koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 1 | import array |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 2 | import os.path |
| 3 | import logging |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 4 | from typing import cast, Any, Tuple, List |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 5 | |
koder aka kdanilov | 6ab4d43 | 2015-06-22 00:26:28 +0300 | [diff] [blame] | 6 | import wally |
koder aka kdanilov | 3b4da8b | 2016-10-17 00:17:53 +0300 | [diff] [blame] | 7 | |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 8 | from ...utils import StopTestError, ssize2b, b2ssize |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 9 | from ...node_interfaces import IRPCNode |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 10 | from ...node_utils import get_os |
koder aka kdanilov | f286517 | 2016-12-30 03:35:11 +0200 | [diff] [blame] | 11 | from ..itest import ThreadedTest |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 12 | from ...result_classes import TimeSeries, DataSource, TestJobConfig |
koder aka kdanilov | f286517 | 2016-12-30 03:35:11 +0200 | [diff] [blame] | 13 | from .fio_task_parser import execution_time, fio_cfg_compile, FioJobConfig, FioParams, get_log_files |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 14 | from . import rpc_plugin |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 15 | from .fio_hist import expected_lat_bins |
koder aka kdanilov | 7f59d56 | 2016-12-26 01:34:23 +0200 | [diff] [blame] | 16 | |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 17 | |
| 18 | logger = logging.getLogger("wally") |
| 19 | |
| 20 | |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 21 | class FioTest(ThreadedTest): |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 22 | soft_runcycle = 5 * 60 |
Michael Semenov | 8ba6e23 | 2015-08-28 10:57:18 +0000 | [diff] [blame] | 23 | retry_time = 30 |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 24 | configs_dir = os.path.dirname(__file__) # type: str |
koder aka kdanilov | 7f59d56 | 2016-12-26 01:34:23 +0200 | [diff] [blame] | 25 | name = 'fio' |
koder aka kdanilov | f286517 | 2016-12-30 03:35:11 +0200 | [diff] [blame] | 26 | job_config_cls = FioJobConfig |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 27 | |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 28 | def __init__(self, *args, **kwargs) -> None: |
| 29 | super().__init__(*args, **kwargs) |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 30 | |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 31 | get = self.suite.params.get |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 32 | |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 33 | self.remote_task_file = self.join_remote("task.fio") |
| 34 | self.remote_output_file = self.join_remote("fio_result.json") |
| 35 | self.use_system_fio = get('use_system_fio', False) # type: bool |
| 36 | self.use_sudo = get("use_sudo", True) # type: bool |
| 37 | self.force_prefill = get('force_prefill', False) # type: bool |
| 38 | |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 39 | self.load_profile_name = self.suite.params['load'] # type: str |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 40 | |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 41 | if os.path.isfile(self.load_profile_name): |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 42 | self.load_profile_path = self.load_profile_name # type: str |
koder aka kdanilov | 6ab4d43 | 2015-06-22 00:26:28 +0300 | [diff] [blame] | 43 | else: |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 44 | self.load_profile_path = os.path.join(self.configs_dir, self.load_profile_name+ '.cfg') |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 45 | |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 46 | self.load_profile = open(self.load_profile_path, 'rt').read() # type: str |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 47 | |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 48 | if self.use_system_fio: |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 49 | self.fio_path = "fio" # type: str |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 50 | else: |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 51 | self.fio_path = os.path.join(self.suite.remote_dir, "fio") |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 52 | |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 53 | self.load_params = self.suite.params['params'] |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 54 | self.file_name = self.load_params['FILENAME'] |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 55 | |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 56 | if 'FILESIZE' not in self.load_params: |
| 57 | logger.debug("Getting test file sizes on all nodes") |
| 58 | try: |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 59 | sizes = {node.conn.fs.file_stat(self.file_name)[b'size'] |
| 60 | for node in self.suite.nodes} |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 61 | except Exception: |
| 62 | logger.exception("FILESIZE is not set in config file and fail to detect it." + |
| 63 | "Set FILESIZE or fix error and rerun test") |
| 64 | raise StopTestError() |
koder aka kdanilov | a94dfe1 | 2015-08-19 13:04:51 +0300 | [diff] [blame] | 65 | |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 66 | if len(sizes) != 1: |
| 67 | logger.error("IO target file %r has different sizes on test nodes - %r", |
| 68 | self.file_name, sizes) |
| 69 | raise StopTestError() |
koder aka kdanilov | a94dfe1 | 2015-08-19 13:04:51 +0300 | [diff] [blame] | 70 | |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 71 | self.file_size = list(sizes)[0] |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 72 | logger.info("Detected test file size is %sB", b2ssize(self.file_size)) |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 73 | self.load_params['FILESIZE'] = self.file_size |
| 74 | else: |
| 75 | self.file_size = ssize2b(self.load_params['FILESIZE']) |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 76 | |
koder aka kdanilov | f286517 | 2016-12-30 03:35:11 +0200 | [diff] [blame] | 77 | self.job_configs = list(fio_cfg_compile(self.load_profile, self.load_profile_path, |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 78 | cast(FioParams, self.load_params))) |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 79 | |
koder aka kdanilov | f286517 | 2016-12-30 03:35:11 +0200 | [diff] [blame] | 80 | if len(self.job_configs) == 0: |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 81 | logger.error("Empty fio config provided") |
| 82 | raise StopTestError() |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 83 | |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 84 | self.exec_folder = self.suite.remote_dir |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 85 | |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 86 | def config_node(self, node: IRPCNode) -> None: |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 87 | plugin_code = open(rpc_plugin.__file__.rsplit(".", 1)[0] + ".py", "rb").read() # type: bytes |
| 88 | node.upload_plugin("fio", plugin_code) |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 89 | |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 90 | try: |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 91 | node.conn.fs.rmtree(self.suite.remote_dir) |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 92 | except Exception: |
| 93 | pass |
| 94 | |
| 95 | try: |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 96 | node.conn.fs.makedirs(self.suite.remote_dir) |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 97 | except Exception: |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 98 | msg = "Failed to recreate folder {} on remote {}.".format(self.suite.remote_dir, node) |
koder aka kdanilov | 3b4da8b | 2016-10-17 00:17:53 +0300 | [diff] [blame] | 99 | logger.exception(msg) |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 100 | raise StopTestError() |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 101 | |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 102 | # TODO: check this during config validation |
| 103 | if self.file_size % (4 * (1024 ** 2)) != 0: |
| 104 | logger.error("Test file size must be proportional to 4MiB") |
| 105 | raise StopTestError() |
| 106 | |
koder aka kdanilov | 3b4da8b | 2016-10-17 00:17:53 +0300 | [diff] [blame] | 107 | self.install_utils(node) |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 108 | |
| 109 | mb = int(self.file_size / 1024 ** 2) |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 110 | logger.info("Filling test file %s on node %s with %sMiB of random data", self.file_name, node.info, mb) |
| 111 | is_prefilled, fill_bw = node.conn.fio.fill_file(self.file_name, mb, |
| 112 | force=self.force_prefill, |
| 113 | fio_path=self.fio_path) |
| 114 | |
| 115 | if not is_prefilled: |
| 116 | logger.info("Test file on node %s is already prefilled", node.info) |
| 117 | elif fill_bw is not None: |
| 118 | logger.info("Initial fio fill bw is %s MiBps for %s", fill_bw, node.info) |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 119 | |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 120 | def install_utils(self, node: IRPCNode) -> None: |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 121 | os_info = get_os(node) |
koder aka kdanilov | 6ab4d43 | 2015-06-22 00:26:28 +0300 | [diff] [blame] | 122 | if self.use_system_fio: |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 123 | if os_info.distro != 'ubuntu': |
| 124 | logger.error("Only ubuntu supported on test VM") |
| 125 | raise StopTestError() |
| 126 | node.conn.fio.install('fio', binary='fio') |
| 127 | else: |
| 128 | node.conn.fio.install('bzip2', binary='bzip2') |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 129 | fio_dir = os.path.dirname(os.path.dirname(wally.__file__)) # type: str |
| 130 | fio_dir = os.path.join(os.getcwd(), fio_dir) |
| 131 | fio_dir = os.path.join(fio_dir, 'fio_binaries') |
| 132 | fname = 'fio_{0.release}_{0.arch}.bz2'.format(os_info) |
| 133 | fio_path = os.path.join(fio_dir, fname) # type: str |
| 134 | |
| 135 | if not os.path.exists(fio_path): |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 136 | logger.error("No prebuild fio binary available for {0}".format(os_info)) |
| 137 | raise StopTestError() |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 138 | |
| 139 | bz_dest = self.join_remote('fio.bz2') # type: str |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 140 | node.copy_file(fio_path, bz_dest, compress=False) |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 141 | node.run("bzip2 --decompress {} ; chmod a+x {}".format(bz_dest, self.join_remote("fio"))) |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 142 | |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 143 | def get_expected_runtime(self, job_config: TestJobConfig) -> int: |
koder aka kdanilov | f286517 | 2016-12-30 03:35:11 +0200 | [diff] [blame] | 144 | return execution_time(cast(FioJobConfig, job_config)) |
koder aka kdanilov | 7022706 | 2016-11-26 23:23:21 +0200 | [diff] [blame] | 145 | |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 146 | def prepare_iteration(self, node: IRPCNode, job: TestJobConfig) -> None: |
| 147 | node.put_to_file(self.remote_task_file, str(job).encode("utf8")) |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 148 | |
| 149 | # TODO: get a link to substorage as a parameter |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 150 | def run_iteration(self, node: IRPCNode, job: TestJobConfig) -> List[TimeSeries]: |
| 151 | exec_time = execution_time(cast(FioJobConfig, job)) |
| 152 | |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 153 | |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 154 | fio_cmd_templ = "cd {exec_folder}; " + \ |
| 155 | "{fio_path} --output-format=json --output={out_file} --alloc-size=262144 {job_file}" |
| 156 | |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 157 | cmd = fio_cmd_templ.format(exec_folder=self.exec_folder, |
| 158 | fio_path=self.fio_path, |
| 159 | out_file=self.remote_output_file, |
| 160 | job_file=self.remote_task_file) |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 161 | must_be_empty = node.run(cmd, timeout=exec_time + max(300, exec_time), check_timeout=1).strip() |
koder aka kdanilov | bbbe1dc | 2016-12-20 01:19:56 +0200 | [diff] [blame] | 162 | |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 163 | if must_be_empty: |
| 164 | logger.error("Unexpected fio output: %r", must_be_empty) |
koder aka kdanilov | bc2c898 | 2015-06-13 02:50:43 +0300 | [diff] [blame] | 165 | |
koder aka kdanilov | f286517 | 2016-12-30 03:35:11 +0200 | [diff] [blame] | 166 | # put fio output into storage |
| 167 | fio_out = node.get_file_content(self.remote_output_file) |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 168 | |
| 169 | path = DataSource(suite_id=self.suite.storage_id, |
| 170 | job_id=job.storage_id, |
| 171 | node_id=node.node_id, |
| 172 | dev='fio', |
| 173 | sensor='stdout', |
| 174 | tag='json') |
| 175 | |
| 176 | self.storage.put_extra(fio_out, path) |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 177 | node.conn.fs.unlink(self.remote_output_file) |
| 178 | |
| 179 | files = [name for name in node.conn.fs.listdir(self.exec_folder)] |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 180 | result = [] |
| 181 | for name, file_path in get_log_files(cast(FioJobConfig, job)): |
| 182 | log_files = [fname for fname in files if fname.startswith(file_path)] |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 183 | if len(log_files) != 1: |
| 184 | logger.error("Found %s files, match log pattern %s(%s) - %s", |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 185 | len(log_files), file_path, name, ",".join(log_files[10:])) |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 186 | raise StopTestError() |
| 187 | |
| 188 | fname = os.path.join(self.exec_folder, log_files[0]) |
| 189 | raw_result = node.get_file_content(fname) # type: bytes |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 190 | node.conn.fs.unlink(fname) |
| 191 | |
| 192 | try: |
| 193 | log_data = raw_result.decode("utf8").split("\n") |
| 194 | except UnicodeEncodeError: |
| 195 | logger.exception("Error during parse %s fio log file - can't decode usint UTF8", name) |
| 196 | raise StopTestError() |
| 197 | |
| 198 | parsed = array.array('L' if name == 'lat' else 'Q') |
koder aka kdanilov | f286517 | 2016-12-30 03:35:11 +0200 | [diff] [blame] | 199 | times = array.array('Q') |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 200 | |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 201 | for idx, line in enumerate(log_data): |
| 202 | line = line.strip() |
| 203 | if line: |
| 204 | try: |
| 205 | time_ms_s, val_s, _, *rest = line.split(",") |
koder aka kdanilov | f286517 | 2016-12-30 03:35:11 +0200 | [diff] [blame] | 206 | times.append(int(time_ms_s.strip())) |
koder aka kdanilov | ffaf48d | 2016-12-27 02:25:29 +0200 | [diff] [blame] | 207 | |
koder aka kdanilov | 23e6bdf | 2016-12-24 02:18:54 +0200 | [diff] [blame] | 208 | if name == 'lat': |
| 209 | vals = [int(i.strip()) for i in rest] |
| 210 | |
| 211 | if len(vals) != expected_lat_bins: |
| 212 | logger.error("Expect {} bins in latency histogram, but found {} at time {}" |
| 213 | .format(expected_lat_bins, len(vals), time_ms_s)) |
| 214 | raise StopTestError() |
| 215 | |
| 216 | parsed.extend(vals) |
| 217 | else: |
| 218 | parsed.append(int(val_s.strip())) |
| 219 | except ValueError: |
| 220 | logger.exception("Error during parse %s fio log file in line %s: %r", name, idx, line) |
| 221 | raise StopTestError() |
koder aka kdanilov | ffaf48d | 2016-12-27 02:25:29 +0200 | [diff] [blame] | 222 | |
koder aka kdanilov | 108ac36 | 2017-01-19 20:17:16 +0200 | [diff] [blame^] | 223 | result.append(TimeSeries(name=name, |
| 224 | raw=raw_result, |
| 225 | second_axis_size=expected_lat_bins if name == 'lat' else 1, |
| 226 | data=parsed, |
| 227 | times=times, |
| 228 | source=path(sensor=name, tag=None))) |
| 229 | return result |
koder aka kdanilov | 7f59d56 | 2016-12-26 01:34:23 +0200 | [diff] [blame] | 230 | |
| 231 | def format_for_console(self, data: Any) -> str: |
| 232 | raise NotImplementedError() |