This is an automated email from the ASF dual-hosted git repository.
zhouyuan pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/gluten.git
The following commit(s) were added to refs/heads/main by this push:
new 4a24955103 [GLUTEN-12288][CI] Move more jobs to nightly (#12814)
4a24955103 is described below
commit 4a249551035d58de9e95c1227e886cf8ca694edb
Author: Yuan <[email protected]>
AuthorDate: Wed Aug 19 08:31:14 2026 +0100
[GLUTEN-12288][CI] Move more jobs to nightly (#12814)
---
.github/workflows/velox_backend_x86.yml | 137 +++++++++++++++++++++++++-------
1 file changed, 108 insertions(+), 29 deletions(-)
diff --git a/.github/workflows/velox_backend_x86.yml
b/.github/workflows/velox_backend_x86.yml
index 89d759e432..951f55cc9f 100644
--- a/.github/workflows/velox_backend_x86.yml
+++ b/.github/workflows/velox_backend_x86.yml
@@ -38,6 +38,17 @@ on:
- 'cpp/**'
- 'dev/**'
- 'build/mvn'
+ # Nightly full run. PR runs trim the TPC matrices to representative JDKs and
+ # skip the OOM / random-kill chaos jobs unless native code changed (see
+ # detect-changes' `full_run` output); this scheduled run restores the full
+ # breadth once a day, so a combo dropped from the per-PR tier can stay broken
+ # for at most a day. It also gives the default branch a post-merge signal,
+ # which the per-PR-only trigger above never provided. 02:00 avoids the
+ # 00:00 velox_nightly.yml and 05:00 delta_spark_ut.yml slots.
+ schedule:
+ - cron: '0 2 * * *'
+ # On-demand full run (e.g. to re-check an extended combo before a release).
+ workflow_dispatch:
env:
ACTIONS_ALLOW_USE_UNSECURE_NODE_VERSION: true
@@ -67,9 +78,19 @@ jobs:
# • gluten-core change → java=true → all spark-test jobs run ✓
# • shims/spark35 only → shims35=true → only spark35 jobs run ✓
# • pure cpp change → cpp=true → only native/tpc jobs run ✓
+ #
+ # The `full_run` flag is true on schedule/workflow_dispatch runs: those have
no
+ # PR diff to filter on, so every path flag is forced true AND the jobs below
+ # widen their matrices to the extended combos that PR runs skip (alternative
+ # JDKs, older Celeborn, chaos tests on java-only changes).
detect-changes:
+ # Every job chains off this one, so the fork guard here skips the whole
+ # workflow: a fork with Actions enabled would otherwise burn ~1800 runner
+ # minutes nightly on a signal nobody reads.
+ if: github.event_name != 'schedule' || github.repository_owner == 'apache'
runs-on: ubuntu-22.04
outputs:
+ full_run: ${{ steps.filter.outputs.full_run }}
cpp: ${{ steps.filter.outputs.cpp }}
java: ${{ steps.filter.outputs.java }}
shims33: ${{ steps.filter.outputs.shims33 }}
@@ -85,6 +106,30 @@ jobs:
- name: Detect changed paths
id: filter
run: |
+ event="${{ github.event_name }}"
+ if [ "$event" = "schedule" ] || [ "$event" = "workflow_dispatch" ];
then
+ # schedule / workflow_dispatch: no PR diff to inspect.
+ # For scheduled runs, skip entirely when no commits landed in the
+ # last 24 hours — avoids burning runner minutes on idle days.
+ if [ "$event" = "schedule" ]; then
+ recent=$(git log --oneline --after="24 hours ago" HEAD | wc -l)
+ if [ "$recent" -eq 0 ]; then
+ echo "No commits in the last 24 hours -- skipping nightly run."
+ for flag in cpp java shims33 shims34 shims35 shims40 shims41
tools_it; do
+ echo "$flag=false" >> $GITHUB_OUTPUT
+ done
+ echo "full_run=false" >> $GITHUB_OUTPUT
+ exit 0
+ fi
+ fi
+ # Run everything, including the extended matrices PR runs skip.
+ for flag in cpp java shims33 shims34 shims35 shims40 shims41
tools_it; do
+ echo "$flag=true" >> $GITHUB_OUTPUT
+ done
+ echo "full_run=true" >> $GITHUB_OUTPUT
+ exit 0
+ fi
+ echo "full_run=false" >> $GITHUB_OUTPUT
BASE=${{ github.event.pull_request.base.sha }}
HEAD=${{ github.sha }}
changed=$(git diff --name-only "$BASE" "$HEAD")
@@ -145,8 +190,9 @@ jobs:
- uses: actions/upload-artifact@v7
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
if-no-files-found: error
+ retention-days: 1
tpc-test-ubuntu:
needs: [detect-changes, build-native-lib-centos-7]
@@ -159,7 +205,16 @@ jobs:
matrix:
os: [ "ubuntu:22.04" ]
spark: [ "spark-3.3", "spark-3.4", "spark-3.5", "spark-4.0",
"spark-4.1" ]
- java: [ "java-8", "java-11", "java-17", "java-21", "java-25" ]
+ # PR runs test the primary JDK per Spark line (8 for 3.x, 17 for 4.x --
+ # after the static excludes below that is 6 combos instead of 11). The
+ # alternative-JDK combos (11/21/25) validate build + TPC run under JDKs
+ # that Gluten changes rarely break in a JDK-specific way; the nightly
+ # full_run keeps covering them daily so a break surfaces within a day
+ # instead of costing ~75 runner-minutes on every PR.
+ java: >-
+ ${{ fromJSON(needs.detect-changes.outputs.full_run == 'true'
+ && '["java-8", "java-11", "java-17", "java-21", "java-25"]'
+ || '["java-8", "java-17"]') }}
# Spark supports JDK17 since 3.3.
exclude:
- spark: spark-3.3
@@ -199,7 +254,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Setup tzdata
run: |
sed -i 's|http://archive|http://eu.archive|g' /etc/apt/sources.list
@@ -289,10 +344,21 @@ jobs:
needs.detect-changes.outputs.tools_it == 'true'
strategy:
fail-fast: false
+ # This job's per-PR value is OS compatibility of the same native lib and
+ # TPC run that tpc-test-ubuntu already exercises combo-by-combo, so PRs
+ # test one representative combo per Spark line ({3.5, 4.1} x primary JDK
+ # = 2 jobs after the excludes below) and the nightly full_run restores
+ # the whole 7-combo fan-out.
matrix:
os: [ "centos:8" ]
- spark: [ "spark-3.3", "spark-3.4", "spark-3.5", "spark-4.0",
"spark-4.1" ]
- java: [ "java-8", "java-11", "java-17" ]
+ spark: >-
+ ${{ fromJSON(needs.detect-changes.outputs.full_run == 'true'
+ && '["spark-3.3", "spark-3.4", "spark-3.5", "spark-4.0",
"spark-4.1"]'
+ || '["spark-3.5", "spark-4.1"]') }}
+ java: >-
+ ${{ fromJSON(needs.detect-changes.outputs.full_run == 'true'
+ && '["java-8", "java-11", "java-17"]'
+ || '["java-8", "java-17"]') }}
# Spark supports JDK17 since 3.3.
exclude:
- spark: spark-3.4
@@ -320,7 +386,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Update mirror list
run: |
if [ "${{ matrix.os }}" = "centos:7" ] || [ "${{ matrix.os }}" =
"centos:8" ]; then
@@ -386,7 +452,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Build and run TPCH/DS tests
run: |
docker pull centos:7
@@ -417,10 +483,13 @@ jobs:
tpc-test-ubuntu-oom:
needs: [detect-changes, build-native-lib-centos-7]
+ # Chaos/stress tier: this job probes native memory-pressure behaviour, so
+ # per-PR it only runs when native code changed (cpp=true). For java-only
+ # changes its ~60 runner-minutes rarely produce PR-specific signal; the
+ # nightly full_run still exercises it daily.
if: >-
- needs.detect-changes.outputs.cpp == 'true' ||
- needs.detect-changes.outputs.java == 'true' ||
- needs.detect-changes.outputs.tools_it == 'true'
+ needs.detect-changes.outputs.full_run == 'true' ||
+ needs.detect-changes.outputs.cpp == 'true'
strategy:
fail-fast: false
matrix:
@@ -444,7 +513,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Setup java
run: |
sudo apt-get update
@@ -531,10 +600,13 @@ jobs:
tpc-test-ubuntu-randomkill:
needs: [detect-changes, build-native-lib-centos-7]
+ # Chaos/stress tier (see tpc-test-ubuntu-oom): random task kills mainly
+ # probe native/JNI shutdown robustness, and at 3 shards this is the single
+ # most expensive TPC job (~190 runner-minutes). Per-PR only for native
+ # changes; nightly full_run covers the rest.
if: >-
- needs.detect-changes.outputs.cpp == 'true' ||
- needs.detect-changes.outputs.java == 'true' ||
- needs.detect-changes.outputs.tools_it == 'true'
+ needs.detect-changes.outputs.full_run == 'true' ||
+ needs.detect-changes.outputs.cpp == 'true'
strategy:
fail-fast: false
matrix:
@@ -559,7 +631,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Setup java
run: |
sudo apt-get update
@@ -601,7 +673,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Install Uniffle ${{ matrix.uniffle }}
run: |
export JAVA_HOME=/usr/lib/jvm/java-1.8.0-openjdk
@@ -641,7 +713,14 @@ jobs:
fail-fast: false
matrix:
spark: [ "spark-3.3" ]
- celeborn: [ "celeborn-0.6.3", "celeborn-0.5.4"]
+ # Both shuffle writers stay per-PR (they exercise different Gluten
+ # writer code paths); the older Celeborn release is client-compat
+ # coverage that the nightly full_run re-checks daily instead of every
+ # PR paying for both versions.
+ celeborn: >-
+ ${{ fromJSON(needs.detect-changes.outputs.full_run == 'true'
+ && '["celeborn-0.6.3", "celeborn-0.5.4"]'
+ || '["celeborn-0.6.3"]') }}
writer: ["sort", "hash"]
runs-on: ubuntu-22.04
container: centos:8
@@ -651,7 +730,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Update mirror list
run: |
sed -i -e "s|mirrorlist=|#mirrorlist=|g" /etc/yum.repos.d/CentOS-*
|| true
@@ -724,7 +803,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Prepare
run: |
dnf module -y install python39 && \
@@ -781,7 +860,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Build and Run unit test for Spark 3.3.1 (slow tests)
run: |
@@ -830,7 +909,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Prepare spark.test.home for Spark 3.4.4 (other tests)
run: |
dnf module -y install python39 && \
@@ -901,7 +980,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Build and Run unit test for Spark 3.4.4 (slow tests, ${{
matrix.suite }})
run: |
cd $GITHUB_WORKSPACE/
@@ -955,7 +1034,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Prepare
run: |
dnf module -y install python39 && \
@@ -1027,7 +1106,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Prepare
run: |
dnf module -y install python39 && \
@@ -1093,7 +1172,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Build and Run unit test for Spark 3.5.5 (slow tests, ${{
matrix.suite }})
run: |
cd $GITHUB_WORKSPACE/
@@ -1331,7 +1410,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Prepare
run: |
dnf module -y install python39 && \
@@ -1397,7 +1476,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Build and Run unit test for Spark 4.0 (slow tests, ${{
matrix.suite }})
run: |
cd $GITHUB_WORKSPACE/
@@ -1454,7 +1533,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Prepare
run: |
dnf install -y python3.11 python3.11-pip python3.11-devel && \
@@ -1526,7 +1605,7 @@ jobs:
uses: actions/download-artifact@v8
with:
name: velox-native-lib-centos-7-${{github.sha}}
- path: ./cpp/build/
+ path: ./cpp/build/releases/
- name: Build and Run unit test for Spark 4.1 (slow tests, ${{
matrix.suite }})
run: |
cd $GITHUB_WORKSPACE/
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
