[utilization] pipeline to create clean db records (#145327)

upload_utilization_script to generate db-ready-insert records to s3
- generate two files: metadata and timeseries in ossci-utilization buckets
- convert log record to db format ones
- add unit test job for tools/stats/

Related Prs:
setup composite action for data pipeline: https://github.com/pytorch/pytorch/pull/145310
add permission for composite action to access S3 bucket: https://github.com/pytorch-labs/pytorch-gha-infra/pull/595
add insert logic in s3 replicator: https://github.com/pytorch/test-infra/pull/6217
Pull Request resolved: https://github.com/pytorch/pytorch/pull/145327
Approved by: https://github.com/huydhn

Co-authored-by: Huy Do <huydhn@gmail.com>

tools/stats/upload_stats_lib.py

@@ -23,6 +23,9 @@ def get_s3_resource() -> Any:
     return boto3.resource("s3")
 
 
+GHA_ARTIFACTS_BUCKET = "gha-artifacts"
+
+
 # NB: In CI, a flaky test is usually retried 3 times, then the test file would be rerun
 # 2 more times
 MAX_RETRY_IN_NON_DISABLED_MODE = 3 * 3
@@ -84,16 +87,22 @@ def _download_artifact(
 
 
 def download_s3_artifacts(
-    prefix: str, workflow_run_id: int, workflow_run_attempt: int
+    prefix: str,
+    workflow_run_id: int,
+    workflow_run_attempt: int,
+    job_id: Optional[int] = None,
 ) -> list[Path]:
-    bucket = get_s3_resource().Bucket("gha-artifacts")
+    bucket = get_s3_resource().Bucket(GHA_ARTIFACTS_BUCKET)
     objs = bucket.objects.filter(
         Prefix=f"pytorch/pytorch/{workflow_run_id}/{workflow_run_attempt}/artifact/{prefix}"
     )
-
     found_one = False
     paths = []
     for obj in objs:
+        object_name = Path(obj.key).name
+        # target an artifact for a specific job_id if provided, otherwise skip the download.
+        if job_id is not None and str(job_id) not in object_name:
+            continue
         found_one = True
         p = Path(Path(obj.key).name)
         print(f"Downloading {p}")