Add a GHA to upload benchmark results (#5742)

This adds a GHA to upload benchmark results. I'm using the existing
`torchci-oss-ci-benchmark` database for now until the new generic
database from https://fburl.com/gdoc/ossgtvte is ready. So, it's backed
by dynamoDB instead of S3 for now.

I also include a quick test with some dummy Android benchmark records.

* [x] Get the list of artifacts
#5727
* [x] On ExecuTorch side, extract the benchmark JSON from the artifacts
pytorch/executorch#5808
* [x] Create a new GitHub action on test-infra to upload the JSON
#5742
* [x] Create a new GHA role to upload the results
meta-pytorch/pytorch-gha-infra#483

Author: huydhn

.github/actions/upload-benchmark-results/action.yml

@@ -0,0 +1,36 @@
+name: Upload benchmark results
+
+inputs:
+  benchmark-results-dir:
+    description: 'The path to the directory with all the results in JSON format'
+    required: True
+  dry-run:
+    default: 'true'
+
+runs:
+  using: composite
+  steps:
+    - name: Install dependencies
+      shell: bash
+      run: |
+        set -eux
+        python3 -mpip install boto3==1.35.33
+
+    # TODO (huydhn): Once the generic benchmark database is ready, this will be
+    # uploaded to S3 instead
+    - name: Upload benchmark results to DynamoDB
+      shell: bash
+      env:
+        BENCHMARK_RESULTS_DIR: ${{ inputs.benchmark-results-dir }}
+        DRY_RUN: ${{ inputs.dry-run }}
+      run: |
+        set -eux
+
+        if [[ "${DRY_RUN}" == "true" ]]; then
+          python3 "${GITHUB_ACTION_PATH}/../../scripts/upload_benchmark_results.py" \
+            --benchmark-results-dir "${BENCHMARK_RESULTS_DIR}" \
+            --dry-run
+        else
+          python3 "${GITHUB_ACTION_PATH}/../../scripts/upload_benchmark_results.py" \
+            --benchmark-results-dir "${BENCHMARK_RESULTS_DIR}"
+        fi

.github/scripts/benchmark-results-dir-for-testing/android-artifacts-31017223108.json

@@ -0,0 +1 @@
+[{"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "load_status", "actual": 0.0, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "model_load_time(ms)", "actual": 286.98526, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "generate_time(ms)", "actual": 405.055521, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "token_per_sec", "actual": 306.53265, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "load_status", "actual": 0.0, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 13", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "model_load_time(ms)", "actual": 308.598385, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 13", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "generate_time(ms)", "actual": 522.4928639999999, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 13", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "token_per_sec", "actual": 243.51297, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 13", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "load_status", "actual": 0.0, "target": 0.0, "device": "Samsung Galaxy S22 Ultra 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 Ultra 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "model_load_time(ms)", "actual": 280.957917, "target": 0.0, "device": "Samsung Galaxy S22 Ultra 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 Ultra 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "generate_time(ms)", "actual": 1366.6017709999999, "target": 0.0, "device": "Samsung Galaxy S22 Ultra 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 Ultra 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "token_per_sec", "actual": 90.70632, "target": 0.0, "device": "Samsung Galaxy S22 Ultra 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 Ultra 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "load_status", "actual": 0.0, "target": 0.0, "device": "Samsung Galaxy S22+ 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22+ 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "model_load_time(ms)", "actual": 289.104427, "target": 0.0, "device": "Samsung Galaxy S22+ 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22+ 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "generate_time(ms)", "actual": 733.888385, "target": 0.0, "device": "Samsung Galaxy S22+ 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22+ 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223108, "name": "llama2", "dtype": "q8", "metric": "token_per_sec", "actual": 174.28572, "target": 0.0, "device": "Samsung Galaxy S22+ 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22+ 5G"}]

.github/scripts/benchmark-results-dir-for-testing/android-artifacts-31017223431.json

@@ -0,0 +1 @@
+[{"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223431, "name": "mv2 xnnpack", "dtype": "q8", "metric": "avg_inference_latency(ms)", "actual": 23.0566825, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223431, "name": "mv2 xnnpack", "dtype": "q8", "metric": "model_load_time(ms)", "actual": 18.705938, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223431, "name": "mv2 xnnpack", "dtype": "q8", "metric": "load_status", "actual": 0.0, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223431, "name": "mv2 xnnpack", "dtype": "q8", "metric": "avg_inference_latency(ms)", "actual": 42.8922188, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 13", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223431, "name": "mv2 xnnpack", "dtype": "q8", "metric": "model_load_time(ms)", "actual": 14.722292, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 13", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223431, "name": "mv2 xnnpack", "dtype": "q8", "metric": "load_status", "actual": 0.0, "target": 0.0, "device": "Samsung Galaxy S22 5G", "arch": "Android 13", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223431, "name": "mv2 xnnpack", "dtype": "q8", "metric": "avg_inference_latency(ms)", "actual": 10.9387811, "target": 0.0, "device": "Samsung Galaxy S22 Ultra 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 Ultra 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223431, "name": "mv2 xnnpack", "dtype": "q8", "metric": "model_load_time(ms)", "actual": 13.23677, "target": 0.0, "device": "Samsung Galaxy S22 Ultra 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 Ultra 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223431, "name": "mv2 xnnpack", "dtype": "q8", "metric": "load_status", "actual": 0.0, "target": 0.0, "device": "Samsung Galaxy S22 Ultra 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22 Ultra 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223431, "name": "mv2 xnnpack", "dtype": "q8", "metric": "avg_inference_latency(ms)", "actual": 63.4430574, "target": 0.0, "device": "Samsung Galaxy S22+ 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22+ 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223431, "name": "mv2 xnnpack", "dtype": "q8", "metric": "model_load_time(ms)", "actual": 16.301875, "target": 0.0, "device": "Samsung Galaxy S22+ 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22+ 5G"}, {"repo": "pytorch/executorch", "head_branch": "main", "workflow_id": 12345, "run_attempt": 1, "job_id": 31017223431, "name": "mv2 xnnpack", "dtype": "q8", "metric": "load_status", "actual": 0.0, "target": 0.0, "device": "Samsung Galaxy S22+ 5G", "arch": "Android 12", "filename": "android-perf", "test_name": "ANDROID_APP", "runner": "Samsung Galaxy S22+ 5G"}]

.github/scripts/upload_benchmark_results.py

@@ -0,0 +1,130 @@
+#!/usr/bin/env python3
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+import hashlib
+import json
+import logging
+import os
+import time
+from argparse import Action, ArgumentParser, Namespace
+from decimal import Decimal
+
+from logging import info
+from typing import Any, Callable, Dict, List, Optional
+
+import boto3
+
+logging.basicConfig(level=logging.INFO)
+
+
+class ValidateDir(Action):
+    def __call__(
+        self,
+        parser: ArgumentParser,
+        namespace: Namespace,
+        values: Any,
+        option_string: Optional[str] = None,
+    ) -> None:
+        if os.path.isdir(values):
+            setattr(namespace, self.dest, values)
+            return
+
+        parser.error(f"{values} is not a valid directory")
+
+
+def parse_args() -> Any:
+    from argparse import ArgumentParser
+
+    parser = ArgumentParser("upload the benchmark results to OSS benchmark database")
+    parser.add_argument(
+        "--benchmark-results-dir",
+        type=str,
+        required=True,
+        action=ValidateDir,
+        help="the directory with all the benchmark results in JSON format",
+    )
+    parser.add_argument(
+        "--dry-run",
+        action="store_true",
+    )
+    parser.add_argument(
+        "--dynamodb-table",
+        type=str,
+        default="torchci-oss-ci-benchmark",
+        help="the name of the DynamoDB table to upload to",
+    )
+
+    return parser.parse_args()
+
+
+# DynamoDB use Decimal, not float
+class DecimalEncoder(json.JSONEncoder):
+    def default(self, o: Any) -> Any:
+        if isinstance(o, Decimal):
+            return str(o)
+        return super().default(o)
+
+
+# TODO (huydhn): This can be replaced by S3 path once we move to S3
+def generate_partition_key(doc: Dict[str, Any]) -> str:
+    """
+    Generate an unique partition key for the document on DynamoDB
+    """
+    repo = doc["repo"]
+    workflow_id = doc["workflow_id"]
+    job_id = doc["job_id"]
+    test_name = doc["test_name"]
+    filename = doc["filename"]
+
+    hash_content = hashlib.md5(
+        json.dumps(doc, cls=DecimalEncoder).encode("utf-8")
+    ).hexdigest()
+    return f"{repo}/{workflow_id}/{job_id}/{test_name}/{filename}/{hash_content}"
+
+
+def upload_to_dynamodb(
+    dynamodb_table: str,
+    docs: List[Any],
+    generate_partition_key: Optional[Callable[[Dict[str, Any]], str]],
+    dry_run: bool = True,
+) -> None:
+    """
+    Copied from upload stats script
+    """
+    info(f"Writing {len(docs)} documents to DynamoDB {dynamodb_table}")
+    if not dry_run:
+        # https://boto3.amazonaws.com/v1/documentation/api/latest/guide/dynamodb.html#batch-writing
+        with boto3.resource("dynamodb").Table(dynamodb_table).batch_writer() as batch:
+            for doc in docs:
+                doc["timestamp"] = int(round(time.time() * 1000))
+                if generate_partition_key:
+                    doc["dynamoKey"] = generate_partition_key(doc)
+                batch.put_item(Item=doc)
+
+
+def main() -> None:
+    args = parse_args()
+
+    for file in os.listdir(args.benchmark_results_dir):
+        if not file.endswith(".json"):
+            continue
+
+        filepath = os.path.join(args.benchmark_results_dir, file)
+        info(f"Loading {filepath}")
+
+        with open(filepath) as f:
+            upload_to_dynamodb(
+                dynamodb_table=args.dynamodb_table,
+                # NB: DynamoDB only accepts decimal number, not float
+                docs=json.load(f, parse_float=Decimal),
+                generate_partition_key=generate_partition_key,
+                dry_run=args.dry_run,
+            )
+
+
+if __name__ == "__main__":
+    main()

.github/workflows/test_upload_benchmark_results.yml

@@ -0,0 +1,20 @@
+name: Test upload-benchmark-results
+
+on:
+  pull_request:
+    paths:
+      - .github/scripts/upload_benchmark_results.py
+      - .github/workflows/test_upload_benchmark_results.ym
+      - .github/actions/upload-benchmark-results/*
+
+jobs:
+  test:
+    runs-on: linux.2xlarge
+    steps:
+      - uses: actions/checkout@v3
+
+      - name: Test upload the benchmark results
+        uses: ./.github/actions/upload-benchmark-results
+        with:
+          benchmark-results-dir: .github/scripts/benchmark-results-dir-for-testing
+          dry-run: true