From c1c0bd7a251865e82b1e8d0292451f78f19f22a8 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Jakub=20Ber=C3=A1nek?= Date: Thu, 26 Sep 2024 13:44:06 +0200 Subject: [PATCH] Upload average CPU consumption of CI jobs to DataDog --- .github/workflows/ci.yml | 10 ++++ src/ci/scripts/upload-artifacts.sh | 2 +- src/ci/scripts/upload-build-metrics.py | 81 ++++++++++++++++++++++++++ 3 files changed, 92 insertions(+), 1 deletion(-) create mode 100644 src/ci/scripts/upload-build-metrics.py diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml index 8032154a736..b6dc27f1234 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -212,6 +212,16 @@ jobs: # erroring about invalid credentials instead. if: github.event_name == 'push' || env.DEPLOY == '1' || env.DEPLOY_ALT == '1' + - name: upload job metrics to DataDog + if: needs.calculate_matrix.outputs.run_type != 'pr' + env: + DATADOG_SITE: datadoghq.com + DATADOG_API_KEY: ${{ secrets.DATADOG_API_KEY }} + DD_GITHUB_JOB_NAME: ${{ matrix.name }} + run: | + npm install -g @datadog/datadog-ci@^2.x.x + python3 src/ci/scripts/upload-build-metrics.py build/cpu-usage.csv + # This job isused to tell bors the final status of the build, as there is no practical way to detect # when a workflow is successful listening to webhooks only in our current bors implementation (homu). outcome: diff --git a/src/ci/scripts/upload-artifacts.sh b/src/ci/scripts/upload-artifacts.sh index 61c187fa77c..129ede636f3 100755 --- a/src/ci/scripts/upload-artifacts.sh +++ b/src/ci/scripts/upload-artifacts.sh @@ -23,7 +23,7 @@ if [[ "${DEPLOY-0}" -eq "1" ]] || [[ "${DEPLOY_ALT-0}" -eq "1" ]]; then fi # CPU usage statistics. -mv build/cpu-usage.csv "${upload_dir}/cpu-${CI_JOB_NAME}.csv" +cp build/cpu-usage.csv "${upload_dir}/cpu-${CI_JOB_NAME}.csv" # Build metrics generated by x.py. mv "${build_dir}/metrics.json" "${upload_dir}/metrics-${CI_JOB_NAME}.json" diff --git a/src/ci/scripts/upload-build-metrics.py b/src/ci/scripts/upload-build-metrics.py new file mode 100644 index 00000000000..a95e0949d70 --- /dev/null +++ b/src/ci/scripts/upload-build-metrics.py @@ -0,0 +1,81 @@ +""" +This script postprocesses data gathered during a CI run, computes certain metrics +from them, and uploads these metrics to DataDog. + +This script is expected to be executed from within a GitHub Actions job. + +It expects the following environment variables: +- DATADOG_SITE: path to the DataDog API endpoint +- DATADOG_API_KEY: DataDog API token +- DD_GITHUB_JOB_NAME: Name of the current GitHub Actions job + +And it also expects the presence of a binary called `datadog-ci` to be in PATH. +It can be installed with `npm install -g @datadog/datadog-ci`. + +Usage: +```bash +$ python3 upload-build-metrics.py +``` + +`path-to-CPU-usage-CSV` is a path to a CSV generated by the `src/ci/cpu-usage-over-time.py` script. +""" +import argparse +import csv +import os +import subprocess +import sys +from pathlib import Path +from typing import List + + +def load_cpu_usage(path: Path) -> List[float]: + usage = [] + with open(path) as f: + reader = csv.reader(f, delimiter=',') + for row in reader: + # The log might contain incomplete rows or some Python exception + if len(row) == 2: + try: + idle = float(row[1]) + usage.append(100.0 - idle) + except ValueError: + pass + return usage + + +def upload_datadog_measure(name: str, value: float): + """ + Uploads a single numeric metric for the current GitHub Actions job to DataDog. + """ + print(f"Metric {name}: {value:.4f}") + + datadog_cmd = "datadog-ci" + if os.getenv("GITHUB_ACTIONS") is not None and sys.platform.lower().startswith("win"): + # Due to weird interaction of MSYS2 and Python, we need to use an absolute path, + # and also specify the ".cmd" at the end. See https://github.com/rust-lang/rust/pull/125771. + datadog_cmd = "C:\\npm\\prefix\\datadog-ci.cmd" + + subprocess.run([ + datadog_cmd, + "measure", + "--level", "job", + "--measures", f"{name}:{value}" + ], + check=False + ) + + +if __name__ == "__main__": + parser = argparse.ArgumentParser( + prog="DataDog metric uploader" + ) + parser.add_argument("cpu-usage-history-csv") + args = parser.parse_args() + + build_usage_csv = vars(args)["cpu-usage-history-csv"] + usage_timeseries = load_cpu_usage(Path(build_usage_csv)) + if len(usage_timeseries) > 0: + avg_cpu_usage = sum(usage_timeseries) / len(usage_timeseries) + else: + avg_cpu_usage = 0 + upload_datadog_measure("avg-cpu-usage", avg_cpu_usage)