mirror of
https://github.com/saymrwulf/pytorch.git
synced 2026-05-14 20:57:59 +00:00
04c1c2b791
There is a bug in the test workflow where it could fail to find the new Docker image when the image hasn't yet became available on ECR, for example e71ab21422. This basically is a race condition where the test job starts before the docker-build workflow could finish successfully. The fix here is to make sure that the test job has the opportunity to build the image if it doesn't exist, same as what the build workflow does atm. Once the docker-build workflow finishes pushing the new image to ECR, that can then be used instead.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102562
Approved by: https://github.com/PaliC
369 lines
16 KiB
YAML
369 lines
16 KiB
YAML
name: linux-test
|
|
|
|
on:
|
|
workflow_call:
|
|
inputs:
|
|
build-environment:
|
|
required: true
|
|
type: string
|
|
description: Top-level label for what's being built/tested.
|
|
test-matrix:
|
|
required: true
|
|
type: string
|
|
description: JSON description of what test configs to run.
|
|
docker-image:
|
|
required: true
|
|
type: string
|
|
description: Docker image to run in.
|
|
sync-tag:
|
|
required: false
|
|
type: string
|
|
default: ""
|
|
description: |
|
|
If this is set, our linter will use this to make sure that every other
|
|
job with the same `sync-tag` is identical.
|
|
timeout-minutes:
|
|
required: false
|
|
type: number
|
|
default: 240
|
|
description: |
|
|
Set the maximum (in minutes) how long the workflow should take to finish
|
|
use-gha:
|
|
required: false
|
|
type: string
|
|
default: ""
|
|
description: If set to any value, upload to GHA. Otherwise upload to S3.
|
|
dashboard-tag:
|
|
required: false
|
|
type: string
|
|
default: ""
|
|
|
|
env:
|
|
GIT_DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
|
|
|
|
jobs:
|
|
test:
|
|
# Don't run on forked repos or empty test matrix
|
|
if: github.repository_owner == 'pytorch' && toJSON(fromJSON(inputs.test-matrix).include) != '[]'
|
|
strategy:
|
|
matrix: ${{ fromJSON(inputs.test-matrix) }}
|
|
fail-fast: false
|
|
runs-on: ${{ matrix.runner }}
|
|
timeout-minutes: ${{ inputs.timeout-minutes }}
|
|
steps:
|
|
- name: Setup SSH (Click me for login details)
|
|
uses: pytorch/test-infra/.github/actions/setup-ssh@main
|
|
if: ${{ !contains(matrix.runner, 'gcp.a100') }}
|
|
with:
|
|
github-secret: ${{ secrets.GITHUB_TOKEN }}
|
|
instructions: |
|
|
All testing is done inside the container, to start an interactive session run:
|
|
docker exec -it $(docker container ps --format '{{.ID}}') bash
|
|
|
|
- name: Checkout PyTorch
|
|
uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
|
|
|
|
- name: Setup Linux
|
|
uses: ./.github/actions/setup-linux
|
|
|
|
- name: Calculate docker image
|
|
id: calculate-docker-image
|
|
uses: ./.github/actions/calculate-docker-image
|
|
with:
|
|
docker-image-name: ${{ inputs.docker-image }}
|
|
xla: ${{ contains(inputs.build-environment, 'xla') }}
|
|
|
|
- name: Pull docker image
|
|
uses: pytorch/test-infra/.github/actions/pull-docker-image@main
|
|
with:
|
|
docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
|
|
|
|
- name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
|
|
id: install-nvidia-driver
|
|
uses: pytorch/test-infra/.github/actions/setup-nvidia@main
|
|
if: contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu')
|
|
|
|
- name: Lock NVIDIA A100 40GB Frequency
|
|
run: |
|
|
sudo nvidia-smi -pm 1
|
|
sudo nvidia-smi -ac 1215,1410
|
|
nvidia-smi
|
|
if: contains(matrix.runner, 'a100')
|
|
|
|
- name: Start monitoring script
|
|
id: monitor-script
|
|
shell: bash
|
|
continue-on-error: true
|
|
run: |
|
|
python3 -m pip install psutil==5.9.1 nvidia-ml-py==11.525.84
|
|
python3 -m tools.stats.monitor > usage_log.txt 2>&1 &
|
|
echo "monitor-script-pid=${!}" >> "${GITHUB_OUTPUT}"
|
|
|
|
- name: Download build artifacts
|
|
uses: ./.github/actions/download-build-artifacts
|
|
with:
|
|
name: ${{ inputs.build-environment }}
|
|
|
|
- name: Parse ref
|
|
id: parse-ref
|
|
run: .github/scripts/parse_ref.py
|
|
|
|
- name: Check for keep-going label
|
|
# This uses the filter-test-configs action because it conviniently
|
|
# checks for labels. It does not actually do any filtering. All
|
|
# filtering is done in the build step.
|
|
id: keep-going
|
|
uses: ./.github/actions/filter-test-configs
|
|
with:
|
|
github-token: ${{ secrets.GITHUB_TOKEN }}
|
|
test-matrix: ${{ inputs.test-matrix }}
|
|
|
|
- name: Download pytest cache
|
|
uses: ./.github/actions/pytest-cache-download
|
|
continue-on-error: true
|
|
with:
|
|
cache_dir: .pytest_cache
|
|
job_identifier: ${{ github.workflow }}_${{ inputs.build-environment }}_${{ github.job }}_${{ matrix.config }}
|
|
|
|
- name: Set Test step time
|
|
id: test-timeout
|
|
shell: bash
|
|
env:
|
|
JOB_TIMEOUT: ${{ inputs.timeout-minutes }}
|
|
run: |
|
|
echo "timeout=$((JOB_TIMEOUT-30))" >> "${GITHUB_OUTPUT}"
|
|
|
|
- name: Test
|
|
id: test
|
|
timeout-minutes: ${{ fromJson(steps.test-timeout.outputs.timeout) }}
|
|
env:
|
|
BUILD_ENVIRONMENT: ${{ inputs.build-environment }}
|
|
PR_NUMBER: ${{ github.event.pull_request.number }}
|
|
BRANCH: ${{ steps.parse-ref.outputs.branch }}
|
|
SHA1: ${{ github.event.pull_request.head.sha || github.sha }}
|
|
BASE_SHA: ${{ github.event.pull_request.base.sha || github.sha }}
|
|
PYTORCH_RETRY_TEST_CASES: 1
|
|
PYTORCH_OVERRIDE_FLAKY_SIGNAL: 1
|
|
TEST_CONFIG: ${{ matrix.config }}
|
|
SHARD_NUMBER: ${{ matrix.shard }}
|
|
NUM_TEST_SHARDS: ${{ matrix.num_shards }}
|
|
PR_BODY: ${{ github.event.pull_request.body }}
|
|
CONTINUE_THROUGH_ERROR: ${{ steps.keep-going.outputs.keep-going }}
|
|
SCCACHE_BUCKET: ossci-compiler-cache-circleci-v2
|
|
SCCACHE_S3_KEY_PREFIX: ${{ github.workflow }}
|
|
SHM_SIZE: ${{ contains(inputs.build-environment, 'cuda') && '2g' || '1g' }}
|
|
DOCKER_IMAGE: ${{ inputs.docker-image }}
|
|
XLA_CUDA: ${{ contains(inputs.build-environment, 'xla') && '0' || '' }}
|
|
XLA_CLANG_CACHE_S3_BUCKET_NAME: ossci-compiler-clang-cache-circleci-xla
|
|
PYTORCH_TEST_CUDA_MEM_LEAK_CHECK: ${{ matrix.mem_leak_check && '1' || '0' }}
|
|
PYTORCH_TEST_RERUN_DISABLED_TESTS: ${{ matrix.rerun_disabled_tests && '1' || '0' }}
|
|
DASHBOARD_TAG: ${{ inputs.dashboard-tag }}
|
|
run: |
|
|
set -x
|
|
|
|
if [[ $TEST_CONFIG == 'multigpu' ]]; then
|
|
TEST_COMMAND=.ci/pytorch/multigpu-test.sh
|
|
elif [[ $BUILD_ENVIRONMENT == *onnx* ]]; then
|
|
TEST_COMMAND=.ci/onnx/test.sh
|
|
else
|
|
TEST_COMMAND=.ci/pytorch/test.sh
|
|
fi
|
|
|
|
COMMIT_MESSAGES=$(git cherry -v "origin/${GIT_DEFAULT_BRANCH:-main}")
|
|
|
|
# sanitize the input commit message and PR body here:
|
|
#
|
|
# trim all new lines from commit messages + PR_BODY to avoid issues with batch environment
|
|
# variable copying. see https://github.com/pytorch/pytorch/pull/80043#issuecomment-1167796028
|
|
COMMIT_MESSAGES="${COMMIT_MESSAGES//[$'\n\r']}"
|
|
PR_BODY="${PR_BODY//[$'\n\r']}"
|
|
|
|
# then trim all special characters like single and double quotes to avoid unescaped inputs to
|
|
# wreak havoc internally
|
|
export COMMIT_MESSAGES="${COMMIT_MESSAGES//[\'\"]}"
|
|
export PR_BODY="${PR_BODY//[\'\"]}"
|
|
|
|
# detached container should get cleaned up by teardown_ec2_linux
|
|
# TODO: Stop building test binaries as part of the build phase
|
|
# Used for GPU_FLAG since that doesn't play nice
|
|
# shellcheck disable=SC2086,SC2090
|
|
container_name=$(docker run \
|
|
${GPU_FLAG:-} \
|
|
-e BUILD_ENVIRONMENT \
|
|
-e PR_NUMBER \
|
|
-e GITHUB_ACTIONS \
|
|
-e BASE_SHA \
|
|
-e BRANCH \
|
|
-e SHA1 \
|
|
-e AWS_DEFAULT_REGION \
|
|
-e IN_WHEEL_TEST \
|
|
-e SHARD_NUMBER \
|
|
-e TEST_CONFIG \
|
|
-e NUM_TEST_SHARDS \
|
|
-e PR_BODY \
|
|
-e COMMIT_MESSAGES \
|
|
-e CONTINUE_THROUGH_ERROR \
|
|
-e PYTORCH_RETRY_TEST_CASES \
|
|
-e PYTORCH_OVERRIDE_FLAKY_SIGNAL \
|
|
-e PR_LABELS \
|
|
-e MAX_JOBS="$(nproc --ignore=2)" \
|
|
-e SCCACHE_BUCKET \
|
|
-e SCCACHE_S3_KEY_PREFIX \
|
|
-e XLA_CUDA \
|
|
-e XLA_CLANG_CACHE_S3_BUCKET_NAME \
|
|
-e PYTORCH_TEST_CUDA_MEM_LEAK_CHECK \
|
|
-e PYTORCH_TEST_RERUN_DISABLED_TESTS \
|
|
-e SKIP_SCCACHE_INITIALIZATION=1 \
|
|
-e DASHBOARD_TAG \
|
|
--env-file="/tmp/github_env_${GITHUB_RUN_ID}" \
|
|
--ulimit stack=10485760:83886080 \
|
|
--security-opt seccomp=unconfined \
|
|
--cap-add=SYS_PTRACE \
|
|
--ipc=host \
|
|
--shm-size="${SHM_SIZE}" \
|
|
--tty \
|
|
--detach \
|
|
--name="${container_name}" \
|
|
--user jenkins \
|
|
-v "${GITHUB_WORKSPACE}:/var/lib/jenkins/workspace" \
|
|
-w /var/lib/jenkins/workspace \
|
|
"${DOCKER_IMAGE}"
|
|
)
|
|
# Propagate download.pytorch.org IP to container
|
|
grep download.pytorch.org /etc/hosts | docker exec -i "${container_name}" sudo bash -c "/bin/cat >> /etc/hosts"
|
|
echo "DOCKER_CONTAINER_ID=${container_name}" >> "${GITHUB_ENV}"
|
|
docker exec -t "${container_name}" sh -c "pip install $(echo dist/*.whl)[opt-einsum] && ${TEST_COMMAND}"
|
|
|
|
- name: Upload pytest cache if tests failed
|
|
uses: ./.github/actions/pytest-cache-upload
|
|
continue-on-error: true
|
|
if: failure() && steps.test.conclusion && steps.test.conclusion == 'failure'
|
|
with:
|
|
cache_dir: .pytest_cache
|
|
shard: ${{ matrix.shard }}
|
|
job_identifier: ${{ github.workflow }}_${{ inputs.build-environment }}_${{ github.job }}_${{ matrix.config }}
|
|
|
|
- name: Print remaining test logs
|
|
shell: bash
|
|
if: always() && steps.test.conclusion
|
|
run: |
|
|
cat test/**/*.log || true
|
|
|
|
- name: Get workflow job id
|
|
id: get-job-id
|
|
uses: ./.github/actions/get-workflow-job-id
|
|
if: always()
|
|
with:
|
|
github-token: ${{ secrets.GITHUB_TOKEN }}
|
|
|
|
- name: Stop monitoring script
|
|
if: always() && steps.monitor-script.outputs.monitor-script-pid
|
|
shell: bash
|
|
continue-on-error: true
|
|
env:
|
|
MONITOR_SCRIPT_PID: ${{ steps.monitor-script.outputs.monitor-script-pid }}
|
|
run: |
|
|
kill "$MONITOR_SCRIPT_PID"
|
|
|
|
- name: Upload test artifacts
|
|
uses: ./.github/actions/upload-test-artifacts
|
|
if: always() && steps.test.conclusion && steps.test.conclusion != 'skipped'
|
|
with:
|
|
file-suffix: ${{ github.job }}-${{ matrix.config }}-${{ matrix.shard }}-${{ matrix.num_shards }}-${{ matrix.runner }}_${{ steps.get-job-id.outputs.job-id }}
|
|
use-gha: ${{ inputs.use-gha }}
|
|
|
|
- name: Collect backtraces from coredumps (if any)
|
|
if: always()
|
|
run: |
|
|
# shellcheck disable=SC2156
|
|
find . -iname "core.[1-9]*" -exec docker exec "${DOCKER_CONTAINER_ID}" sh -c "gdb python {} -ex 'bt' -ex 'q'" \;
|
|
|
|
- name: Store Core dumps on S3
|
|
uses: seemethere/upload-artifact-s3@v5
|
|
if: failure()
|
|
with:
|
|
name: coredumps-${{ matrix.config }}-${{ matrix.shard }}-${{ matrix.num_shards }}-${{ matrix.runner }}
|
|
retention-days: 14
|
|
if-no-files-found: ignore
|
|
path: ./**/core.[1-9]*
|
|
|
|
- name: Teardown Linux
|
|
uses: pytorch/test-infra/.github/actions/teardown-linux@main
|
|
if: always()
|
|
|
|
# NB: We are currently having an intermittent GPU-related issue on G5 runners with
|
|
# A10G GPU. Once this happens, trying to reset the GPU as done in setup-nvidia does
|
|
# not seem to help. Here are some symptoms:
|
|
# * Calling nvidia-smi timeouts after 60 second
|
|
# * Fail to run nvidia-smi with an unable to determine the device handle for GPU
|
|
# unknown error
|
|
# * Test fails with a missing CUDA GPU error when initializing CUDA in PyTorch
|
|
# * Run docker --gpus all fails with error response from daemon
|
|
#
|
|
# As both the root cause and recovery path are unclear, let's take the runner out of
|
|
# service so that it doesn't get any more jobs
|
|
- name: Check NVIDIA driver installation step
|
|
if: failure() && steps.install-nvidia-driver.outcome && steps.install-nvidia-driver.outcome != 'skipped'
|
|
shell: bash
|
|
env:
|
|
RUNNER_WORKSPACE: ${{ runner.workspace }}
|
|
run: |
|
|
set +e
|
|
set -x
|
|
|
|
nvidia-smi
|
|
# NB: Surprisingly, nvidia-smi command returns successfully with return code 0 even in
|
|
# the case where the driver has already crashed as it still can get the driver version
|
|
# and some basic information like the bus ID. However, the rest of the information
|
|
# would be missing (ERR!), for example:
|
|
#
|
|
# +-----------------------------------------------------------------------------+
|
|
# | NVIDIA-SMI 525.89.02 Driver Version: 525.89.02 CUDA Version: 12.0 |
|
|
# |-------------------------------+----------------------+----------------------+
|
|
# | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
|
|
# | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|
|
# | | | MIG M. |
|
|
# |===============================+======================+======================|
|
|
# | 0 ERR! Off | 00000000:00:1E.0 Off | ERR! |
|
|
# |ERR! ERR! ERR! ERR! / ERR! | 4184MiB / 23028MiB | ERR! Default |
|
|
# | | | ERR! |
|
|
# +-------------------------------+----------------------+----------------------+
|
|
#
|
|
# +-----------------------------------------------------------------------------+
|
|
# | Processes: |
|
|
# | GPU GI CI PID Type Process name GPU Memory |
|
|
# | ID ID Usage |
|
|
# |=============================================================================|
|
|
# +-----------------------------------------------------------------------------+
|
|
#
|
|
# This should be reported as a failure instead as it will guarantee to fail when
|
|
# Docker tries to run with --gpus all
|
|
#
|
|
# So, the correct check here is to query one of the missing piece of info like
|
|
# GPU name, so that the command can fail accordingly
|
|
nvidia-smi --query-gpu=gpu_name --format=csv,noheader --id=0
|
|
NVIDIA_SMI_STATUS=$?
|
|
|
|
# These are acceptable return code from nvidia-smi as copied from setup-nvidia GitHub action
|
|
if [ "$NVIDIA_SMI_STATUS" -ne 0 ] && [ "$NVIDIA_SMI_STATUS" -ne 14 ]; then
|
|
echo "NVIDIA driver installation has failed, shutting down the runner..."
|
|
.github/scripts/stop_runner_service.sh
|
|
fi
|
|
|
|
# For runner with multiple GPUs, we also want to confirm that the number of GPUs are the
|
|
# power of 2, i.e. 1, 2, 4, or 8. This is to avoid flaky test issue when one GPU fails
|
|
# https://github.com/pytorch/test-infra/issues/4000
|
|
GPU_COUNT=$(nvidia-smi --list-gpus | wc -l)
|
|
NVIDIA_SMI_STATUS=$?
|
|
|
|
# These are acceptable return code from nvidia-smi as copied from setup-nvidia GitHub action
|
|
if [ "$NVIDIA_SMI_STATUS" -ne 0 ] && [ "$NVIDIA_SMI_STATUS" -ne 14 ]; then
|
|
echo "NVIDIA driver installation has failed, shutting down the runner..."
|
|
.github/scripts/stop_runner_service.sh
|
|
fi
|
|
|
|
# Check the GPU count to be a power of 2
|
|
if [ "$GPU_COUNT" -le 8 ] && [ "$GPU_COUNT" -ne 1 ] && [ "$GPU_COUNT" -ne 2 ] && [ "$GPU_COUNT" -ne 4 ] && [ "$GPU_COUNT" -ne 8 ]; then
|
|
echo "NVIDIA driver detects $GPU_COUNT GPUs. The runner has a broken GPU, shutting it down..."
|
|
.github/scripts/stop_runner_service.sh
|
|
fi
|