tools/perf/benchmarks

#!/usr/bin/env python3
# Copyright (C) 2023 The Android Open Source Project
#
# 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
#
#      http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import sys
if __name__ == "__main__":
    sys.dont_write_bytecode = True

import argparse
import dataclasses
import datetime
import json
import os
import pathlib
import shutil
import subprocess
import time

import pretty
import utils


class FatalError(Exception):
    def __init__(self):
        pass


class OptionsError(Exception):
    def __init__(self, message):
        self.message = message


@dataclasses.dataclass(frozen=True)
class Lunch:
    "Lunch combination"

    target_product: str
    "TARGET_PRODUCT"

    target_release: str
    "TARGET_RELEASE"

    target_build_variant: str
    "TARGET_BUILD_VARIANT"

    def ToDict(self):
        return {
            "TARGET_PRODUCT": self.target_product,
            "TARGET_RELEASE": self.target_release,
            "TARGET_BUILD_VARIANT": self.target_build_variant,
        }

    def Combine(self):
        return f"{self.target_product}-{self.target_release}-{self.target_build_variant}"


@dataclasses.dataclass(frozen=True)
class Change:
    "A change that we make to the tree, and how to undo it"
    label: str
    "String to print in the log when the change is made"

    change: callable
    "Function to change the source tree"

    undo: callable
    "Function to revert the source tree to its previous condition in the most minimal way possible."


@dataclasses.dataclass(frozen=True)
class Benchmark:
    "Something we measure"

    id: str
    "Short ID for the benchmark, for the command line"

    title: str
    "Title for reports"

    change: Change
    "Source tree modification for the benchmark that will be measured"

    modules: list[str]
    "Build modules to build on soong command line"

    preroll: int
    "Number of times to run the build command to stabilize"

    postroll: int
    "Number of times to run the build command after reverting the action to stabilize"


@dataclasses.dataclass(frozen=True)
class FileSnapshot:
    "Snapshot of a file's contents."

    filename: str
    "The file that was snapshottened"

    contents: str
    "The contents of the file"

    def write(self):
        "Write the contents back to the file"
        with open(self.filename, "w") as f:
            f.write(self.contents)


def Snapshot(filename):
    """Return a FileSnapshot with the file's current contents."""
    with open(filename) as f:
        contents = f.read()
    return FileSnapshot(filename, contents)


def Clean():
    """Remove the out directory."""
    def remove_out():
        if os.path.exists("out"):
            shutil.rmtree("out")
    return Change(label="Remove out", change=remove_out, undo=lambda: None)


def NoChange():
    """No change to the source tree."""
    return Change(label="No change", change=lambda: None, undo=lambda: None)


def Modify(filename, contents, before=None):
    """Create an action to modify `filename` by appending `contents` before the last instances
    of `before` in the file.

    Raises an error if `before` doesn't appear in the file.
    """
    orig = Snapshot(filename)
    if before:
        index = orig.contents.rfind(before)
        if index < 0:
            report_error(f"{filename}: Unable to find string '{before}' for modify operation.")
            raise FatalError()
    else:
        index = len(orig.contents)
    modified = FileSnapshot(filename, orig.contents[:index] + contents + orig.contents[index:])
    return Change(
            label="Modify " + filename,
            change=lambda: modified.write(),
            undo=lambda: orig.write()
        )


class BenchmarkReport():
    "Information about a run of the benchmark"

    lunch: Lunch
    "lunch combo"

    benchmark: Benchmark
    "The benchmark object."

    iteration: int
    "Which iteration of the benchmark"

    log_dir: str
    "Path the the log directory, relative to the root of the reports directory"

    preroll_duration_ns: [int]
    "Durations of the in nanoseconds."

    duration_ns: int
    "Duration of the measured portion of the benchmark in nanoseconds."

    postroll_duration_ns: [int]
    "Durations of the postrolls in nanoseconds."

    complete: bool
    "Whether the benchmark made it all the way through the postrolls."

    def __init__(self, lunch, benchmark, iteration, log_dir):
        self.lunch = lunch
        self.benchmark = benchmark
        self.iteration = iteration
        self.log_dir = log_dir
        self.preroll_duration_ns = []
        self.duration_ns = -1
        self.postroll_duration_ns = []
        self.complete = False

    def ToDict(self):
        return {
            "lunch": self.lunch.ToDict(),
            "id": self.benchmark.id,
            "title": self.benchmark.title,
            "modules": self.benchmark.modules,
            "change": self.benchmark.change.label,
            "iteration": self.iteration,
            "log_dir": self.log_dir,
            "preroll_duration_ns": self.preroll_duration_ns,
            "duration_ns": self.duration_ns,
            "postroll_duration_ns": self.postroll_duration_ns,
            "complete": self.complete,
        }

class Runner():
    """Runs the benchmarks."""

    def __init__(self, options):
        self._options = options
        self._reports = []
        self._complete = False

    def Run(self):
        """Run all of the user-selected benchmarks."""
        # Clean out the log dir or create it if necessary
        prepare_log_dir(self._options.LogDir())

        try:
            for lunch in self._options.Lunches():
                print(lunch)
                for benchmark in self._options.Benchmarks():
                    for iteration in range(self._options.Iterations()):
                        self._run_benchmark(lunch, benchmark, iteration)
            self._complete = True
        finally:
            self._write_summary()


    def _run_benchmark(self, lunch, benchmark, iteration):
        """Run a single benchmark."""
        benchmark_log_subdir = self._log_dir(lunch, benchmark, iteration)
        benchmark_log_dir = self._options.LogDir().joinpath(benchmark_log_subdir)

        sys.stderr.write(f"STARTING BENCHMARK: {benchmark.id}\n")
        sys.stderr.write(f"             lunch: {lunch.Combine()}\n")
        sys.stderr.write(f"         iteration: {iteration}\n")
        sys.stderr.write(f" benchmark_log_dir: {benchmark_log_dir}\n")

        report = BenchmarkReport(lunch, benchmark, iteration, benchmark_log_subdir)
        self._reports.append(report)

        # Preroll builds
        for i in range(benchmark.preroll):
            ns = self._run_build(lunch, benchmark_log_dir.joinpath(f"pre_{i}"), benchmark.modules)
            report.preroll_duration_ns.append(ns)

        sys.stderr.write(f"PERFORMING CHANGE: {benchmark.change.label}\n")
        if not self._options.DryRun():
            benchmark.change.change()
        try:

            # Measured build
            ns = self._run_build(lunch, benchmark_log_dir.joinpath("measured"), benchmark.modules)
            report.duration_ns = ns

            # Postroll builds
            for i in range(benchmark.preroll):
                ns = self._run_build(lunch, benchmark_log_dir.joinpath(f"post_{i}"),
                                     benchmark.modules)
                report.postroll_duration_ns.append(ns)

        finally:
            # Always undo, even if we crashed or the build failed and we stopped.
            sys.stderr.write(f"UNDOING CHANGE: {benchmark.change.label}\n")
            if not self._options.DryRun():
                benchmark.change.undo()

        self._write_summary()
        sys.stderr.write(f"FINISHED BENCHMARK: {benchmark.id}\n")

    def _log_dir(self, lunch, benchmark, iteration):
        """Construct the log directory fir a benchmark run."""
        path = f"{lunch.Combine()}/{benchmark.id}"
        # Zero pad to the correct length for correct alpha sorting
        path += ("/%0" + str(len(str(self._options.Iterations()))) + "d") % iteration
        return path

    def _run_build(self, lunch, build_log_dir, modules):
        """Builds the modules.  Saves interesting log files to log_dir.  Raises FatalError
        if the build fails.
        """
        sys.stderr.write(f"STARTING BUILD {modules}\n")

        before_ns = time.perf_counter_ns()
        if not self._options.DryRun():
            cmd = [
                "build/soong/soong_ui.bash",
                "--build-mode",
                "--all-modules",
                f"--dir={self._options.root}",
            ] + modules
            env = dict(os.environ)
            env["TARGET_PRODUCT"] = lunch.target_product
            env["TARGET_RELEASE"] = lunch.target_release
            env["TARGET_BUILD_VARIANT"] = lunch.target_build_variant
            returncode = subprocess.call(cmd, env=env)
            if returncode != 0:
                report_error(f"Build failed: {' '.join(cmd)}")
                raise FatalError()

        after_ns = time.perf_counter_ns()

        # TODO: Copy some log files.

        sys.stderr.write(f"FINISHED BUILD {modules}\n")

        return after_ns - before_ns

    def _write_summary(self):
        # Write the results, even if the build failed or we crashed, including
        # whether we finished all of the benchmarks.
        data = {
            "start_time": self._options.Timestamp().isoformat(),
            "branch": self._options.Branch(),
            "tag": self._options.Tag(),
            "benchmarks": [report.ToDict() for report in self._reports],
            "complete": self._complete,
        }
        with open(self._options.LogDir().joinpath("summary.json"), "w", encoding="utf-8") as f:
            json.dump(data, f, indent=2, sort_keys=True)


def benchmark_table(benchmarks):
    rows = [("ID", "DESCRIPTION", "REBUILD"),]
    rows += [(benchmark.id, benchmark.title, " ".join(benchmark.modules)) for benchmark in
             benchmarks]
    return rows


def prepare_log_dir(directory):
    if os.path.exists(directory):
        # If it exists and isn't a directory, fail.
        if not os.path.isdir(directory):
            report_error(f"Log directory already exists but isn't a directory: {directory}")
            raise FatalError()
        # Make sure the directory is empty. Do this rather than deleting it to handle
        # symlinks cleanly.
        for filename in os.listdir(directory):
            entry = os.path.join(directory, filename)
            if os.path.isdir(entry):
                shutil.rmtree(entry)
            else:
                os.unlink(entry)
    else:
        # Create it
        os.makedirs(directory)


class Options():
    def __init__(self):
        self._had_error = False

        # Wall time clock when we started
        self._timestamp = datetime.datetime.now(datetime.timezone.utc)

        # Move to the root of the tree right away. Everything must happen from there.
        self.root = utils.get_root()
        if not self.root:
            report_error("Unable to find root of tree from cwd.")
            raise FatalError()
        os.chdir(self.root)

        # Initialize the Benchmarks. Note that this pre-loads all of the files, etc.
        # Doing all that here forces us to fail fast if one of them can't load a required
        # file, at the cost of a small startup speed. Don't make this do something slow
        # like scan the whole tree.
        self._init_benchmarks()

        # Argument parsing
        epilog = f"""
benchmarks:
{pretty.FormatTable(benchmark_table(self._benchmarks), prefix="  ")}
"""

        parser = argparse.ArgumentParser(
                prog="benchmarks",
                allow_abbrev=False, # Don't let people write unsupportable scripts.
                formatter_class=argparse.RawDescriptionHelpFormatter,
                epilog=epilog,
                description="Run build system performance benchmarks.")
        self.parser = parser

        parser.add_argument("--log-dir",
                            help="Directory for logs. Default is $TOP/../benchmarks/.")
        parser.add_argument("--dated-logs", action="store_true",
                            help="Append timestamp to log dir.")
        parser.add_argument("-n", action="store_true", dest="dry_run",
                            help="Dry run. Don't run the build commands but do everything else.")
        parser.add_argument("--tag",
                            help="Variant of the run, for when there are multiple perf runs.")
        parser.add_argument("--lunch", nargs="*",
                            help="Lunch combos to test")
        parser.add_argument("--iterations", type=int, default=1,
                            help="Number of iterations of each test to run.")
        parser.add_argument("--branch", type=str,
                            help="Specify branch. Otherwise a guess will be made based on repo.")
        parser.add_argument("--benchmark", nargs="*", default=[b.id for b in self._benchmarks],
                            metavar="BENCHMARKS",
                            help="Benchmarks to run.  Default suite will be run if omitted.")

        self._args = parser.parse_args()

        self._branch = self._branch()
        self._log_dir = self._log_dir()
        self._lunches = self._lunches()

        # Validate the benchmark ids
        all_ids = [benchmark.id for benchmark in self._benchmarks]
        bad_ids = [id for id in self._args.benchmark if id not in all_ids]
        if bad_ids:
            for id in bad_ids:
                self._error(f"Invalid benchmark: {id}")

        if self._had_error:
            raise FatalError()

    def Timestamp(self):
        return self._timestamp

    def _branch(self):
        """Return the branch, either from the command line or by guessing from repo."""
        if self._args.branch:
            return self._args.branch
        try:
            branch = subprocess.check_output(f"cd {self.root}/.repo/manifests"
                        + " && git rev-parse --abbrev-ref --symbolic-full-name @{u}",
                    shell=True, encoding="utf-8")
            return branch.strip().split("/")[-1]
        except subprocess.CalledProcessError as ex:
            report_error("Can't get branch from .repo dir. Specify --branch argument")
            report_error(str(ex))
            raise FatalError()

    def Branch(self):
        return self._branch

    def _log_dir(self):
        "The log directory to use, based on the current options"
        if self._args.log_dir:
            d = pathlib.Path(self._args.log_dir).resolve().absolute()
        else:
            d = self.root.joinpath("..", utils.DEFAULT_REPORT_DIR)
        if self._args.dated_logs:
            d = d.joinpath(self._timestamp.strftime('%Y-%m-%d'))
        d = d.joinpath(self._branch)
        if self._args.tag:
            d = d.joinpath(self._args.tag)
        return d.resolve().absolute()

    def LogDir(self):
        return self._log_dir

    def Benchmarks(self):
        return [b for b in self._benchmarks if b.id in self._args.benchmark]

    def Tag(self):
        return self._args.tag

    def DryRun(self):
        return self._args.dry_run

    def _lunches(self):
        def parse_lunch(lunch):
            parts = lunch.split("-")
            if len(parts) != 3:
                raise OptionsError(f"Invalid lunch combo: {lunch}")
            return Lunch(parts[0], parts[1], parts[2])
        # If they gave lunch targets on the command line use that
        if self._args.lunch:
            result = []
            # Split into Lunch objects
            for lunch in self._args.lunch:
                try:
                    result.append(parse_lunch(lunch))
                except OptionsError as ex:
                    self._error(ex.message)
            return result
        # Use whats in the environment
        product = os.getenv("TARGET_PRODUCT")
        release = os.getenv("TARGET_RELEASE")
        variant = os.getenv("TARGET_BUILD_VARIANT")
        if (not product) or (not release) or (not variant):
            # If they didn't give us anything, fail rather than guessing. There's no good
            # default for AOSP.
            self._error("No lunch combo specified. Either pass --lunch argument or run lunch.")
            return []
        return [Lunch(product, release, variant),]

    def Lunches(self):
        return self._lunches

    def Iterations(self):
        return self._args.iterations

    def _init_benchmarks(self):
        """Initialize the list of benchmarks."""
        # Assumes that we've already chdired to the root of the tree.
        self._benchmarks = [
            Benchmark(id="full",
                title="Full build",
                change=Clean(),
                modules=["droid"],
                preroll=0,
                postroll=3
            ),
            Benchmark(id="nochange",
                title="No change",
                change=NoChange(),
                modules=["droid"],
                preroll=2,
                postroll=3
            ),
            Benchmark(id="modify_bp",
                title="Modify Android.bp",
                change=Modify("bionic/libc/Android.bp", "// Comment"),
                modules=["droid"],
                preroll=1,
                postroll=3
            ),
        ]

    def _error(self, message):
        report_error(message)
        self._had_error = True


def report_error(message):
    sys.stderr.write(f"error: {message}\n")


def main(argv):
    try:
        options = Options()
        runner = Runner(options)
        runner.Run()
    except FatalError:
        sys.stderr.write(f"FAILED\n")


if __name__ == "__main__":
    main(sys.argv)