUB_STEP_SUMMARY

View File

@@ -1,157 +0,0 @@
name: Deploy to Orchestrator (lab)
# Creates or updates the Argo CD Application and the hub route for the lab, from score-lab.yaml.
# Run after "Build and Push to ACR (lab)", which renders deploy/lab. Dispatch only.
on:
workflow_dispatch: {}
env:
PO_API_URL: https://api.dev.orchestrator.crucible.kyndemo.live
PO_ORG_ID: crucible
PO_AUTH_TOKEN: ${{ secrets.PO_AUTH_TOKEN }}
# ONE ORCHESTRATOR PROJECT PER APPLICATION.
#
# This used to be the shared `apps-cluster` project with one environment per app, which
# made every app a peer of every other: the Orchestrator tab on any component listed the
# entire estate, and an app had exactly one environment named after itself, so there was
# nowhere for dev/staging/prod to live.
#
# That shape existed to avoid a Terraform pull request against config/projects.tf for every
# scaffolded app. That constraint turned out not to be real -- `octl create project`,
# `octl create runner-rule` and `octl create environment` are all runtime operations, so
# the workflow below builds the whole thing on first deploy and needs no repository change.
# Its own project, NOT demo-vivek-11. Sharing it would reuse deploy.yml's runner rule (the dev
# apps runner) and overwrite the dev workload with this file.
#
# Bound to the lab HUB's runner: what the orchestrator creates here is the Argo CD Application
# and the public route, and both live on the hub. Argo CD delivers the workload to the spoke
# from deploy/lab.
PROJECT_ID: demo-vivek-11-lab
# The runner a project's workloads execute on. This is NOT cosmetic: the Kubernetes and
# Helm providers are ambient, so a workload lands in whichever cluster its runner lives in,
# and Terraform state is keyed per runner. A project bound to the wrong runner deploys to
# the wrong cluster, and repointing it afterwards orphans the state it already owns.
PO_RUNNER_ID: crucible-orchestrator-dev-platform-lab-runner
OCTL_VERSION: 1.0.0
IMAGE: cruciblecoredevacr.azurecr.io/demo-vivek-11
jobs:
guard:
name: Platform guard
runs-on: ubuntu-latest
outputs:
ready: ${{ steps.check.outputs.ready }}
steps:
- uses: actions/checkout@v4
- name: Check platform initialized
id: check
run: |
if [ -f ".platform/initialized.md" ]; then
echo "ready=true" >> $GITHUB_OUTPUT
else
echo "ready=false" >> $GITHUB_OUTPUT
echo "Skipping: .platform/initialized.md not found"
fi
deploy:
name: Deploy to Orchestrator
needs: guard
if: >-
(github.event_name == 'workflow_run' && github.event.workflow_run.conclusion == 'success' && needs.guard.outputs.ready == 'true') ||
(github.event_name == 'workflow_dispatch')
runs-on: ubuntu-latest
steps:
- name: Checkout code
uses: actions/checkout@v4
- name: Install octl
run: |
set -euo pipefail
curl -fsSLo /tmp/octl.tar.gz \
"https://github.com/stellwerk-labs/platform-orchestrator-cli/releases/download/v${OCTL_VERSION}/platform-orchestrator-cli_${OCTL_VERSION}_linux_amd64.tar.gz"
tar xzf /tmp/octl.tar.gz -C /tmp
install -m 755 /tmp/octl /usr/local/bin/octl
octl --version
- name: Set environment
run: |
# One environment, of type `lab`. The type is what selects the Argo CD implementations:
# the workload becomes an Application and workload-ingress becomes the hub route.
echo "ENV_ID=lab" >> $GITHUB_ENV
- name: Ensure the project and its runner binding exist
run: |
set -euo pipefail
# Created on first deploy rather than by a pull request against config/projects.tf.
# None of this needs a repository change: project, runner rule and environment are
# all runtime objects.
#
# Neither create is idempotent, so both fall through to a read on the second run.
octl create project "$PROJECT_ID" \
--set display_name='demo-vivek-11 (lab)' || \
octl get project "$PROJECT_ID"
# The runner rule is what routes this project's deployments to the apps cluster.
# Creating it twice would leave two rules matching the same project, so it is
# created only when absent -- `create` would happily add a duplicate.
if octl get runner-rules -o json 2>/dev/null | grep -q "\"project_id\": *\"$PROJECT_ID\""; then
echo "Runner rule for '$PROJECT_ID' already exists."
else
octl create runner-rule \
--set project_id="$PROJECT_ID" \
--set runner_id="$PO_RUNNER_ID" \
--no-prompt
fi
- name: Ensure the environment exists
run: |
set -euo pipefail
octl create environment "$PROJECT_ID" "$ENV_ID" \
--set env_type_id=lab \
--set display_name="Lab" || \
octl get environment "$PROJECT_ID" "$ENV_ID"
- name: Deploy the Score workload
run: |
set -euo pipefail
# `octl score deploy` is ADDITIVE — it adds or updates a workload in the manifest and
# never removes one. That is the opposite of `octl deploy`, where omission is
# deletion. Do not substitute one for the other.
# No --show-logs: octl 1.0.0 has no such flag and exits 1 with `unknown flag`
# BEFORE contacting the orchestrator, so the whole deploy dies on an argument
# typo. Its nearest relatives are --runner-logs-level (default `info`, already
# what we want) and --skip-logs (which suppresses storage). Neither streams the
# runner's logs into this job, so there is nothing to substitute -- the runner
# logs are read from the orchestrator, not from here.
# The tag must be the commit the BUILD built, and it must be the WHOLE sha.
#
# build-push.yml tags with `` -- all 40 characters -- so the
# 7-character `${GITHUB_SHA:0:7}` this used to pass named a tag that has never
# existed in the registry.
#
# And on a workflow_run, GITHUB_SHA is the DEFAULT branch's head, while the build
# that produced the image ran on `dev`. They coincide only while the branches are
# level. `workflow_run.head_sha` is the triggering run's own commit, which is by
# definition the one that was built; `github.sha` covers the workflow_dispatch case,
# where there is no triggering run.
IMAGE_TAG="${{ github.event.workflow_run.head_sha || github.sha }}"
echo "Deploying ${IMAGE}:${IMAGE_TAG}"
octl score deploy "$PROJECT_ID" "$ENV_ID" score-lab.yaml \
--default-image "${IMAGE}:${IMAGE_TAG}" \
--no-prompt
- name: Deployment summary
if: always()
run: |
# Same commit the deploy step resolved, abbreviated for reading only -- the
# deployed tag is the full sha.
DEPLOYED_SHA="${{ github.event.workflow_run.head_sha || github.sha }}"
SHORT_SHA="${DEPLOYED_SHA:0:7}"
echo "## Deployment Result" >> $GITHUB_STEP_SUMMARY
echo "| Field | Value |" >> $GITHUB_STEP_SUMMARY
echo "|---|---|" >> $GITHUB_STEP_SUMMARY
echo "| Project | \`$PROJECT_ID\` |" >> $GITHUB_STEP_SUMMARY
echo "| Environment | \`$ENV_ID\` |" >> $GITHUB_STEP_SUMMARY
echo "| Commit | \`$SHORT_SHA\` |" >> $GITHUB_STEP_SUMMARY
echo "[View in Orchestrator Console](https://console.dev.orchestrator.crucible.kyndemo.live/orgs/$PO_ORG_ID/projects/$PROJECT_ID/environments/$ENV_ID)" >> $GITHUB_STEP_SUMMARY

View File

@@ -14,15 +14,17 @@ browser ─► hub pe-gatewayapi (demo-vivek-11.apps.lab…) ─► Private Link
| Argo CD Application `team-demo/demo-vivek-11` | lab **hub** | the orchestrator, from the workload in `score-lab.yaml` |
| Public route (Backend + HTTPRoute) | lab **hub** | the orchestrator, from the `workload-ingress` resource |
## Workflows
## Workflow
1. **Build and Push to ACR (lab)** builds the image, renders `score-lab.yaml` into `lab/` and commits
it. The image tag is the commit that was built. Argo CD then rolls it out.
2. **Deploy to Orchestrator (lab)** creates or updates the Application and the public route. It is
needed the first time and whenever the hostname, team or repository in `score-lab.yaml` changes.
**Build, Render and Deploy (lab)** runs three jobs for the dispatched commit:
1. **build** the image, tagged with the commit.
2. **render** `score-lab.yaml` into `lab/` and commit it. Argo CD then rolls it out.
3. **deploy** `score-lab.yaml` to the orchestrator, which keeps the Application and the public route
in place. It changes something only when the hostname, team or repository changes.
`lab/` is **generated**: never edit it. To change what is deployed, change `score-lab.yaml` and run
the build workflow. The provisioners version is pinned in that workflow.
the workflow. The provisioners version is pinned in that workflow.
## What `score-lab.yaml` must carry