From d8ec3f8673346c0dc93e567159771f24c1bf74fc Mon Sep 17 00:00:00 2001 From: Vratko Polak Date: Fri, 8 Sep 2023 17:22:20 +0200 Subject: feat(perpatch): parse results from json + Use test names in output. - Methodology updated in subsequent change. Change-Id: I6a62f87249ea79262778f68d00f9bb81134f0b02 Signed-off-by: Vratko Polak --- resources/tools/integrated/compare_perpatch.py | 173 +++++++++++++++++-------- 1 file changed, 116 insertions(+), 57 deletions(-) (limited to 'resources/tools/integrated') diff --git a/resources/tools/integrated/compare_perpatch.py b/resources/tools/integrated/compare_perpatch.py index b4d52dcdfe..0adb6ae73e 100644 --- a/resources/tools/integrated/compare_perpatch.py +++ b/resources/tools/integrated/compare_perpatch.py @@ -1,4 +1,4 @@ -# Copyright (c) 2021 Cisco and/or its affiliates. +# Copyright (c) 2023 Cisco and/or its affiliates. # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at: @@ -13,72 +13,127 @@ """Script for determining whether per-patch perf test votes -1. -This script assumes there exist two text files with processed BMRR results, -located at hardcoded relative paths (subdirs thereof), having several lines -of json-parseable lists of float values, corresponding to testcase results. +This script expects a particular tree created on a filesystem by +per_patch_perf.sh bootstrap script, including test results +exported as json files according to a current model schema. +This script extracts the results (according to tresult type) +and joins them into one list of floats for parent and one for current. + This script then uses jumpavg library to determine whether there was a regression, progression or no change for each testcase. -If number of tests does not match, or there was a regression, + +If the set of test names does not match, or there was a regression, this script votes -1 (by exiting with code 1), otherwise it votes +1 (exit 0). """ import json +import os import sys +from typing import Dict, List + from resources.libraries.python import jumpavg -def main(): - """Execute the main logic, return the code to return as return code. +def parse(dirpath: str, fake_value: float) -> Dict[str, List[float]]: + """Looks for test jsons, extract scalar results. + + Files other than .json are skipped, jsons without test_id are skipped. + If the test failed, four fake values are used as a fake result. + + Units are ignored, as both parent and current are tested + with the same CSIT code so the unit should be identical. + + :param dirpath: Path to the directory tree to examine. + :param fail_value: Fake value to use for test cases that failed. + :type dirpath: str + :returns: Mapping from test IDs to list of measured values. + :rtype: Dict[str, List[float]] + :raises RuntimeError: On duplicate test ID or unknown test type. + """ + results = {} + for root, _, files in os.walk(dirpath): + for filename in files: + if not filename.endswith(".json"): + continue + filepath = os.path.join(root, filename) + with open(filepath, "rt", encoding="utf8") as file_in: + data = json.load(file_in) + if "test_id" not in data: + continue + name = data["test_id"] + if name in results: + raise RuntimeError(f"Duplicate: {name}") + if not data["passed"]: + results[name] = [fake_value] * 4 + continue + result_object = data["result"] + result_type = result_object["type"] + if result_type == "mrr": + results[name] = result_object["receive_rate"]["rate"]["values"] + elif result_type == "ndrpdr": + results[name] = [result_object["pdr"]["lower"]["rate"]["value"]] + elif result_type == "soak": + results[name] = [ + result_object["critical_rate"]["lower"]["rate"]["value"] + ] + elif result_type == "reconf": + results[name] = [result_object["loss"]["time"]["value"]] + elif result_type == "hoststack": + results[name] = [result_object["bandwidth"]["value"]] + else: + raise RuntimeError(f"Unknown result type: {result_type}") + return results + + +def main() -> int: + """Execute the main logic, return a number to return as the return code. + + Call parse to get parent and current data. + Use higher fake value for parent, so changes that keep a test failing + are marked as regressions. + + If there are multiple iterations, the value lists are joined. + For each test, call jumpavg.classify to detect possible regression. + + If there is at least one regression, return 3. :returns: Return code, 0 or 3 based on the comparison result. :rtype: int """ iteration = -1 - parent_iterations = list() - current_iterations = list() - num_tests = None + parent_aggregate = {} + current_aggregate = {} + test_names = None while 1: iteration += 1 - parent_lines = list() - current_lines = list() - filename = f"csit_parent/{iteration}/results.txt" - try: - with open(filename) as parent_file: - parent_lines = parent_file.readlines() - except IOError: + parent_results = {} + current_results = {} + parent_results = parse(f"csit_parent/{iteration}", fake_value=2.0) + parent_names = set(parent_results.keys()) + if test_names is None: + test_names = parent_names + if not parent_names: + # No more iterations. break - num_lines = len(parent_lines) - filename = f"csit_current/{iteration}/results.txt" - with open(filename) as current_file: - current_lines = current_file.readlines() - if num_lines != len(current_lines): - print( - f"Number of tests does not match within iteration {iteration}", - file=sys.stderr - ) - return 1 - if num_tests is None: - num_tests = num_lines - elif num_tests != num_lines: - print( - f"Number of tests does not match previous at iteration " - f"{iteration}", file=sys.stderr - ) - return 1 - parent_iterations.append(parent_lines) - current_iterations.append(current_lines) + assert parent_names == test_names, f"{parent_names} != {test_names}" + current_results = parse(f"csit_current/{iteration}", fake_value=1.0) + current_names = set(current_results.keys()) + assert ( + current_names == parent_names + ), f"{current_names} != {parent_names}" + for name in test_names: + if name not in parent_aggregate: + parent_aggregate[name] = [] + if name not in current_aggregate: + current_aggregate[name] = [] + parent_aggregate[name].extend(parent_results[name]) + current_aggregate[name].extend(current_results[name]) exit_code = 0 - for test_index in range(num_tests): - parent_values = list() - current_values = list() - for iteration_index, _ in enumerate(parent_iterations): - parent_values.extend( - json.loads(parent_iterations[iteration_index][test_index]) - ) - current_values.extend( - json.loads(current_iterations[iteration_index][test_index]) - ) + for name in test_names: + print(f"Test name: {name}") + parent_values = parent_aggregate[name] + current_values = current_aggregate[name] print(f"Time-ordered MRR values for parent build: {parent_values}") print(f"Time-ordered MRR values for current build: {current_values}") parent_values = sorted(parent_values) @@ -87,11 +142,14 @@ def main(): parent_stats = jumpavg.AvgStdevStats.for_runs(parent_values) current_stats = jumpavg.AvgStdevStats.for_runs(current_values) parent_group_list = jumpavg.BitCountingGroupList( - max_value=max_value).append_group_of_runs([parent_stats]) - combined_group_list = parent_group_list.copy( - ).extend_runs_to_last_group([current_stats]) + max_value=max_value + ).append_group_of_runs([parent_stats]) + combined_group_list = ( + parent_group_list.copy().extend_runs_to_last_group([current_stats]) + ) separated_group_list = parent_group_list.append_group_of_runs( - [current_stats]) + [current_stats] + ) print(f"Value-ordered MRR values for parent build: {parent_values}") print(f"Value-ordered MRR values for current build: {current_values}") avg_diff = (current_stats.avg - parent_stats.avg) / parent_stats.avg @@ -103,7 +161,7 @@ def main(): f" {combined_group_list[0].stats}" ) bits_diff = separated_group_list.bits - combined_group_list.bits - compared = u"longer" if bits_diff >= 0 else u"shorter" + compared = "longer" if bits_diff >= 0 else "shorter" print( f"Separate groups are {compared} than single group" f" by {abs(bits_diff)} bits" @@ -112,16 +170,17 @@ def main(): # That matters if only stats (not list of floats) are given. classified_list = jumpavg.classify([parent_values, current_values]) if len(classified_list) < 2: - print(f"Test test_index {test_index}: normal (no anomaly)") + print(f"Test {name}: normal (no anomaly)") continue anomaly = classified_list[1].comment - if anomaly == u"regression": - print(f"Test test_index {test_index}: anomaly regression") + if anomaly == "regression": + print(f"Test {name}: anomaly regression") exit_code = 3 # 1 or 2 can be caused by other errors continue - print(f"Test test_index {test_index}: anomaly {anomaly}") + print(f"Test {name}: anomaly {anomaly}") print(f"Exit code: {exit_code}") return exit_code -if __name__ == u"__main__": + +if __name__ == "__main__": sys.exit(main()) -- cgit 1.2.3-korg