prometheus/.github/workflows/prombench.yml
machine424 91ef9a50b9
chore(prombench): re-roder checks on nodes to avoid deadlocks start and cancel and in restart
the start job requires the node pools to be gone before creating them, and
the cleanup and restart jobs require the pools to be running to delete them,
When the initial start is partial (only one pool was created), no command can
move forward...

the preconditions should be relaxed, for more robustness.

Signed-off-by: machine424 <ayoubmrini424@gmail.com>
2025-09-02 12:26:08 +02:00

131 lines
5.8 KiB
YAML

on:
repository_dispatch:
types: [prombench_start, prombench_restart, prombench_stop]
name: Prombench Workflow
permissions:
contents: read
env:
AUTH_FILE: ${{ secrets.TEST_INFRA_PROVIDER_AUTH }}
CLUSTER_NAME: test-infra
DOMAIN_NAME: prombench.prometheus.io
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
GITHUB_ORG: prometheus
GITHUB_REPO: prometheus
GITHUB_STATUS_TARGET_URL: https://github.com/${{github.repository}}/actions/runs/${{github.run_id}}
LAST_COMMIT_SHA: ${{ github.event.client_payload.LAST_COMMIT_SHA }}
GKE_PROJECT_ID: macro-mile-203600
PR_NUMBER: ${{ github.event.client_payload.PR_NUMBER }}
PROVIDER: gke
RELEASE: ${{ github.event.client_payload.RELEASE }}
BENCHMARK_VERSION: ${{ github.event.client_payload.BENCHMARK_VERSION }}
BENCHMARK_DIRECTORY: ${{ github.event.client_payload.BENCHMARK_DIRECTORY }}
ZONE: europe-west3-a
jobs:
benchmark_start:
name: Benchmark Start
if: github.event.action == 'prombench_start'
runs-on: ubuntu-latest
steps:
- name: Update status to pending
run: >-
curl -i -X POST
-H "Authorization: Bearer $GITHUB_TOKEN"
-H "Content-Type: application/json"
--data '{"state":"pending", "context": "prombench-status-update-start", "target_url": "'$GITHUB_STATUS_TARGET_URL'"}'
"https://api.github.com/repos/$GITHUB_REPOSITORY/statuses/$LAST_COMMIT_SHA"
- name: Run make deploy to start test
id: make_deploy
uses: docker://prominfra/prombench:master
with:
args: >-
make deploy;
until make all_nodes_running; do echo "waiting for nodepools to be created"; sleep 10; done;
- name: Update status to failure
if: failure()
run: >-
curl -i -X POST
-H "Authorization: Bearer $GITHUB_TOKEN"
-H "Content-Type: application/json"
--data '{"state":"failure", "context": "prombench-status-update-start", "target_url": "'$GITHUB_STATUS_TARGET_URL'"}'
"https://api.github.com/repos/$GITHUB_REPOSITORY/statuses/$LAST_COMMIT_SHA"
- name: Update status to success
if: success()
run: >-
curl -i -X POST
-H "Authorization: Bearer $GITHUB_TOKEN"
-H "Content-Type: application/json"
--data '{"state":"success", "context": "prombench-status-update-start", "target_url": "'$GITHUB_STATUS_TARGET_URL'"}'
"https://api.github.com/repos/$GITHUB_REPOSITORY/statuses/$LAST_COMMIT_SHA"
benchmark_cancel:
name: Benchmark Cancel
if: github.event.action == 'prombench_stop'
runs-on: ubuntu-latest
steps:
- name: Update status to pending
run: >-
curl -i -X POST
-H "Authorization: Bearer $GITHUB_TOKEN"
-H "Content-Type: application/json"
--data '{"state":"pending", "context": "prombench-status-update-cancel", "target_url": "'$GITHUB_STATUS_TARGET_URL'"}'
"https://api.github.com/repos/$GITHUB_REPOSITORY/statuses/$LAST_COMMIT_SHA"
- name: Run make clean to stop test
id: make_clean
uses: docker://prominfra/prombench:master
with:
args: >-
make clean;
until make all_nodes_deleted; do echo "waiting for nodepools to be deleted"; sleep 10; done;
- name: Update status to failure
if: failure()
run: >-
curl -i -X POST
-H "Authorization: Bearer $GITHUB_TOKEN"
-H "Content-Type: application/json"
--data '{"state":"failure", "context": "prombench-status-update-cancel", "target_url": "'$GITHUB_STATUS_TARGET_URL'"}'
"https://api.github.com/repos/$GITHUB_REPOSITORY/statuses/$LAST_COMMIT_SHA"
- name: Update status to success
if: success()
run: >-
curl -i -X POST
-H "Authorization: Bearer $GITHUB_TOKEN"
-H "Content-Type: application/json"
--data '{"state":"success", "context": "prombench-status-update-cancel", "target_url": "'$GITHUB_STATUS_TARGET_URL'"}'
"https://api.github.com/repos/$GITHUB_REPOSITORY/statuses/$LAST_COMMIT_SHA"
benchmark_restart:
name: Benchmark Restart
if: github.event.action == 'prombench_restart'
runs-on: ubuntu-latest
steps:
- name: Update status to pending
run: >-
curl -i -X POST
-H "Authorization: Bearer $GITHUB_TOKEN"
-H "Content-Type: application/json"
--data '{"state":"pending", "context": "prombench-status-update-restart", "target_url": "'$GITHUB_STATUS_TARGET_URL'"}'
"https://api.github.com/repos/$GITHUB_REPOSITORY/statuses/$LAST_COMMIT_SHA"
- name: Run make clean then make deploy to restart test
id: make_restart
uses: docker://prominfra/prombench:master
with:
args: >-
make clean;
until make all_nodes_deleted; do echo "waiting for nodepools to be deleted"; sleep 10; done;
make deploy;
until make all_nodes_running; do echo "waiting for nodepools to be created"; sleep 10; done;
- name: Update status to failure
if: failure()
run: >-
curl -i -X POST
-H "Authorization: Bearer $GITHUB_TOKEN"
-H "Content-Type: application/json"
--data '{"state":"failure", "context": "prombench-status-update-restart", "target_url": "'$GITHUB_STATUS_TARGET_URL'"}'
"https://api.github.com/repos/$GITHUB_REPOSITORY/statuses/$LAST_COMMIT_SHA"
- name: Update status to success
if: success()
run: >-
curl -i -X POST
-H "Authorization: Bearer $GITHUB_TOKEN"
-H "Content-Type: application/json"
--data '{"state":"success", "context": "prombench-status-update-restart", "target_url": "'$GITHUB_STATUS_TARGET_URL'"}'
"https://api.github.com/repos/$GITHUB_REPOSITORY/statuses/$LAST_COMMIT_SHA"