Skip to content

Commit 7e268f3

Browse files
sirutBuasaisirutBuasai
authored
port label tests with sagemaker to v2 (#5708)
* port label tests with sagemaker to v2 Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * fix tests to run build only Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * run pr tests Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * fix rayserve Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * trigger all workflows Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * chanage release to prod Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * change workflow name Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * disable vllm Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * use reusable workflow Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * use resuabel workflow for sglang Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * echo build cmd Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * add concurrency Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * use pformat Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * add vllm auto release test Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * align check label with build image Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * test pr test-change only path logic Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * revert pr test change Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * split sanity and remove sagemaker test change on ec2 Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> * rename auto release files Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com> --------- Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com>
1 parent 67ee3c4 commit 7e268f3

32 files changed

+1879
-1464
lines changed

‎.github/config/sglang-ec2.yml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -27,4 +27,4 @@ release:
2727
public_registry: true
2828
private_registry: true
2929
enable_soci: true
30-
environment: gamma # gamma or production
30+
environment: production # gamma or production

‎.github/config/sglang-sagemaker.yml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -27,4 +27,4 @@ release:
2727
public_registry: true
2828
private_registry: true
2929
enable_soci: true
30-
environment: gamma # gamma or production
30+
environment: production # gamma or production

‎.github/scripts/build_image.sh‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -117,6 +117,7 @@ BUILD_CMD="${BUILD_CMD} \
117117

118118
# Execute build
119119
echo "Executing build command..."
120+
echo "${BUILD_CMD}"
120121
eval ${BUILD_CMD}
121122

122123
# Clean up local image

‎.github/workflows/auto-release-sglang-ec2.yml‎

Lines changed: 195 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,195 @@
1+
name: Auto Release - SGLang EC2
2+
3+
on:
4+
# schedule run must be on the default branch
5+
schedule:
6+
# Runs at 10:00 AM PST/PDT on Monday and Wednesday
7+
- cron: '00 17 * * 2,4' # Tuesday and Thursday at 10:00 AM PDT / 9:00 AM PST
8+
9+
# # PR triggers for testing
10+
# pull_request:
11+
# paths:
12+
# - "**sglang**"
13+
# - "!docs/**"
14+
15+
# Manual trigger
16+
workflow_dispatch:
17+
18+
permissions:
19+
contents: read
20+
pull-requests: read
21+
22+
env:
23+
# CI environment configuration
24+
FORCE_COLOR: "1"
25+
26+
# Config file path
27+
CONFIG_FILE: ".github/config/sglang-ec2.yml"
28+
29+
jobs:
30+
load-config:
31+
runs-on: ubuntu-latest
32+
concurrency:
33+
group: ${{ github.workflow }}-load-config-${{ github.event.pull_request.number || github.run_id }}
34+
cancel-in-progress: true
35+
outputs:
36+
config: ${{ steps.load.outputs.config }}
37+
framework: ${{ steps.parse.outputs.framework }}
38+
framework-version: ${{ steps.parse.outputs.framework-version }}
39+
python-version: ${{ steps.parse.outputs.python-version }}
40+
cuda-version: ${{ steps.parse.outputs.cuda-version }}
41+
os-version: ${{ steps.parse.outputs.os-version }}
42+
container-type: ${{ steps.parse.outputs.container-type }}
43+
device-type: ${{ steps.parse.outputs.device-type }}
44+
arch-type: ${{ steps.parse.outputs.arch-type }}
45+
contributor: ${{ steps.parse.outputs.contributor }}
46+
customer-type: ${{ steps.parse.outputs.customer-type }}
47+
prod-image: ${{ steps.parse.outputs.prod-image }}
48+
steps:
49+
- name: Checkout code
50+
uses: actions/checkout@v5
51+
52+
- name: Load configuration
53+
id: load
54+
uses: ./.github/actions/load-config
55+
with:
56+
config-file: ${{ env.CONFIG_FILE }}
57+
58+
- name: Parse configuration
59+
id: parse
60+
run: |
61+
echo '${{ steps.load.outputs.config }}' > config.json
62+
echo "framework=$(jq -r '.common.framework' config.json)" >> $GITHUB_OUTPUT
63+
echo "framework-version=$(jq -r '.common.framework_version' config.json)" >> $GITHUB_OUTPUT
64+
echo "python-version=$(jq -r '.common.python_version' config.json)" >> $GITHUB_OUTPUT
65+
echo "cuda-version=$(jq -r '.common.cuda_version' config.json)" >> $GITHUB_OUTPUT
66+
echo "os-version=$(jq -r '.common.os_version' config.json)" >> $GITHUB_OUTPUT
67+
echo "container-type=$(jq -r '.common.job_type' config.json)" >> $GITHUB_OUTPUT
68+
echo "device-type=$(jq -r '.common.device_type // "gpu"' config.json)" >> $GITHUB_OUTPUT
69+
echo "arch-type=$(jq -r '.common.arch_type // "x86"' config.json)" >> $GITHUB_OUTPUT
70+
echo "contributor=$(jq -r '.common.contributor // "None"' config.json)" >> $GITHUB_OUTPUT
71+
echo "customer-type=$(jq -r '.common.customer_type // ""' config.json)" >> $GITHUB_OUTPUT
72+
echo "prod-image=$(jq -r '.common.prod_image' config.json)" >> $GITHUB_OUTPUT
73+
74+
build-image:
75+
needs: [load-config]
76+
runs-on:
77+
- codebuild-runner-${{ github.run_id }}-${{ github.run_attempt }}
78+
fleet:x86-build-runner
79+
buildspec-override:true
80+
concurrency:
81+
group: ${{ github.workflow }}-build-image-${{ github.event.pull_request.number || github.run_id }}
82+
cancel-in-progress: true
83+
outputs:
84+
ci-image: ${{ steps.build.outputs.image-uri }}
85+
steps:
86+
- name: Checkout code
87+
uses: actions/checkout@v5
88+
89+
- name: Build image
90+
id: build
91+
uses: ./.github/actions/build-image
92+
with:
93+
framework: ${{ needs.load-config.outputs.framework }}
94+
target: sglang-ec2
95+
base-image: lmsysorg/sglang:v${{ needs.load-config.outputs.framework-version }}-${{ needs.load-config.outputs.cuda-version }}-amd64
96+
framework-version: ${{ needs.load-config.outputs.framework-version }}
97+
container-type: ${{ needs.load-config.outputs.container-type }}
98+
aws-account-id: ${{ vars.CI_AWS_ACCOUNT_ID }}
99+
aws-region: ${{ vars.AWS_REGION }}
100+
tag-pr: ${{ needs.load-config.outputs.framework }}-${{ needs.load-config.outputs.framework-version }}-gpu-${{ needs.load-config.outputs.python-version }}-${{ needs.load-config.outputs.cuda-version }}-${{ needs.load-config.outputs.os-version }}-ec2-${{ github.run_id }}
101+
dockerfile-path: docker/${{ needs.load-config.outputs.framework }}/Dockerfile
102+
arch-type: ${{ needs.load-config.outputs.arch-type }}
103+
device-type: ${{ needs.load-config.outputs.device-type }}
104+
cuda-version: ${{ needs.load-config.outputs.cuda-version }}
105+
python-version: ${{ needs.load-config.outputs.python-version }}
106+
os-version: ${{ needs.load-config.outputs.os-version }}
107+
contributor: ${{ needs.load-config.outputs.contributor }}
108+
customer-type: ${{ needs.load-config.outputs.customer-type }}
109+
110+
sanity-test:
111+
needs: [build-image, load-config]
112+
concurrency:
113+
group: ${{ github.workflow }}-sanity-test-${{ github.event.pull_request.number || github.run_id }}
114+
cancel-in-progress: true
115+
uses: ./.github/workflows/reusable-sanity-tests.yml
116+
with:
117+
image-uri: ${{ needs.build-image.outputs.ci-image }}
118+
aws-account-id: ${{ vars.CI_AWS_ACCOUNT_ID }}
119+
aws-region: ${{ vars.AWS_REGION }}
120+
framework: ${{ needs.load-config.outputs.framework }}
121+
framework-version: ${{ needs.load-config.outputs.framework-version }}
122+
python-version: ${{ needs.load-config.outputs.python-version }}
123+
cuda-version: ${{ needs.load-config.outputs.cuda-version }}
124+
os-version: ${{ needs.load-config.outputs.os-version }}
125+
customer-type: ${{ needs.load-config.outputs.customer-type }}
126+
arch-type: ${{ needs.load-config.outputs.arch-type }}
127+
device-type: ${{ needs.load-config.outputs.device-type }}
128+
contributor: ${{ needs.load-config.outputs.contributor }}
129+
container-type: ${{ needs.load-config.outputs.container-type }}
130+
131+
upstream-tests:
132+
needs: [build-image, load-config]
133+
concurrency:
134+
group: ${{ github.workflow }}-upstream-tests-${{ github.event.pull_request.number || github.run_id }}
135+
cancel-in-progress: true
136+
uses: ./.github/workflows/reusable-sglang-upstream-tests.yml
137+
with:
138+
image-uri: ${{ needs.build-image.outputs.ci-image }}
139+
aws-account-id: ${{ vars.CI_AWS_ACCOUNT_ID }}
140+
aws-region: ${{ vars.AWS_REGION }}
141+
framework-version: ${{ needs.load-config.outputs.framework-version }}
142+
benchmark-start-command: >-
143+
docker run -d -it --rm --gpus=all
144+
-v ${HOME}/.cache/huggingface:/root/.cache/huggingface
145+
-v ${TEST_ARTIFACTS_DIRECTORY}/dataset:/dataset
146+
-p 30000:30000
147+
-e HF_TOKEN=${HF_TOKEN}
148+
${IMAGE}
149+
--model-path Qwen/Qwen3-0.6B
150+
--reasoning-parser qwen3
151+
--host 127.0.0.1
152+
--port 30000
153+
secrets: inherit
154+
155+
generate-release-spec:
156+
needs: [load-config, build-image, sanity-test, upstream-tests]
157+
runs-on: ubuntu-latest
158+
concurrency:
159+
group: ${{ github.workflow }}-generate-release-spec-${{ github.event.pull_request.number || github.run_id }}
160+
cancel-in-progress: true
161+
outputs:
162+
release-spec: ${{ steps.generate.outputs.release-spec }}
163+
should-release: ${{ steps.check-release.outputs.should-release }}
164+
steps:
165+
- name: Checkout code
166+
uses: actions/checkout@v5
167+
168+
- name: Check if release is enabled
169+
id: check-release
170+
run: |
171+
echo '${{ needs.load-config.outputs.config }}' > config.json
172+
RELEASE_ENABLED=$(jq -r '.release.release // false' config.json)
173+
echo "Release enabled: ${RELEASE_ENABLED}"
174+
echo "should-release=${RELEASE_ENABLED}" >> $GITHUB_OUTPUT
175+
176+
- name: Generate release spec
177+
id: generate
178+
if: steps.check-release.outputs.should-release == 'true'
179+
uses: ./.github/actions/generate-release-spec
180+
with:
181+
config-json: ${{ needs.load-config.outputs.config }}
182+
183+
release-image:
184+
needs: [load-config, build-image, generate-release-spec]
185+
if: needs.generate-release-spec.outputs.should-release == 'true'
186+
concurrency:
187+
group: ${{ github.workflow }}-release-image-${{ github.event.pull_request.number || github.run_id }}
188+
cancel-in-progress: false
189+
uses: ./.github/workflows/reusable-release-image.yml
190+
with:
191+
source-image-uri: ${{ needs.build-image.outputs.ci-image }}
192+
release-spec: ${{ needs.generate-release-spec.outputs.release-spec }}
193+
environment: ${{ github.event_name == 'pull_request' && 'gamma' || fromJson(needs.load-config.outputs.config).release.environment }}
194+
aws-region: ${{ vars.AWS_REGION }}
195+
runner-fleet: default-runner

0 commit comments

Comments
 (0)