SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Refactor daily CI workflow (#30012)

Browse Source 
* separate jobs

* separate jobs

* use channel name directly instead of ID

* use channel name directly instead of ID

* use channel name directly instead of ID

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
This commit is contained in:
Yih-Dar
2024-04-05 15:49:51 +02:00
committed by GitHub
parent 17cd7a9d28
commit b17b54d3dd
4 changed files with 267 additions and 137 deletions
Download Patch File Download Diff File Expand all files Collapse all files

88
utils/notification_service.py
View File

@@ -227,10 +227,13 @@ class Message:
button_text = "Check warnings (Link not found)"
# Use the workflow run link
job_link = f"https://github.com/huggingface/transformers/actions/runs/{os.environ['GITHUB_RUN_ID']}"
if "Extract warnings in CI artifacts" in github_actions_job_links:
button_text = "Check warnings"
# Use the actual job link
job_link = f"{github_actions_job_links['Extract warnings in CI artifacts']}"
for job in github_actions_jobs:
if "Extract warnings in CI artifacts" in job["name"] and job["conclusion"] == "success":
button_text = "Check warnings"
# Use the actual job link
job_link = job["html_url"]
break
huggingface_hub_warnings = [x for x in self.selected_warnings if "huggingface_hub" in x]
text = f"There are {len(self.selected_warnings)} warnings being selected."
@@ -573,7 +576,7 @@ class Message:
print(json.dumps({"blocks": blocks}))
client.chat_postMessage(
channel=os.environ["CI_SLACK_REPORT_CHANNEL_ID"],
channel=SLACK_REPORT_CHANNEL_ID,
text=text,
blocks=payload,
)
@@ -586,7 +589,7 @@ class Message:
text = f"{self.n_failures} failures out of {self.n_tests} tests," if self.n_failures else "All tests passed."
self.thread_ts = client.chat_postMessage(
channel=os.environ["CI_SLACK_REPORT_CHANNEL_ID"],
channel=SLACK_REPORT_CHANNEL_ID,
blocks=payload,
text=text,
)
@@ -712,7 +715,7 @@ class Message:
print(json.dumps({"blocks": blocks}))
client.chat_postMessage(
channel=os.environ["CI_SLACK_REPORT_CHANNEL_ID"],
channel=SLACK_REPORT_CHANNEL_ID,
text=f"Results for {job}",
blocks=blocks,
thread_ts=self.thread_ts["ts"],
@@ -735,7 +738,7 @@ class Message:
print(json.dumps({"blocks": blocks}))
client.chat_postMessage(
channel=os.environ["CI_SLACK_REPORT_CHANNEL_ID"],
channel=SLACK_REPORT_CHANNEL_ID,
text=f"Results for {job}",
blocks=blocks,
thread_ts=self.thread_ts["ts"],
@@ -749,7 +752,7 @@ class Message:
print(json.dumps({"blocks": blocks}))
client.chat_postMessage(
channel=os.environ["CI_SLACK_REPORT_CHANNEL_ID"],
channel=SLACK_REPORT_CHANNEL_ID,
text="Results for new failures",
blocks=blocks,
thread_ts=self.thread_ts["ts"],
@@ -852,6 +855,8 @@ def prepare_reports(title, header, reports, to_truncate=True):
if __name__ == "__main__":
SLACK_REPORT_CHANNEL_ID = os.environ["SLACK_REPORT_CHANNEL"]
# runner_status = os.environ.get("RUNNER_STATUS")
# runner_env_status = os.environ.get("RUNNER_ENV_STATUS")
setup_status = os.environ.get("SETUP_STATUS")
@@ -861,7 +866,8 @@ if __name__ == "__main__":
# Let's keep the lines regardig runners' status (we might be able to use them again in the future)
runner_not_available = False
runner_failed = False
setup_failed = True if setup_status is not None and setup_status != "success" else False
# Some jobs don't depend (`needs`) on the job `setup`: in this case, the status of the job `setup` is `skipped`.
setup_failed = False if setup_status in ["skipped", "success"] else True
org = "huggingface"
repo = "transformers"
@@ -929,14 +935,21 @@ if __name__ == "__main__":
Message.error_out(title, ci_title, runner_not_available, runner_failed, setup_failed)
exit(0)
arguments = sys.argv[1:][0]
try:
folder_slices = ast.literal_eval(arguments)
# Need to change from elements like `models/bert` to `models_bert` (the ones used as artifact names).
models = [x.replace("models/", "models_") for folders in folder_slices for x in folders]
except SyntaxError:
Message.error_out(title, ci_title)
raise ValueError("Errored out.")
# sys.argv[0] is always `utils/notification_service.py`.
arguments = sys.argv[1:]
# In our usage in `.github/workflows/slack-report.yml`, we always pass an argument when calling this script.
# The argument could be an empty string `""` if a job doesn't depend on the job `setup`.
if arguments[0] == "":
models = []
else:
model_list_as_str = arguments[0]
try:
folder_slices = ast.literal_eval(model_list_as_str)
# Need to change from elements like `models/bert` to `models_bert` (the ones used as artifact names).
models = [x.replace("models/", "models_") for folders in folder_slices for x in folders]
except Exception:
Message.error_out(title, ci_title)
raise ValueError("Errored out.")
github_actions_jobs = get_jobs(
workflow_run_id=os.environ["GITHUB_RUN_ID"], token=os.environ["ACCESS_REPO_INFO_TOKEN"]
@@ -1039,9 +1052,9 @@ if __name__ == "__main__":
# Additional runs
additional_files = {
"Examples directory": "run_examples_gpu",
"PyTorch pipelines": "run_tests_torch_pipeline_gpu",
"TensorFlow pipelines": "run_tests_tf_pipeline_gpu",
"Examples directory": "run_examples_gpu",
"Torch CUDA extension tests": "run_tests_torch_cuda_extensions_gpu_test_reports",
"Quantization tests": "run_tests_quantization_torch_gpu",
}
@@ -1056,6 +1069,24 @@ if __name__ == "__main__":
elif ci_event.startswith("Push CI (AMD)"):
additional_files = {}
# A map associating the job names (specified by `inputs.job` in a workflow file) with the keys of
# `additional_files`. This is used to remove some entries in `additional_files` that are not concerned by a
# specific job. See below.
job_to_test_map = {
"run_pipelines_torch_gpu": "PyTorch pipelines",
"run_pipelines_tf_gpu": "TensorFlow pipelines",
"run_examples_gpu": "Examples directory",
"run_all_tests_torch_cuda_extensions_gpu": "Torch CUDA extension tests",
"run_tests_quantization_torch_gpu": "Quantization tests",
}
# Remove some entries in `additional_files` if they are not concerned.
test_name = None
job_name = os.getenv("CI_TEST_JOB")
if job_name in job_to_test_map:
test_name = job_to_test_map[job_name]
additional_files = {k: v for k, v in additional_files.items() if k == test_name}
additional_results = {
key: {
"failed": {"unclassified": 0, "single": 0, "multi": 0},
@@ -1103,17 +1134,24 @@ if __name__ == "__main__":
{"line": line, "trace": stacktraces.pop(0)}
)
# Let's only check the warning for the model testing job. Currently, the job `run_extract_warnings` is only run
# when `inputs.job` (in the workflow file) is `run_tests_gpu`. The reason is: otherwise we need to save several
# artifacts with different names which complicates the logic for an insignificant part of the CI workflow reporting.
selected_warnings = []
if "warnings_in_ci" in available_artifacts:
directory = available_artifacts["warnings_in_ci"].paths[0]["path"]
with open(os.path.join(directory, "selected_warnings.json")) as fp:
selected_warnings = json.load(fp)
if job_name == "run_tests_gpu":
if "warnings_in_ci" in available_artifacts:
directory = available_artifacts["warnings_in_ci"].paths[0]["path"]
with open(os.path.join(directory, "selected_warnings.json")) as fp:
selected_warnings = json.load(fp)
if not os.path.isdir(os.path.join(os.getcwd(), "prev_ci_results")):
os.makedirs(os.path.join(os.getcwd(), "prev_ci_results"))
with open("prev_ci_results/model_results.json", "w", encoding="UTF-8") as fp:
json.dump(model_results, fp, indent=4, ensure_ascii=False)
# Only the model testing job is concerned: this condition is to avoid other jobs to upload the empty list as
# results.
if job_name == "run_tests_gpu":
with open("prev_ci_results/model_results.json", "w", encoding="UTF-8") as fp:
json.dump(model_results, fp, indent=4, ensure_ascii=False)
prev_ci_artifacts = None
target_workflow = "huggingface/transformers/.github/workflows/self-scheduled.yml@refs/heads/main"