This is an automated email from the ASF dual-hosted git repository.
potiuk pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/airflow.git
The following commit(s) were added to refs/heads/main by this push:
new 586e658ee30 Run lang-SDK K8s system test once in a dedicated job, not
6x per run (#69500)
586e658ee30 is described below
commit 586e658ee30ef650c1affd29ff11149be22d94a1
Author: Jarek Potiuk <[email protected]>
AuthorDate: Tue Jul 7 02:53:21 2026 +0200
Run lang-SDK K8s system test once in a dedicated job, not 6x per run
(#69500)
The Multi-Lang KubernetesExecutor coordinator system test was gated on
KubernetesExecutor + use-standard-naming==false, but that condition is true
once per kubernetes-combo, so it ran redundantly across all six Python/K8s
combos on ci-amd (and again on ci-arm) - adding ~4 minutes of Go/Java build
and coordinator run to each of those six system-test jobs and pushing them
to
~20 minutes, well above the rest of the matrix.
The test is executor/naming/version-independent, so run it once instead: a
dedicated tests-kubernetes-lang-sdk job on a single default
Python-Kubernetes
combo that provisions the native Go/Java toolchains (cached) and runs only
the
lang-SDK test (-k test_lang_sdk_combined_dag_succeeds), not the full suite
or
the upgrade path. The regular system-test matrix no longer runs it at all,
so
those six jobs drop back to ~15 minutes.
Also scope the Go/Java build caches by runner.arch: the previous key used
only
runner.os (Linux), which would collide between amd64 and arm64 runners now
that
the dedicated job runs on both.
---
.github/workflows/ci-amd.yml | 2 +
.github/workflows/ci-arm.yml | 2 +
.github/workflows/k8s-tests.yml | 133 +++++++++++++++++++++++++++---------
kubernetes-tests/lang_sdk/README.md | 34 +++++----
4 files changed, 126 insertions(+), 45 deletions(-)
diff --git a/.github/workflows/ci-amd.yml b/.github/workflows/ci-amd.yml
index 750c430b780..ab0a1a7ff60 100644
--- a/.github/workflows/ci-amd.yml
+++ b/.github/workflows/ci-amd.yml
@@ -934,6 +934,8 @@ jobs:
debug-resources: ${{ needs.build-info.outputs.debug-resources }}
kubernetes-combos: ${{ needs.build-info.outputs.kubernetes-combos }}
java-sdk-version: ${{ needs.build-info.outputs.java-sdk-version }}
+ lang-sdk-kubernetes-combo: >-
+ ${{ needs.build-info.outputs.default-python-version }}-${{
needs.build-info.outputs.default-kubernetes-version }}
if: >
( needs.build-info.outputs.run-kubernetes-tests == 'true' ||
needs.build-info.outputs.run-helm-tests == 'true')
diff --git a/.github/workflows/ci-arm.yml b/.github/workflows/ci-arm.yml
index 56385406a76..354155192a2 100644
--- a/.github/workflows/ci-arm.yml
+++ b/.github/workflows/ci-arm.yml
@@ -927,6 +927,8 @@ jobs:
debug-resources: ${{ needs.build-info.outputs.debug-resources }}
kubernetes-combos: ${{ needs.build-info.outputs.kubernetes-combos }}
java-sdk-version: ${{ needs.build-info.outputs.java-sdk-version }}
+ lang-sdk-kubernetes-combo: >-
+ ${{ needs.build-info.outputs.default-python-version }}-${{
needs.build-info.outputs.default-kubernetes-version }}
if: >
( needs.build-info.outputs.run-kubernetes-tests == 'true' ||
needs.build-info.outputs.run-helm-tests == 'true')
diff --git a/.github/workflows/k8s-tests.yml b/.github/workflows/k8s-tests.yml
index 613357004ab..54b4d3dc9c4 100644
--- a/.github/workflows/k8s-tests.yml
+++ b/.github/workflows/k8s-tests.yml
@@ -52,6 +52,10 @@ on: # yamllint disable-line rule:truthy
description: "JDK version used to build the lang-SDK Java artifacts
natively in CI"
required: true
type: string
+ lang-sdk-kubernetes-combo:
+ description: "Single Python-Kubernetes combo (<python>-<k8s-version>)
for the dedicated lang-SDK job"
+ required: true
+ type: string
permissions:
contents: read
jobs:
@@ -100,60 +104,131 @@ jobs:
make-mnt-writeable-and-cleanup: true
id: breeze
# preparing k8s environment with uv takes < 15 seconds with `uv` - there
is no point in caching it.
- # Provision the lang-SDK Go/Java toolchains on the host only for the
single variant that runs the
- # lang-SDK test (KubernetesExecutor, standard-naming off).
LANG_SDK_NATIVE_TOOLCHAIN=true then makes
- # breeze build the artifacts natively instead of in throwaway toolchain
containers, skipping the
- # image pulls and reusing the caches restored below. The module/build
and Gradle caches are keyed
+ # The lang-SDK coordinator system test is intentionally NOT run in this
matrix. It has its own
+ # dedicated single-combo job (tests-kubernetes-lang-sdk below), so it no
longer adds the ~6 minutes
+ # of Go/Java build + coordinator run to each of these system-test jobs.
+ - name: "\
+ Run complete K8S tests ${{ matrix.executor }}-${{
env.PYTHON_MAJOR_MINOR_VERSION }}-\
+ ${{env.KUBERNETES_VERSION}}-${{ matrix.use-standard-naming }}"
+ run: breeze k8s run-complete-tests --upgrade --no-copy-local-sources
+ env:
+ EXECUTOR: ${{ matrix.executor }}
+ USE_STANDARD_NAMING: ${{ matrix.use-standard-naming }}
+ VERBOSE: "false"
+ - name: "\
+ Print logs ${{ matrix.executor }}-${{ matrix.kubernetes-combo }}-\
+ ${{ matrix.use-standard-naming }}"
+ run: |
+ for file in `find /tmp/kind_logs_*/ -type f` ; do
+ echo "::group::${file}"
+ cat $file
+ echo "::endgroup::"
+ done
+ if: failure() || cancelled() || inputs.include-success-outputs ==
'true'
+ - name: "\
+ Upload KinD logs ${{ matrix.executor }}-${{ matrix.kubernetes-combo
}}-\
+ ${{ matrix.use-standard-naming }}"
+ uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a
# v7.0.1
+ with:
+ name: "\
+ kind-logs-${{ matrix.kubernetes-combo }}-${{ matrix.executor }}-\
+ ${{ matrix.use-standard-naming }}"
+ path: /tmp/kind_logs_*
+ retention-days: '7'
+ if: failure() || cancelled() || inputs.include-success-outputs ==
'true'
+ - name: "Delete clusters just in case they are left"
+ run: breeze k8s delete-cluster --all
+ if: always()
+
+ # The Multi-Lang KubernetesExecutor coordinator system test builds a Go
bundle and a Java jar and runs
+ # them as workers under KubernetesExecutor. It is
executor/naming/version-independent, so instead of
+ # bolting it onto all six ``KubernetesExecutor``/``use-standard-naming ==
false`` system-test jobs above
+ # (which ran it redundantly and added ~6 minutes each), it runs once here on
a single default combo and
+ # executes only the lang-SDK test (``-k
test_lang_sdk_combined_dag_succeeds``) - not the full suite.
+ tests-kubernetes-lang-sdk:
+ timeout-minutes: 60
+ name: "K8S Lang-SDK:${{ inputs.lang-sdk-kubernetes-combo }}"
+ runs-on: ${{ fromJSON(inputs.runners) }}
+ env:
+ DEBUG_RESOURCES: ${{ inputs.debug-resources }}
+ INCLUDE_SUCCESS_OUTPUTS: ${{ inputs.include-success-outputs }}
+ GITHUB_REPOSITORY: ${{ github.repository }}
+ GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+ GITHUB_USERNAME: ${{ github.actor }}
+ VERBOSE: "true"
+ steps:
+ - name: "Cleanup repo"
+ shell: bash
+ run: sudo rm -rf ${GITHUB_WORKSPACE}/*
+ - name: "Prepare PYTHON_MAJOR_MINOR_VERSION and KUBERNETES_VERSION"
+ id: prepare-versions
+ env:
+ KUBERNETES_COMBO: ${{ inputs.lang-sdk-kubernetes-combo }}
+ run: |
+ echo "PYTHON_MAJOR_MINOR_VERSION=${KUBERNETES_COMBO}" | sed
's/-.*//' >> $GITHUB_ENV
+ echo "KUBERNETES_VERSION=${KUBERNETES_COMBO}" | sed 's/=[^-]*-/=/'
>> $GITHUB_ENV
+ - name: "Checkout ${{ github.ref }} ( ${{ github.sha }} )"
+ uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 #
v7.0.0
+ with:
+ persist-credentials: false
+ # env.PYTHON_MAJOR_MINOR_VERSION, env.KUBERNETES_VERSION are set in the
previous
+ # step id: prepare-versions
+ - name: "Prepare breeze & PROD image: ${{ env.PYTHON_MAJOR_MINOR_VERSION
}}"
+ uses: ./.github/actions/prepare_breeze_and_image
+ with:
+ platform: ${{ inputs.platform }}
+ image-type: "prod"
+ python: ${{ env.PYTHON_MAJOR_MINOR_VERSION }}
+ use-uv: ${{ inputs.use-uv }}
+ make-mnt-writeable-and-cleanup: true
+ id: breeze
+ # Provision the lang-SDK Go/Java toolchains on the host so breeze builds
the artifacts natively
+ # (LANG_SDK_NATIVE_TOOLCHAIN=true) instead of in throwaway toolchain
containers, skipping the image
+ # pulls and reusing the caches restored below. The module/build and
Gradle caches are keyed
# explicitly (rather than via the setup-* built-in caching) so the key
carries an ``-vN-`` salt:
# bump it to force-invalidate a poisoned cache without waiting for a
dependency change to rotate it.
# keep go-version in sync with go-sdk/go.mod and the sibling Go SDK job
in ci-amd.yml.
- name: "Setup Go for lang-SDK build"
- if: matrix.executor == 'KubernetesExecutor' &&
matrix.use-standard-naming == false
uses: actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16 #
v6.5.0
with:
go-version: 1.25
cache: false
- name: "Cache lang-SDK Go module + build cache"
- if: matrix.executor == 'KubernetesExecutor' &&
matrix.use-standard-naming == false
uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
with:
path: |
~/.cache/go-build
~/go/pkg/mod
- key: lang-sdk-go-v1-${{ runner.os }}-${{
hashFiles('kubernetes-tests/lang_sdk/go_example/go.sum') }}
+ key: lang-sdk-go-v1-${{ runner.os }}-${{ runner.arch }}-${{
hashFiles('kubernetes-tests/lang_sdk/go_example/go.sum') }} # yamllint
disable-line rule:line-length
restore-keys: |
- lang-sdk-go-v1-${{ runner.os }}-
+ lang-sdk-go-v1-${{ runner.os }}-${{ runner.arch }}-
- name: "Setup Java for lang-SDK build"
- if: matrix.executor == 'KubernetesExecutor' &&
matrix.use-standard-naming == false
uses: actions/setup-java@1bcf9fb12cf4aa7d266a90ae39939e61372fe520 #
v5.4.0
with:
distribution: 'temurin'
java-version: ${{ inputs.java-sdk-version }}
- name: "Cache lang-SDK Gradle distribution + dependencies"
- if: matrix.executor == 'KubernetesExecutor' &&
matrix.use-standard-naming == false
uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
with:
path: |
~/.gradle/caches
~/.gradle/wrapper
- key: lang-sdk-gradle-v1-${{ runner.os }}-${{
hashFiles('java-sdk/**/*.gradle*', 'java-sdk/**/gradle-wrapper.properties',
'kubernetes-tests/lang_sdk/java_example/**/*.gradle*') }} # yamllint
disable-line rule:line-length
+ key: lang-sdk-gradle-v1-${{ runner.os }}-${{ runner.arch }}-${{
hashFiles('java-sdk/**/*.gradle*', 'java-sdk/**/gradle-wrapper.properties',
'kubernetes-tests/lang_sdk/java_example/**/*.gradle*') }} # yamllint
disable-line rule:line-length
restore-keys: |
- lang-sdk-gradle-v1-${{ runner.os }}-
- - name: "\
- Run complete K8S tests ${{ matrix.executor }}-${{
env.PYTHON_MAJOR_MINOR_VERSION }}-\
- ${{env.KUBERNETES_VERSION}}-${{ matrix.use-standard-naming }}"
- run: breeze k8s run-complete-tests --upgrade --no-copy-local-sources
+ lang-sdk-gradle-v1-${{ runner.os }}-${{ runner.arch }}-
+ # No --upgrade here: the redeploy/upgrade path is already exercised
across the main matrix above;
+ # this job only needs a single deploy to validate the Multi-Lang
coordinator, saving ~1.5 minutes.
+ - name: "Run lang-SDK K8S test ${{ inputs.lang-sdk-kubernetes-combo }}"
+ run: >-
+ breeze k8s run-complete-tests --no-copy-local-sources
+ -- -k test_lang_sdk_combined_dag_succeeds
env:
- EXECUTOR: ${{ matrix.executor }}
- USE_STANDARD_NAMING: ${{ matrix.use-standard-naming }}
- # Provision + run the lang-SDK coordinator system test in a single
variant only
- # (KubernetesExecutor, standard-naming off) to keep it off the other
five k8s jobs.
- RUN_LANG_SDK_K8S_TESTS: ${{ (matrix.executor == 'KubernetesExecutor'
&& matrix.use-standard-naming == false) && 'true' || 'false' }} # yamllint
disable-line rule:line-length
- LANG_SDK_NATIVE_TOOLCHAIN: ${{ (matrix.executor ==
'KubernetesExecutor' && matrix.use-standard-naming == false) && 'true' ||
'false' }} # yamllint disable-line rule:line-length
+ EXECUTOR: "KubernetesExecutor"
+ USE_STANDARD_NAMING: "false"
+ RUN_LANG_SDK_K8S_TESTS: "true"
+ LANG_SDK_NATIVE_TOOLCHAIN: "true"
VERBOSE: "false"
- - name: "\
- Print logs ${{ matrix.executor }}-${{ matrix.kubernetes-combo }}-\
- ${{ matrix.use-standard-naming }}"
+ - name: "Print logs lang-sdk-${{ inputs.lang-sdk-kubernetes-combo }}"
run: |
for file in `find /tmp/kind_logs_*/ -type f` ; do
echo "::group::${file}"
@@ -161,14 +236,10 @@ jobs:
echo "::endgroup::"
done
if: failure() || cancelled() || inputs.include-success-outputs ==
'true'
- - name: "\
- Upload KinD logs ${{ matrix.executor }}-${{ matrix.kubernetes-combo
}}-\
- ${{ matrix.use-standard-naming }}"
+ - name: "Upload KinD logs lang-sdk-${{ inputs.lang-sdk-kubernetes-combo
}}"
uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a
# v7.0.1
with:
- name: "\
- kind-logs-${{ matrix.kubernetes-combo }}-${{ matrix.executor }}-\
- ${{ matrix.use-standard-naming }}"
+ name: "kind-logs-lang-sdk-${{ inputs.lang-sdk-kubernetes-combo }}"
path: /tmp/kind_logs_*
retention-days: '7'
if: failure() || cancelled() || inputs.include-success-outputs ==
'true'
diff --git a/kubernetes-tests/lang_sdk/README.md
b/kubernetes-tests/lang_sdk/README.md
index c43a14f07dc..55e0d697935 100644
--- a/kubernetes-tests/lang_sdk/README.md
+++ b/kubernetes-tests/lang_sdk/README.md
@@ -95,19 +95,25 @@ RUN_LANG_SDK_K8S_TESTS=true breeze k8s tests --executor
KubernetesExecutor \
-- -k test_lang_sdk_combined_dag_succeeds
```
-In CI (and for a one-shot local run) steps 2-3 are folded into the standard
k8s job via
-`breeze k8s run-complete-tests --lang-sdk-test`: it provisions the lang-SDK
env after the base
-deploy, then runs the test. The `k8s-tests.yml` workflow enables it
(`RUN_LANG_SDK_K8S_TESTS=true`,
-which `--lang-sdk-test` reads) for a single variant only -- KubernetesExecutor
with standard-naming
-off -- so the other five k8s jobs skip the test. The provisioning builds (Go
bundle, Java jar, Java
-worker image) and the localstack deploy run in parallel.
+In CI (and for a one-shot local run) steps 2-3 are folded into a single
`run-complete-tests` call via
+`breeze k8s run-complete-tests --lang-sdk-test`: it provisions the lang-SDK
env after the base deploy,
+then runs the test. Rather than bolting this onto the regular k8s system-test
matrix (which ran it
+redundantly on all six `KubernetesExecutor` / standard-naming-off jobs and
added ~6 minutes each), the
+`k8s-tests.yml` workflow runs it in a **dedicated `tests-kubernetes-lang-sdk`
job** on a single default
+Python-Kubernetes combo (the `lang-sdk-kubernetes-combo` input, wired from the
`default-python-version`
+and `default-kubernetes-version` build-info outputs). That job sets
`RUN_LANG_SDK_K8S_TESTS=true`
+(which `--lang-sdk-test` reads) and runs only the lang-SDK test (`-k
+test_lang_sdk_combined_dag_succeeds`), not the full suite; the regular
system-test matrix no longer runs
+it at all. The provisioning builds (Go bundle, Java jar, Java worker image)
and the localstack deploy
+run in parallel.
By default the Go bundle and Java jar are built inside ephemeral toolchain
containers so a dev host
-needs neither Go nor a JDK installed. In CI that variant sets
`LANG_SDK_NATIVE_TOOLCHAIN=true`, which
-makes breeze build both artifacts with the host `go` / `./gradlew` instead:
the workflow installs the
-toolchains via `actions/setup-go` and `actions/setup-java` and restores the Go
module/build cache and
-the Gradle distribution + dependency cache with `actions/cache`, so the build
skips the per-run
-toolchain-image pulls and cold dependency downloads. The cache keys carry a
`-v1-` salt (see
-`lang-sdk-go-v1-` / `lang-sdk-gradle-v1-` in `k8s-tests.yml`) — bump it to
force-invalidate a poisoned
-cache. The JDK version comes from the `java-sdk-version` build-info output
(the `JAVA_SDK_VERSION`
-breeze constant).
+needs neither Go nor a JDK installed. In CI the dedicated job sets
`LANG_SDK_NATIVE_TOOLCHAIN=true`,
+which makes breeze build both artifacts with the host `go` / `./gradlew`
instead: the workflow installs
+the toolchains via `actions/setup-go` and `actions/setup-java` and restores
the Go module/build cache
+and the Gradle distribution + dependency cache with `actions/cache`, so the
build skips the per-run
+toolchain-image pulls and cold dependency downloads. The cache keys carry a
`-v1-` salt and a
+`runner.arch` segment (see `lang-sdk-go-v1-` / `lang-sdk-gradle-v1-` in
`k8s-tests.yml`) — bump the salt
+to force-invalidate a poisoned cache; the arch segment keeps the amd64 and
arm64 caches separate. The
+JDK version comes from the `java-sdk-version` build-info output (the
`JAVA_SDK_VERSION` breeze
+constant).
