Keep going display on HUD: lambda to trigger log classifier when temp log comes in (#6811)

Please approve/merge
https://github.com/pytorch/test-infra/pull/6809/files first

See the PR mentioned in the README for context

The lambda is really simple but idk how else to trigger log classifier
whenever something gets uploaded to s3 cuz the log classifier is usually
triggered just by curling

Author: clee2000

.github/workflows/_lambda-do-release-runners.yml

@@ -89,7 +89,8 @@ jobs:
           { dir-name: 'oss_ci_job_queue_time',      zip-name: 'oss-ci-job-queue-time' },
           { dir-name: 'oss_ci_cur',                 zip-name: 'oss-ci-cur' },
           { dir-name: 'benchmark-results-uploader', zip-name: 'benchmark-results-uploader' },
-          { dir-name: 'pytorch-auto-revert',        zip-name: 'pytorch-auto-revert' }
+          { dir-name: 'pytorch-auto-revert',        zip-name: 'pytorch-auto-revert' },
+          { dir-name: 'keep-going-call-log-classifier', zip-name: 'keep-going-call-log-classifier' },
         ]
     name: Upload Release for ${{ matrix.dir-name }} lambda
     runs-on: ubuntu-latest

aws/lambda/keep-going-call-log-classifier/Makefile

@@ -0,0 +1,17 @@
+all: run-local
+
+clean:
+	rm -rf deployment
+	rm -rf venv
+	rm -rf deployment.zip
+
+venv/bin/python:
+	virtualenv venv
+
+deployment.zip:
+	mkdir -p deployment
+	cp lambda_function.py ./deployment/.
+	cd ./deployment && zip -q -r ../deployment.zip .
+
+.PHONY: create-deployment-package
+create-deployment-package: deployment.zip

aws/lambda/keep-going-call-log-classifier/README.md

@@ -0,0 +1,19 @@
+# Keep Going Call Log Classifier
+
+This is a simple AWS Lambda function that uploads the temporary values for
+keep-going in DynamoDB and ClickHouse by calling the log classifier when a new
+object is added to `s3://gha-artifacts/temp_logs`.
+
+Please see https://github.com/pytorch/pytorch/pull/155371 for more context.
+
+## Testing
+
+To test the Lambda function locally:
+
+```bash
+# Run test
+python test_lambda_function.py
+```
+
+Page maintainers: @pytorch/pytorch-dev-infra
+Last verified: 2025-06-24

aws/lambda/keep-going-call-log-classifier/lambda_function.py

@@ -0,0 +1,29 @@
+from typing import Any
+from urllib.error import HTTPError
+from urllib.request import urlopen
+
+
+LOG_CLASSIFIER_URL = (
+    "https://vwg52br27lx5oymv4ouejwf4re0akoeg.lambda-url.us-east-1.on.aws/"
+)
+PYTORCH_REPO = "pytorch/pytorch"
+
+
+def lambda_handler(event: Any, context: Any) -> None:
+    # Entry point for the lambda function
+    for record in event["Records"]:
+        key = record["s3"]["object"]["key"]
+        job_id = key.split("/")[-1]
+        try:
+            job_id = int(job_id)
+        except ValueError:
+            print(f"Failed to convert job id into int job_id={job_id}, key={key}")
+            continue
+        try:
+            urlopen(
+                f"{LOG_CLASSIFIER_URL}?job_id={job_id}&repo={PYTORCH_REPO}&temp_log=true"
+            )
+        except HTTPError as e:
+            print(
+                f"Failed to call log classifier for job_id={job_id}, key={key}, error={e}"
+            )

aws/lambda/keep-going-call-log-classifier/test_lambda_function.py

@@ -0,0 +1,40 @@
+import unittest
+from unittest.mock import patch
+
+from lambda_function import lambda_handler, LOG_CLASSIFIER_URL
+
+
+def get_test_event(job_id: int) -> dict:
+    return {
+        "Records": [
+            {
+                "s3": {
+                    "bucket": {"name": "gha-artifacts"},
+                    "object": {"key": f"temp_logs/{job_id}"},
+                }
+            }
+        ]
+    }
+
+
+class TestKeepGoingUploadCallLogClassifier(unittest.TestCase):
+    def test_lambda_handler(self):
+        # Mock urlopen to avoid actual HTTP requests
+        # Check urlopen called with the expected URL
+        expected_url = (
+            f"{LOG_CLASSIFIER_URL}?job_id=123345&repo=pytorch/pytorch&temp_log=true"
+        )
+        with patch("lambda_function.urlopen") as mock_urlopen:
+            lambda_handler(get_test_event(123345), None)
+            mock_urlopen.assert_called_once_with(expected_url)
+
+    def test_fails_with_invalid_job_id(self):
+        # Mock urlopen to avoid actual HTTP requests
+        # Check urlopen called with the expected URL
+        with patch("lambda_function.urlopen") as mock_urlopen:
+            lambda_handler(get_test_event("not a number"), None)
+            mock_urlopen.assert_not_called()
+
+
+if __name__ == "__main__":
+    unittest.main()