Skip to content

Commit ded4302

Browse files
mamy-CSmamy-CS
authored
make full e2e tests required on pr ci run (llm-d#842)
* make full e2e tests required on pr ci run Signed-off-by: Mohammed Abdi <mohammed.munir.abdi@ibm.com> * update smoke tests name Signed-off-by: Mohammed Abdi <mohammed.munir.abdi@ibm.com> --------- Signed-off-by: Mohammed Abdi <mohammed.munir.abdi@ibm.com>
1 parent 35808f1 commit ded4302

1 file changed

Lines changed: 100 additions & 49 deletions

File tree

‎.github/workflows/ci-pr-checks.yaml‎

Lines changed: 100 additions & 49 deletions
Original file line numberDiff line numberDiff line change
@@ -35,7 +35,7 @@ on:
3535
workflow_dispatch:
3636
inputs:
3737
run_full_tests:
38-
description: 'Run full e2e test suite on Kind (default: smoke tests only)'
38+
description: 'Run full e2e test suite on Kind (check to run full; unchecked runs no e2e)'
3939
required: false
4040
default: false
4141
type: boolean
@@ -254,24 +254,104 @@ jobs:
254254
255255
core.setOutput('run_full', 'false');
256256
257-
# E2E tests - smoke tests run automatically, full tests on approval
258-
# Skip e2e tests if PR only contains docs/metadata changes (unless explicitly triggered via /trigger-e2e-full)
259-
e2e-tests:
257+
# E2E smoke tests - run automatically on every PR with code changes
258+
# Skip if PR only contains docs/metadata changes
259+
e2e-tests-smoke:
260+
runs-on: ubuntu-latest
261+
needs: [lint-and-test, check-code-changes, check-full-tests]
262+
if: github.event_name == 'pull_request' && needs.check-code-changes.result == 'success' && needs.check-code-changes.outputs.has_code_changes == 'true'
263+
timeout-minutes: 60
264+
permissions:
265+
contents: read
266+
steps:
267+
- name: Checkout source
268+
uses: actions/checkout@v4
269+
270+
- name: Extract Go version from go.mod
271+
run: sed -En 's/^go (.*)$/GO_VERSION=\1/p' go.mod >> $GITHUB_ENV
272+
273+
- name: Set up Go with cache
274+
uses: actions/setup-go@v6
275+
with:
276+
go-version: "${{ env.GO_VERSION }}"
277+
cache-dependency-path: ./go.sum
278+
279+
- name: Install dependencies
280+
run: go mod download
281+
282+
- name: Install Kind
283+
run: |
284+
ARCH=$(uname -m)
285+
case "$ARCH" in
286+
x86_64) KIND_ARCH="amd64" ;;
287+
aarch64) KIND_ARCH="arm64" ;;
288+
*) echo "Unsupported architecture: $ARCH"; exit 1 ;;
289+
esac
290+
curl -Lo ./kind "https://kind.sigs.k8s.io/dl/v0.25.0/kind-linux-${KIND_ARCH}"
291+
chmod +x ./kind
292+
sudo mv ./kind /usr/local/bin/kind
293+
kind version
294+
295+
- name: Set up Docker Buildx
296+
uses: docker/setup-buildx-action@v3
297+
298+
- name: Build WVA image locally
299+
id: build-image
300+
env:
301+
CHECKOUT_SHA: ${{ needs.check-code-changes.outputs.pr_head_sha || github.sha }}
302+
run: |
303+
# Generate unique image tag for this PR run (local image, no registry needed)
304+
IMAGE_NAME="llm-d-workload-variant-autoscaler"
305+
IMAGE_TAG="pr-${GITHUB_RUN_ID}-${CHECKOUT_SHA:0:7}"
306+
# Use localhost prefix for local-only image (Kind will load it directly)
307+
FULL_IMAGE="localhost/${IMAGE_NAME}:${IMAGE_TAG}"
308+
309+
echo "Building local image: $FULL_IMAGE"
310+
echo "Image will be loaded into Kind cluster (no push needed)"
311+
312+
# Build image locally (no push needed for Kind)
313+
make docker-build IMG="$FULL_IMAGE"
314+
315+
echo "image=$FULL_IMAGE" >> $GITHUB_OUTPUT
316+
echo "image_tag=${IMAGE_TAG}" >> $GITHUB_OUTPUT
317+
echo "Image built locally: $FULL_IMAGE"
318+
319+
- name: Run e2e tests (smoke)
320+
shell: bash
321+
env:
322+
ENVIRONMENT: kind-emulator
323+
USE_SIMULATOR: "true"
324+
SCALE_TO_ZERO_ENABLED: "false"
325+
CREATE_CLUSTER: "true"
326+
INSTALL_GATEWAY_CTRLPLANE: "true"
327+
E2E_TESTS_ENABLED: "true"
328+
IMG: ${{ steps.build-image.outputs.image }}
329+
SKIP_BUILD: "true"
330+
PROMETHEUS_ADAPTER_WAIT: "false"
331+
DELETE_CLUSTER: "true"
332+
# Lower saturation thresholds for simulator mode — the simulator's
333+
# KV-cache and queue metrics are modest, so default thresholds
334+
# (kvSpareTrigger=0.1, queueSpareTrigger=3) are too high to trigger
335+
# scale-up reliably. These values trigger when kvUsage > 0.30 or
336+
# queueLength > 0.5, which the simulator produces under load.
337+
KV_SPARE_TRIGGER: "0.5"
338+
QUEUE_SPARE_TRIGGER: "4.5"
339+
run: |
340+
make test-e2e-smoke-with-setup
341+
342+
# Full Kind E2E - runs only when triggered (/trigger-e2e-full or workflow_dispatch with run_full_tests).
343+
# Add "e2e-tests-full" as a required status check in branch protection so it appears on every PR
344+
# and must run and pass before merge (shows as "Expected" until triggered).
345+
e2e-tests-full:
260346
runs-on: ubuntu-latest
261347
needs: [lint-and-test, check-code-changes, check-full-tests]
262348
if: >-
263-
always() && (
264-
(github.event_name == 'issue_comment' && needs.check-full-tests.outputs.run_full == 'true' && needs.check-code-changes.result == 'success') ||
265-
(github.event_name == 'workflow_dispatch' && (
266-
needs.check-full-tests.outputs.run_full == 'true' ||
267-
(needs.check-code-changes.result == 'success' && needs.check-code-changes.outputs.has_code_changes == 'true')
268-
)) ||
269-
(github.event_name == 'pull_request' && needs.check-code-changes.result == 'success' && needs.check-code-changes.outputs.has_code_changes == 'true')
270-
)
349+
always() && needs.check-full-tests.outputs.run_full == 'true' && needs.check-code-changes.result == 'success'
350+
&& (github.event_name == 'issue_comment' || github.event_name == 'workflow_dispatch')
271351
timeout-minutes: 60
272352
permissions:
273353
contents: read
274-
statuses: write # For reporting status on fork PR commits
354+
statuses: write
275355
steps:
276356
- name: Set pending status on PR head
277357
if: github.event_name == 'issue_comment' && needs.check-code-changes.outputs.pr_head_sha != ''
@@ -284,7 +364,7 @@ jobs:
284364
sha: '${{ needs.check-code-changes.outputs.pr_head_sha }}',
285365
state: 'pending',
286366
target_url: `https://github.com/${context.repo.owner}/${context.repo.repo}/actions/runs/${context.runId}`,
287-
description: 'E2E tests running...',
367+
description: 'Full E2E tests running...',
288368
context: '${{ github.workflow }} / e2e (comment trigger)'
289369
});
290370
@@ -301,7 +381,6 @@ jobs:
301381
- name: Checkout source
302382
uses: actions/checkout@v4
303383
with:
304-
# For issue_comment on a fork PR: checkout from head repo so the commit exists
305384
repository: ${{ needs.check-code-changes.outputs.pr_head_repo || github.repository }}
306385
ref: ${{ needs.check-code-changes.outputs.pr_head_sha || github.sha }}
307386
token: ${{ secrets.GITHUB_TOKEN }}
@@ -339,65 +418,37 @@ jobs:
339418
env:
340419
CHECKOUT_SHA: ${{ needs.check-code-changes.outputs.pr_head_sha || github.sha }}
341420
run: |
342-
# Generate unique image tag for this PR run (local image, no registry needed)
343421
IMAGE_NAME="llm-d-workload-variant-autoscaler"
344422
IMAGE_TAG="pr-${GITHUB_RUN_ID}-${CHECKOUT_SHA:0:7}"
345-
# Use localhost prefix for local-only image (Kind will load it directly)
346423
FULL_IMAGE="localhost/${IMAGE_NAME}:${IMAGE_TAG}"
347-
348424
echo "Building local image: $FULL_IMAGE"
349-
echo "Image will be loaded into Kind cluster (no push needed)"
350-
351-
# Build image locally (no push needed for Kind)
352425
make docker-build IMG="$FULL_IMAGE"
353-
354426
echo "image=$FULL_IMAGE" >> $GITHUB_OUTPUT
355427
echo "image_tag=${IMAGE_TAG}" >> $GITHUB_OUTPUT
356-
echo "Image built locally: $FULL_IMAGE"
357-
358-
- name: Determine test type
359-
id: test-type
360-
run: |
361-
if [ "${{ needs.check-full-tests.outputs.run_full }}" == "true" ]; then
362-
echo "test_target=test-e2e-full-with-setup" >> $GITHUB_OUTPUT
363-
echo "test_name=full" >> $GITHUB_OUTPUT
364-
echo "scale_to_zero=true" >> $GITHUB_OUTPUT
365-
echo "delete_cluster=false" >> $GITHUB_OUTPUT
366-
else
367-
echo "test_target=test-e2e-smoke-with-setup" >> $GITHUB_OUTPUT
368-
echo "test_name=smoke" >> $GITHUB_OUTPUT
369-
echo "scale_to_zero=false" >> $GITHUB_OUTPUT
370-
echo "delete_cluster=true" >> $GITHUB_OUTPUT
371-
fi
372428
373-
- name: Run e2e tests (${{ steps.test-type.outputs.test_name }})
429+
- name: Run e2e tests (full)
374430
shell: bash
375431
env:
376432
ENVIRONMENT: kind-emulator
377433
USE_SIMULATOR: "true"
378-
SCALE_TO_ZERO_ENABLED: ${{ steps.test-type.outputs.scale_to_zero }}
434+
SCALE_TO_ZERO_ENABLED: "true"
379435
CREATE_CLUSTER: "true"
380436
INSTALL_GATEWAY_CTRLPLANE: "true"
381437
E2E_TESTS_ENABLED: "true"
382438
IMG: ${{ steps.build-image.outputs.image }}
383439
SKIP_BUILD: "true"
384440
PROMETHEUS_ADAPTER_WAIT: "false"
385-
DELETE_CLUSTER: ${{ steps.test-type.outputs.delete_cluster }}
386-
# Lower saturation thresholds for simulator mode — the simulator's
387-
# KV-cache and queue metrics are modest, so default thresholds
388-
# (kvSpareTrigger=0.1, queueSpareTrigger=3) are too high to trigger
389-
# scale-up reliably. These values trigger when kvUsage > 0.30 or
390-
# queueLength > 0.5, which the simulator produces under load.
441+
DELETE_CLUSTER: "false"
391442
KV_SPARE_TRIGGER: "0.5"
392443
QUEUE_SPARE_TRIGGER: "4.5"
393444
run: |
394-
make ${{ steps.test-type.outputs.test_target }}
445+
make test-e2e-full-with-setup
395446
396447
# Report status back to PR for issue_comment triggered runs
397448
# This ensures fork PRs show the correct status after /trigger-e2e-full runs complete
398449
report-status:
399450
runs-on: ubuntu-latest
400-
needs: [check-code-changes, check-full-tests, e2e-tests]
451+
needs: [check-code-changes, check-full-tests, e2e-tests-full]
401452
if: always() && github.event_name == 'issue_comment' && needs.check-full-tests.outputs.run_full == 'true'
402453
permissions:
403454
statuses: write
@@ -407,7 +458,7 @@ jobs:
407458
with:
408459
script: |
409460
const prHeadSha = '${{ needs.check-code-changes.outputs.pr_head_sha }}';
410-
const e2eResult = '${{ needs.e2e-tests.result }}';
461+
const e2eResult = '${{ needs.e2e-tests-full.result }}';
411462
412463
if (!prHeadSha) {
413464
console.log('No PR head SHA available, skipping status report');

0 commit comments

Comments
 (0)