Day 10: Environments, OIDC and Updates
Goal: Create an environment
produktionwith approval and a secret of its own, fetch an OIDC token from the run and read its claims, and let Dependabot keep the dependencies up to date. · Duration: approx. 75 min · Prerequisite: Day 9
Briefing
- An environment is a target such as
produktion. It can require an approval, keep secrets for itself only and define which branches may deploy to it. A job selects it withenvironment:. - OIDC: With
id-token: write, a job fetches a signed JWT from GitHub. It states who the job is: repository, branch, environment. Clouds and vaults exchange it for short-lived credentials – without a stored secret. - Dependabot looks for new versions of Go modules,
actions and base images and opens pull requests for them. Configured in
.github/dependabot.yml.
deploy (environment: produktion)
─► waits ─► approval ─► secret DEPLOY_TOKEN
└► OIDC token: sub = repo:…:environment:produktion
Drill 1: An environment with approval
The approval should come from you. For that, the API needs your numeric ID:
cd ~/gha-lab/notes-actions
gh api user --jq .idExpected (your number differs):
251214518
The rules as JSON – replace the number with yours:
cat > ~/gha-lab/env-produktion.json <<'EOF'
{
"reviewers": [ { "type": "User", "id": 251214518 } ],
"prevent_self_review": false,
"deployment_branch_policy": {
"protected_branches": true,
"custom_branch_policies": false
}
}
EOF
gh api -X PUT repos/{owner}/{repo}/environments/produktion \
--input ~/gha-lab/env-produktion.json --jq '.name, [.protection_rules[].type]'Expected:
produktion
["required_reviewers","branch_policy"]
reviewers: who may approve – people or teams.prevent_self_review: false: whoever triggered the run may approve it themselves. Alone in the repository there is no other way; in a team you set it totrue.protected_branches: true: only protected branches may deploy to this environment – for you that ismain, via the ruleset from Day 4.
A secret just for this environment:
gh secret set DEPLOY_TOKEN --env produktion --body lab-deploy-token-123
gh secret list --env produktionExpected:
NAME UPDATED
DEPLOY_TOKEN less than a minute ago
gh never shows the value – even you can only overwrite a
secret, not read it.
Save your progress? Sign up for free
Drill 2: A deployment that waits
cat > .github/workflows/deploy.yml <<'EOF'
name: deploy
on: workflow_dispatch
permissions:
contents: read
jobs:
deploy:
runs-on: ubuntu-24.04
environment: produktion
permissions:
contents: read
id-token: write
steps:
- name: Secret nutzen
env:
DEPLOY_TOKEN: ${{ secrets.DEPLOY_TOKEN }}
run: |
echo "Token: $DEPLOY_TOKEN"
echo "Länge: ${#DEPLOY_TOKEN}"
- name: OIDC-Token
run: >-
curl -fsS -H "Authorization: bearer $ACTIONS_ID_TOKEN_REQUEST_TOKEN"
"$ACTIONS_ID_TOKEN_REQUEST_URL&audience=lab"
| jq -r .value
| jq -R 'split(".")[1] | @base64d | fromjson
| {sub, aud, ref, environment, job_workflow_ref}'
EOF
git add .github
git commit -qm "deploy: Environment produktion"
git push -q
gh workflow run deploy.ymlenvironment: produktionbinds the job to the environment: it waits for the approval and afterwards sees its secrets.- With
id-token: write, the runner sets two variables:ACTIONS_ID_TOKEN_REQUEST_URLandACTIONS_ID_TOKEN_REQUEST_TOKEN. With them,curlrequests a token for the audience (audience)lab. - The second
jqsplits the JWT and decodes only the middle part, the claims. The token itself does not end up in the log.
With the ID from
gh run list --workflow deploy.yml --limit 1:
gh run view <ID>Expected (shortened):
* main deploy · 41626025600
Triggered via workflow_dispatch less than a minute ago
JOBS
* deploy (ID 125883462374)
* – the run is not finished.
gh run list --workflow deploy.yml shows the status
waiting: the job is waiting for approval. You can approve
in the browser – or via the API. First the ID of the environment:
gh api repos/{owner}/{repo}/actions/runs/<ID>/pending_deployments \
--jq '.[].environment | .name, .id'Expected (the ID differs):
produktion
26123220715
gh api -X POST repos/{owner}/{repo}/actions/runs/<ID>/pending_deployments \
-F 'environment_ids[]=26123220715' -f state=approved -f comment='Lab-Freigabe' \
--jq '.[].environment'Expected:
produktion
-Fpasses the number as a number,-fas text.environment_ids[]is a list – here with one entry.state=rejectedwould cancel the run.
As soon as the run is finished:
gh run view <ID> --log | grep -E 'Z (Token|Länge):'Expected (timestamps shortened):
deploy Secret nutzen ... Token: ***
deploy Secret nutzen ... Länge: 20
GitHub replaces every secret in the log with ***. The
length shows: the value was there.
Save your progress? Sign up for free
Drill 3: What the OIDC token contains
gh run view <ID> --log | grep -A6 'Z {$' | cut -f3 | cut -c30-Expected (the numbers differ):
{
"sub": "repo:<du>@251214518/notes-actions@1444038371:environment:produktion",
"aud": "lab",
"ref": "refs/heads/main",
"environment": "produktion",
"job_workflow_ref": "<du>/notes-actions/.github/workflows/deploy.yml@refs/heads/main"
}
cut -f3takes the third column of the log (the output),cut -c30-cuts off the timestamp.subis the most important claim. A cloud does not trust “GitHub”, but exactly this value: “only jobs from<du>/notes-actionsin theproduktionenvironment”. A pull request from a fork gets a differentsuband therefore no access.- After owner and repository come their immutable IDs
(
@251214518,@1444038371). If someone renames the repository or creates a new one with the same name, thesubno longer matches – a protection against hijacked names. auddefines whom the token is intended for. A token forlabis useless with a different service.
In AWS, the condition of a role would look something like this:
"token.actions.githubusercontent.com:sub":
"repo:<du>@251214518/notes-actions@1444038371:environment:produktion"
"token.actions.githubusercontent.com:aud": "sts.amazonaws.com"
No access key in the repository, no secret that expires or must be rotated.
Save your progress? Sign up for free
Drill 4: Dependabot
mkdir -p .github
cat > .github/dependabot.yml <<'EOF'
version: 2
updates:
- package-ecosystem: gomod
directory: /
schedule:
interval: weekly
- package-ecosystem: github-actions
directory: /
schedule:
interval: weekly
groups:
actions:
patterns: ['*']
- package-ecosystem: docker
directory: /
schedule:
interval: weekly
EOF
git add .github/dependabot.yml
git commit -qm "dependabot: Go, Actions, Docker"
git push -q- One entry per ecosystem.
github-actionsfinds alluses:– including those pinned by SHA with the comment# v7.0.1from Day 9. groups:combines all action updates into one pull request.dockerreads theFROMlines of the Dockerfile.
Dependabot runs for the first time shortly after the push. After a few minutes:
gh pr list --author app/dependabotExpected (depending on the state of the dependencies):
Showing 1 of 1 pull request in <du>/notes-actions that matches your search
ID TITLE BRANCH CREATED AT
#16 build(deps): bump github.c... dependabot/go_module... about 4 minutes ago
In full:
build(deps): bump github.com/jackc/pgx/v5 from 5.11.0 to 5.12.0.
If nothing shows up, everything is up to date. Whether Dependabot has
run is shown by its runs:
gh run list --workflow 'Dependabot Updates' --limit 3 --json displayTitle \
--jq '.[].displayTitle'Expected (the numbers differ):
docker in /. - Update #1677437037
github_actions in /. - Update #1677437022
go_modules in /. - Update #1677437015
Dependabot Updates is a workflow that GitHub manages
itself – it shows up in gh workflow list --all, but does
not live in your repository. One run per ecosystem. For actions and
Docker there was nothing new – you pinned the actions to the latest
state on Day 9.
Dependabot pull requests go through ci like any other –
the ruleset requires green checks before you merge them.
Save your progress? Sign up for free
Checkpoint
gh api repos/{owner}/{repo}/environments --jq '.environments[].name'Expected:
produktion
gh run list --workflow deploy.yml --limit 1 --json conclusion --jq '.[0].conclusion'Expected (after the challenge it shows failure):
success
gh api repos/{owner}/{repo}/contents/.github/dependabot.yml --jq .nameExpected:
dependabot.yml
Challenge
Reject a deployment: start deploy.yml once more and
reject the approval with a comment. What result does the run have?
Hint: As in Drill 2, with
state=rejected.
Solution
gh workflow run deploy.ymlWith the new ID from
gh run list --workflow deploy.yml --limit 1:
gh api -X POST repos/{owner}/{repo}/actions/runs/<ID>/pending_deployments \
-F 'environment_ids[]=26123220715' -f state=rejected -f comment='Nicht heute' \
--jq '.[].environment'
gh run view <ID> --json conclusion --jq .conclusionExpected:
produktion
failure
The run ends as failure, the deploy job
never started. In the browser, the comment “Nicht heute” is attached to
the rejection.
Save your progress? Sign up for free
Enterprise Note
- OIDC instead of secrets: AWS, Azure, Google Cloud,
HashiCorp Vault and OpenBao accept GitHub OIDC tokens. The trust rule
checks at least
subandaud– a rule on the owner only (repo:<org>/*) lets every repository of the organization in. - Customizing
sub: Organizations can define which claims appear insub(e.g. additionallyjob_workflow_ref). That way access can be bound to a specific reusable workflow (Day 12). - Environments as a gate: Approvals, wait timers
(
wait_timer) and branch rules are the place where a human consents before production – with an entry in the audit log. - Renovate instead of Dependabot: Renovate can do more – custom data sources, rules across many repositories, automerge after a test run. The sister book “Renovate hands-on” goes into depth.