io_scenario/io.py

import re
import os
import sys
import stat
import time
import json
import Queue
import os.path
import argparse
import warnings
import threading
import subprocess


class BenchmarkOption(object):
    def __init__(self, concurence, iodepth, action, blocksize, size):
        self.iodepth = iodepth
        self.action = action
        self.blocksize = blocksize
        self.concurence = concurence
        self.size = size
        self.direct_io = False
        self.use_hight_io_priority = True
        self.sync = False


def which(program):
    def is_exe(fpath):
        return os.path.isfile(fpath) and os.access(fpath, os.X_OK)

    fpath, fname = os.path.split(program)
    if fpath:
        if is_exe(program):
            return program
    else:
        for path in os.environ["PATH"].split(os.pathsep):
            path = path.strip('"')
            exe_file = os.path.join(path, program)
            if is_exe(exe_file):
                return exe_file

    return None


# ------------------------------ IOZONE SUPPORT ------------------------------


class IOZoneParser(object):
    "class to parse iozone results"

    start_tests = re.compile(r"^\s+KB\s+reclen\s+")
    resuts = re.compile(r"[\s0-9]+")
    mt_iozone_re = re.compile(r"\s+Children see throughput " +
                              r"for\s+\d+\s+(?P<cmd>.*?)\s+=\s+" +
                              r"(?P<perf>[\d.]+)\s+KB/sec")

    cmap = {'initial writers': 'write',
            'rewriters': 'rewrite',
            'initial readers': 'read',
            're-readers': 'reread',
            'random readers': 'random read',
            'random writers': 'random write',
            'readers': 'read'}

    string1 = "                           " + \
              "                   random  random    " + \
              "bkwd   record   stride                                   "

    string2 = "KB  reclen   write rewrite    " + \
              "read    reread    read   write    " + \
              "read  rewrite     read   fwrite frewrite   fread  freread"

    @classmethod
    def apply_parts(cls, parts, string, sep=' \t\n'):
        add_offset = 0
        for part in parts:
            _, start, stop = part
            start += add_offset
            add_offset = 0

            # condition splited to make pylint happy
            while stop + add_offset < len(string):

                # condition splited to make pylint happy
                if not (string[stop + add_offset] not in sep):
                    break

                add_offset += 1

            yield part, string[start:stop + add_offset]

    @classmethod
    def make_positions(cls):
        items = [i for i in cls.string2.split() if i]

        pos = 0
        cls.positions = []

        for item in items:
            npos = cls.string2.index(item, 0 if pos == 0 else pos + 1)
            cls.positions.append([item, pos, npos + len(item)])
            pos = npos + len(item)

        for itm, val in cls.apply_parts(cls.positions, cls.string1):
            if val.strip():
                itm[0] = val.strip() + " " + itm[0]

    @classmethod
    def parse_iozone_res(cls, res, mthreads=False):
        parsed_res = None

        sres = res.split('\n')

        if not mthreads:
            for pos, line in enumerate(sres[1:]):
                if line.strip() == cls.string2 and \
                            sres[pos].strip() == cls.string1.strip():
                    add_pos = line.index(cls.string2)
                    parsed_res = {}

                    npos = [(name, start + add_pos, stop + add_pos)
                            for name, start, stop in cls.positions]

                    for itm, res in cls.apply_parts(npos, sres[pos + 2]):
                        if res.strip() != '':
                            parsed_res[itm[0]] = int(res.strip())

                    del parsed_res['KB']
                    del parsed_res['reclen']
        else:
            parsed_res = {}
            for line in sres:
                rr = cls.mt_iozone_re.match(line)
                if rr is not None:
                    cmd = rr.group('cmd')
                    key = cls.cmap.get(cmd, cmd)
                    perf = int(float(rr.group('perf')))
                    parsed_res[key] = perf
        return parsed_res


IOZoneParser.make_positions()


def iozone_do_prepare(params, filename, pattern_base):
    all_files = []
    threads = int(params.concurence)
    if 1 != threads:
        filename = filename + "_{0}"
        all_files.extend(filename .format(i) for i in range(threads))
    else:
        all_files.append(filename)

    bsz = 1024 if params.size > 1024 else params.size
    if params.size % bsz != 0:
        fsz = (params.size // bsz + 1) * bsz
    else:
        fsz = params.size

    for fname in all_files:
        with open(fname, "wb") as fd:
            if fsz > 1024:
                pattern = pattern_base * 1024 * 1024
                for _ in range(int(fsz / 1024) + 1):
                    fd.write(pattern)
            else:
                fd.write(pattern_base * 1024 * fsz)
            fd.flush()
    return all_files


VERIFY_PATTERN = "\x6d"


def prepare_iozone(params, filename):
    return iozone_do_prepare(params, filename, VERIFY_PATTERN)


def do_run_iozone(params, timeout, all_files,
                  iozone_path='iozone',
                  microsecond_mode=False,
                  prepare_only=False):

    cmd = [iozone_path, "-V", str(ord(VERIFY_PATTERN))]

    if params.sync:
        cmd.append('-o')

    if params.direct_io:
        cmd.append('-I')

    if microsecond_mode:
        cmd.append('-N')

    threads = int(params.concurence)
    if 1 != threads:
        cmd.extend(('-t', str(threads), '-F'))
        cmd.extend(all_files)
    else:
        cmd.extend(('-f', all_files[0]))

    cmd.append('-i')

    if params.action == 'write':
        cmd.append("0")
    elif params.action == 'read':
        cmd.append("1")
    elif params.action == 'randwrite' or params.action == 'randread':
        cmd.append("2")
    else:
        raise ValueError("Unknown action {0!r}".format(params.action))

    cmd.extend(('-s', str(params.size)))
    cmd.extend(('-r', str(params.blocksize)))

    # no retest
    cmd.append('-+n')

    raw_res = subprocess.check_output(cmd)

    try:
        parsed_res = IOZoneParser.parse_iozone_res(raw_res, threads > 1)

        res = {}

        if params.action == 'write':
            res['bw_mean'] = parsed_res['write']
        elif params.action == 'randwrite':
            res['bw_mean'] = parsed_res['random write']
        elif params.action == 'read':
            res['bw_mean'] = parsed_res['read']
        elif params.action == 'randread':
            res['bw_mean'] = parsed_res['random read']
    except:
        raise

    return res, " ".join(cmd)


def run_iozone(benchmark, binary_path, all_files,
               timeout=None):

    if timeout is not None:
        benchmark.size = benchmark.blocksize * 50
        res_time = do_run_iozone(benchmark, timeout, all_files,
                                 iozone_path=binary_path,
                                 microsecond_mode=True)[0]

        size = (benchmark.blocksize * timeout * 1000000)
        size /= res_time["bw_mean"]
        size = (size // benchmark.blocksize + 1) * benchmark.blocksize
        benchmark.size = size

    return do_run_iozone(benchmark, timeout, all_files,
                         iozone_path=binary_path)


def install_iozone_package():
    if which('iozone'):
        return

    is_redhat = os.path.exists('/etc/centos-release')
    is_redhat = is_redhat or os.path.exists('/etc/fedora-release')
    is_redhat = is_redhat or os.path.exists('/etc/redhat-release')

    if is_redhat:
        subprocess.check_output(["yum", "install", 'iozone3'])
        return

    try:
        os_release_cont = open('/etc/os-release').read()

        is_ubuntu = "Ubuntu" in os_release_cont

        if is_ubuntu or "Debian GNU/Linux" in os_release_cont:
            subprocess.check_output(["apt-get", "install", "iozone3"])
            return
    except (IOError, OSError):
        pass

    raise RuntimeError("Unknown host OS.")


def install_iozone_static(iozone_url, dst):
    if not os.path.isfile(dst):
        import urllib
        urllib.urlretrieve(iozone_url, dst)

    st = os.stat(dst)
    os.chmod(dst, st.st_mode | stat.S_IEXEC)


def locate_iozone():
    binary_path = which('iozone')

    if binary_path is None:
        binary_path = which('iozone3')

    if binary_path is None:
        sys.stderr.write("Can't found neither iozone not iozone3 binary"
                         "Provide --binary-path or --binary-url option")
        return None

    return binary_path

# ------------------------------ FIO SUPPORT ---------------------------------


def run_fio_once(benchmark, fio_path, tmpname, timeout=None):
    if benchmark.size is None:
        raise ValueError("You need to specify file size for fio")

    cmd_line = [fio_path,
                "--name=%s" % benchmark.action,
                "--rw=%s" % benchmark.action,
                "--blocksize=%sk" % benchmark.blocksize,
                "--iodepth=%d" % benchmark.iodepth,
                "--filename=%s" % tmpname,
                "--size={0}k".format(benchmark.size),
                "--numjobs={0}".format(benchmark.concurence),
                "--runtime=60",
                "--output-format=json",
                "--sync=" + ('1' if benchmark.sync else '0')]

    if timeout is not None:
        cmd_line.append("--timeout=%d" % timeout)
        cmd_line.append("--runtime=%d" % timeout)

    if benchmark.direct_io:
        cmd_line.append("--direct=1")

    if benchmark.use_hight_io_priority:
        cmd_line.append("--prio=0")

    raw_out = subprocess.check_output(cmd_line)
    return json.loads(raw_out)["jobs"][0], " ".join(cmd_line)


def run_fio(benchmark, binary_path, all_files, timeout=None):
    job_output, cmd_line = run_fio_once(benchmark, binary_path,
                                        all_files[0], timeout)

    if benchmark.action in ('write', 'randwrite'):
        raw_result = job_output['write']
    else:
        raw_result = job_output['read']

    res = {}

    # 'bw_dev bw_mean bw_max bw_min'.split()
    for field in ["bw_mean", "iops"]:
        res[field] = raw_result[field]
    res["lat"] = raw_result["lat"]["mean"]
    res["clat"] = raw_result["clat"]["mean"]
    res["slat"] = raw_result["slat"]["mean"]

    return res, cmd_line


def locate_fio():
    return which('fio')


# ----------------------------------------------------------------------------


def locate_binary(binary_tp, binary_path):
    if binary_path is not None:
        if os.path.isfile(binary_path):
            if not os.access(binary_path, os.X_OK):
                st = os.stat(binary_path)
                os.chmod(binary_path, st.st_mode | stat.S_IEXEC)
        else:
            binary_path = None

    if binary_path is not None:
        return binary_path

    if 'iozone' == binary_tp:
        return locate_iozone()
    else:
        return locate_fio()


def run_benchmark(binary_tp, *argv, **kwargs):
    if 'iozone' == binary_tp:
        return run_iozone(*argv, **kwargs)
    else:
        return run_fio(*argv, **kwargs)


def prepare_benchmark(binary_tp, *argv, **kwargs):
    if 'iozone' == binary_tp:
        return {'all_files': prepare_iozone(*argv, **kwargs)}
    else:
        return {}


def type_size(string):
    try:
        return re.match("\d+[KGBM]?", string, re.I).group(0)
    except:
        msg = "{0!r} don't looks like size-description string".format(string)
        raise ValueError(msg)


def type_size_ext(string):
    if string.startswith("x"):
        int(string[1:])
        return string

    if string.startswith("r"):
        int(string[1:])
        return string

    try:
        return re.match("\d+[KGBM]?", string, re.I).group(0)
    except:
        msg = "{0!r} don't looks like size-description string".format(string)
        raise ValueError(msg)


def ssize_to_kb(ssize):
    try:
        smap = dict(k=1, K=1, M=1024, m=1024, G=1024**2, g=1024**2)
        for ext, coef in smap.items():
            if ssize.endswith(ext):
                return int(ssize[:-1]) * coef

        if int(ssize) % 1024 != 0:
            raise ValueError()

        return int(ssize) / 1024

    except (ValueError, TypeError, AttributeError):
        tmpl = "Unknow size format {0!r} (or size not multiples 1024)"
        raise ValueError(tmpl.format(ssize))


def get_ram_size():
    try:
        with open("/proc/meminfo") as fd:
            for ln in fd:
                if "MemTotal:" in ln:
                    sz, kb = ln.split(':')[1].strip().split(" ")
                    assert kb == 'kB'
                    return int(sz)
    except (ValueError, TypeError, AssertionError):
        raise
        # return None


def parse_args(argv):
    parser = argparse.ArgumentParser(
        description="Run 'iozone' or 'fio' and return result")
    parser.add_argument(
        "--type", metavar="BINARY_TYPE",
        choices=['iozone', 'fio'], required=True)
    parser.add_argument(
        "--iodepth", metavar="IODEPTH", type=int,
        help="I/O requests queue depths", required=True)
    parser.add_argument(
        '-a', "--action", metavar="ACTION", type=str,
        help="actions to run", required=True,
        choices=["read", "write", "randread", "randwrite"])
    parser.add_argument(
        "--blocksize", metavar="BLOCKSIZE", type=type_size,
        help="single operation block size", required=True)
    parser.add_argument(
        "--timeout", metavar="TIMEOUT", type=int,
        help="runtime of a single run", default=None)
    parser.add_argument(
        "--iosize", metavar="SIZE", type=type_size_ext,
        help="file size", default=None)
    parser.add_argument(
        "-s", "--sync", default=False, action="store_true",
        help="exec sync after each write")
    parser.add_argument(
        "-d", "--direct-io", default=False, action="store_true",
        help="use O_DIRECT", dest='directio')
    parser.add_argument(
        "-t", "--sync-time", default=None, type=int, metavar="UTC_TIME",
        help="sleep till sime utc time", dest='sync_time')
    parser.add_argument(
        "--test-file", help="file path to run test on",
        default=None, dest='test_file')
    parser.add_argument(
        "--binary-path", help="path to binary, used for testing",
        default=None, dest='binary_path')
    parser.add_argument(
        "--prepare-only", default=False, action="store_true")
    parser.add_argument("--concurrency", default=1, type=int)

    parser.add_argument("--preparation-results", default="{}")

    return parser.parse_args(argv)


def sensor_thread(sensor_list, cmd_q, data_q):
    while True:
        try:
            cmd_q.get(timeout=0.5)
            data_q.put([])
            return
        except Queue.Empty:
            pass


def clean_prep(preparation_results):
    if 'all_files' in preparation_results:
        for fname in preparation_results['all_files']:
            if os.path.isfile(fname):
                os.unlink(fname)


def main(argv):
    if argv[0] == '--clean':
        clean_prep(json.loads(argv[1]))
        return 0

    argv_obj = parse_args(argv)
    argv_obj.blocksize = ssize_to_kb(argv_obj.blocksize)

    if argv_obj.iosize is not None:
        if argv_obj.iosize.startswith('x'):
            argv_obj.iosize = argv_obj.blocksize * int(argv_obj.iosize[1:])
        elif argv_obj.iosize.startswith('r'):
            rs = get_ram_size()
            if rs is None:
                sys.stderr.write("Can't determine ram size\n")
                exit(1)
            argv_obj.iosize = rs * int(argv_obj.iosize[1:])
        else:
            argv_obj.iosize = ssize_to_kb(argv_obj.iosize)

    benchmark = BenchmarkOption(argv_obj.concurrency,
                                argv_obj.iodepth,
                                argv_obj.action,
                                argv_obj.blocksize,
                                argv_obj.iosize)

    benchmark.direct_io = argv_obj.directio

    if argv_obj.sync:
        benchmark.sync = True

    if argv_obj.prepare_only:
        test_file_name = argv_obj.test_file
        if test_file_name is None:
            with warnings.catch_warnings():
                warnings.simplefilter("ignore")
                test_file_name = os.tmpnam()

        fnames = prepare_benchmark(argv_obj.type,
                                   benchmark,
                                   test_file_name)
        print json.dumps(fnames)
    else:
        preparation_results = json.loads(argv_obj.preparation_results)

        if not isinstance(preparation_results, dict):
            raise ValueError("preparation_results should be a dict" +
                             " with all-string keys")

        for key in preparation_results:
            if not isinstance(key, basestring):
                raise ValueError("preparation_results should be a dict" +
                                 " with all-string keys")

        if argv_obj.test_file and 'all_files' in preparation_results:
            raise ValueError("Either --test-file or --preparation-results" +
                             " options should be provided, not both")

        if argv_obj.test_file is not None:
            preparation_results['all_files'] = [argv_obj.test_file]

        autoremove = False
        if 'all_files' not in preparation_results:
            with warnings.catch_warnings():
                warnings.simplefilter("ignore")
                preparation_results['all_files'] = [os.tmpnam()]
                autoremove = True

        binary_path = locate_binary(argv_obj.type,
                                    argv_obj.binary_path)

        if binary_path is None:
            sys.stderr.write("Can't locate binary {0}\n".format(argv_obj.type))
            return 1

        try:
            if argv_obj.sync_time is not None:
                dt = argv_obj.sync_time - time.time()
                if dt > 0:
                    time.sleep(dt)

            res, cmd = run_benchmark(argv_obj.type,
                                     benchmark=benchmark,
                                     binary_path=binary_path,
                                     timeout=argv_obj.timeout,
                                     **preparation_results)

            res['__meta__'] = benchmark.__dict__.copy()
            res['__meta__']['cmdline'] = cmd

            import pprint
            sys.stdout.write(pprint.pformat(res))

            # sys.stdout.write(json.dumps(res))
            if not argv_obj.prepare_only:
                sys.stdout.write("\n")

        finally:
            if autoremove:
                clean_prep(preparation_results)


if __name__ == '__main__':
    exit(main(sys.argv[1:]))