1#!/usr/bin/env python3 2 3import os 4import re 5import sys 6import json 7import paramiko 8import zipfile 9import threading 10import subprocess 11import itertools 12import time 13import uuid 14import rpc 15import rpc.client 16from common import * 17 18 19class Server: 20 def __init__(self, name, username, password, mode, nic_ips, transport): 21 self.name = name 22 self.mode = mode 23 self.username = username 24 self.password = password 25 self.nic_ips = nic_ips 26 self.transport = transport.lower() 27 28 if not re.match("^[A-Za-z0-9]*$", name): 29 self.log_print("Please use a name which contains only letters or numbers") 30 sys.exit(1) 31 32 def log_print(self, msg): 33 print("[%s] %s" % (self.name, msg), flush=True) 34 35 36class Target(Server): 37 def __init__(self, name, username, password, mode, nic_ips, transport="rdma", 38 use_null_block=False, sar_settings=None): 39 40 super(Target, self).__init__(name, username, password, mode, nic_ips, transport) 41 self.null_block = bool(use_null_block) 42 self.enable_sar = False 43 if sar_settings: 44 self.enable_sar, self.sar_delay, self.sar_interval, self.sar_count = sar_settings 45 46 self.script_dir = os.path.dirname(os.path.abspath(sys.argv[0])) 47 self.spdk_dir = os.path.abspath(os.path.join(self.script_dir, "../../../")) 48 49 def zip_spdk_sources(self, spdk_dir, dest_file): 50 self.log_print("Zipping SPDK source directory") 51 fh = zipfile.ZipFile(dest_file, "w", zipfile.ZIP_DEFLATED) 52 for root, directories, files in os.walk(spdk_dir, followlinks=True): 53 for file in files: 54 fh.write(os.path.relpath(os.path.join(root, file))) 55 fh.close() 56 self.log_print("Done zipping") 57 58 def read_json_stats(self, file): 59 with open(file, "r") as json_data: 60 data = json.load(json_data) 61 job_pos = 0 # job_post = 0 because using aggregated results 62 63 # Check if latency is in nano or microseconds to choose correct dict key 64 def get_lat_unit(key_prefix, dict_section): 65 # key prefix - lat, clat or slat. 66 # dict section - portion of json containing latency bucket in question 67 # Return dict key to access the bucket and unit as string 68 for k, v in dict_section.items(): 69 if k.startswith(key_prefix): 70 return k, k.split("_")[1] 71 72 read_iops = float(data["jobs"][job_pos]["read"]["iops"]) 73 read_bw = float(data["jobs"][job_pos]["read"]["bw"]) 74 lat_key, lat_unit = get_lat_unit("lat", data["jobs"][job_pos]["read"]) 75 read_avg_lat = float(data["jobs"][job_pos]["read"][lat_key]["mean"]) 76 read_min_lat = float(data["jobs"][job_pos]["read"][lat_key]["min"]) 77 read_max_lat = float(data["jobs"][job_pos]["read"][lat_key]["max"]) 78 clat_key, clat_unit = get_lat_unit("clat", data["jobs"][job_pos]["read"]) 79 read_p99_lat = float(data["jobs"][job_pos]["read"][clat_key]["percentile"]["99.000000"]) 80 81 if "ns" in lat_unit: 82 read_avg_lat, read_min_lat, read_max_lat = [x / 1000 for x in [read_avg_lat, read_min_lat, read_max_lat]] 83 if "ns" in clat_unit: 84 read_p99_lat = read_p99_lat / 1000 85 86 write_iops = float(data["jobs"][job_pos]["write"]["iops"]) 87 write_bw = float(data["jobs"][job_pos]["write"]["bw"]) 88 lat_key, lat_unit = get_lat_unit("lat", data["jobs"][job_pos]["write"]) 89 write_avg_lat = float(data["jobs"][job_pos]["write"][lat_key]["mean"]) 90 write_min_lat = float(data["jobs"][job_pos]["write"][lat_key]["min"]) 91 write_max_lat = float(data["jobs"][job_pos]["write"][lat_key]["max"]) 92 clat_key, clat_unit = get_lat_unit("clat", data["jobs"][job_pos]["write"]) 93 write_p99_lat = float(data["jobs"][job_pos]["write"][clat_key]["percentile"]["99.000000"]) 94 95 if "ns" in lat_unit: 96 write_avg_lat, write_min_lat, write_max_lat = [x / 1000 for x in [write_avg_lat, write_min_lat, write_max_lat]] 97 if "ns" in clat_unit: 98 write_p99_lat = write_p99_lat / 1000 99 100 return [read_iops, read_bw, read_avg_lat, read_min_lat, read_max_lat, read_p99_lat, 101 write_iops, write_bw, write_avg_lat, write_min_lat, write_max_lat, write_p99_lat] 102 103 def parse_results(self, results_dir, initiator_count=None, run_num=None): 104 files = os.listdir(results_dir) 105 fio_files = filter(lambda x: ".fio" in x, files) 106 json_files = [x for x in files if ".json" in x] 107 108 # Create empty results file 109 csv_file = "nvmf_results.csv" 110 with open(os.path.join(results_dir, csv_file), "w") as fh: 111 header_line = ",".join(["Name", 112 "read_iops", "read_bw", "read_avg_lat_us", 113 "read_min_lat_us", "read_max_lat_us", "read_p99_lat_us", 114 "write_iops", "write_bw", "write_avg_lat_us", 115 "write_min_lat_us", "write_max_lat_us", "write_p99_lat_us"]) 116 fh.write(header_line + "\n") 117 rows = set() 118 119 for fio_config in fio_files: 120 self.log_print("Getting FIO stats for %s" % fio_config) 121 job_name, _ = os.path.splitext(fio_config) 122 123 # If "_CPU" exists in name - ignore it 124 # Initiators for the same job could have diffrent num_cores parameter 125 job_name = re.sub(r"_\d+CPU", "", job_name) 126 job_result_files = [x for x in json_files if job_name in x] 127 self.log_print("Matching result files for current fio config:") 128 for j in job_result_files: 129 self.log_print("\t %s" % j) 130 131 # There may have been more than 1 initiator used in test, need to check that 132 # Result files are created so that string after last "_" separator is server name 133 inits_names = set([os.path.splitext(x)[0].split("_")[-1] for x in job_result_files]) 134 inits_avg_results = [] 135 for i in inits_names: 136 self.log_print("\tGetting stats for initiator %s" % i) 137 # There may have been more than 1 test run for this job, calculate average results for initiator 138 i_results = [x for x in job_result_files if i in x] 139 140 separate_stats = [] 141 for r in i_results: 142 stats = self.read_json_stats(os.path.join(results_dir, r)) 143 separate_stats.append(stats) 144 self.log_print(stats) 145 146 z = [sum(c) for c in zip(*separate_stats)] 147 z = [c/len(separate_stats) for c in z] 148 inits_avg_results.append(z) 149 150 self.log_print("\tAverage results for initiator %s" % i) 151 self.log_print(z) 152 153 # Sum average results of all initiators running this FIO job 154 self.log_print("\tTotal results for %s from all initiators" % fio_config) 155 for a in inits_avg_results: 156 self.log_print(a) 157 total = ["{0:.3f}".format(sum(c)) for c in zip(*inits_avg_results)] 158 rows.add(",".join([job_name, *total])) 159 160 # Save results to file 161 for row in rows: 162 with open(os.path.join(results_dir, csv_file), "a") as fh: 163 fh.write(row + "\n") 164 self.log_print("You can find the test results in the file %s" % os.path.join(results_dir, csv_file)) 165 166 def measure_sar(self, results_dir, sar_file_name): 167 self.log_print("Waiting %d delay before measuring SAR stats" % self.sar_delay) 168 time.sleep(self.sar_delay) 169 out = subprocess.check_output("sar -P ALL %s %s" % (self.sar_interval, self.sar_count), shell=True).decode(encoding="utf-8") 170 with open(os.path.join(results_dir, sar_file_name), "w") as fh: 171 for line in out.split("\n"): 172 if "Average" in line and "CPU" in line: 173 self.log_print("Summary CPU utilization from SAR:") 174 self.log_print(line) 175 if "Average" in line and "all" in line: 176 self.log_print(line) 177 fh.write(out) 178 179 180class Initiator(Server): 181 def __init__(self, name, username, password, mode, nic_ips, ip, transport="rdma", 182 nvmecli_bin="nvme", workspace="/tmp/spdk", fio_bin="/usr/src/fio/fio"): 183 184 super(Initiator, self).__init__(name, username, password, mode, nic_ips, transport) 185 186 self.ip = ip 187 self.spdk_dir = workspace 188 self.fio_bin = fio_bin 189 self.nvmecli_bin = nvmecli_bin 190 self.ssh_connection = paramiko.SSHClient() 191 self.ssh_connection.set_missing_host_key_policy(paramiko.AutoAddPolicy()) 192 self.ssh_connection.connect(self.ip, username=self.username, password=self.password) 193 self.remote_call("sudo rm -rf %s/nvmf_perf" % self.spdk_dir) 194 self.remote_call("mkdir -p %s" % self.spdk_dir) 195 196 def __del__(self): 197 self.ssh_connection.close() 198 199 def put_file(self, local, remote_dest): 200 ftp = self.ssh_connection.open_sftp() 201 ftp.put(local, remote_dest) 202 ftp.close() 203 204 def get_file(self, remote, local_dest): 205 ftp = self.ssh_connection.open_sftp() 206 ftp.get(remote, local_dest) 207 ftp.close() 208 209 def remote_call(self, cmd): 210 stdin, stdout, stderr = self.ssh_connection.exec_command(cmd) 211 out = stdout.read().decode(encoding="utf-8") 212 err = stderr.read().decode(encoding="utf-8") 213 return out, err 214 215 def copy_result_files(self, dest_dir): 216 self.log_print("Copying results") 217 218 if not os.path.exists(dest_dir): 219 os.mkdir(dest_dir) 220 221 # Get list of result files from initiator and copy them back to target 222 stdout, stderr = self.remote_call("ls %s/nvmf_perf" % self.spdk_dir) 223 file_list = stdout.strip().split("\n") 224 225 for file in file_list: 226 self.get_file(os.path.join(self.spdk_dir, "nvmf_perf", file), 227 os.path.join(dest_dir, file)) 228 self.log_print("Done copying results") 229 230 def discover_subsystems(self, address_list, subsys_no): 231 num_nvmes = range(0, subsys_no) 232 nvme_discover_output = "" 233 for ip, subsys_no in itertools.product(address_list, num_nvmes): 234 self.log_print("Trying to discover: %s:%s" % (ip, 4420 + subsys_no)) 235 nvme_discover_cmd = ["sudo", 236 "%s" % self.nvmecli_bin, 237 "discover", "-t %s" % self.transport, 238 "-s %s" % (4420 + subsys_no), 239 "-a %s" % ip] 240 nvme_discover_cmd = " ".join(nvme_discover_cmd) 241 242 stdout, stderr = self.remote_call(nvme_discover_cmd) 243 if stdout: 244 nvme_discover_output = nvme_discover_output + stdout 245 246 subsystems = re.findall(r'trsvcid:\s(\d+)\s+' # get svcid number 247 r'subnqn:\s+([a-zA-Z0-9\.\-\:]+)\s+' # get NQN id 248 r'traddr:\s+(\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3})', # get IP address 249 nvme_discover_output) # from nvme discovery output 250 subsystems = filter(lambda x: x[-1] in address_list, subsystems) 251 subsystems = list(set(subsystems)) 252 subsystems.sort(key=lambda x: x[1]) 253 self.log_print("Found matching subsystems on target side:") 254 for s in subsystems: 255 self.log_print(s) 256 257 return subsystems 258 259 def gen_fio_config(self, rw, rwmixread, block_size, io_depth, subsys_no, num_jobs=None, ramp_time=0, run_time=10): 260 fio_conf_template = """ 261[global] 262ioengine={ioengine} 263{spdk_conf} 264thread=1 265group_reporting=1 266direct=1 267 268norandommap=1 269rw={rw} 270rwmixread={rwmixread} 271bs={block_size} 272iodepth={io_depth} 273time_based=1 274ramp_time={ramp_time} 275runtime={run_time} 276""" 277 if "spdk" in self.mode: 278 subsystems = self.discover_subsystems(self.nic_ips, subsys_no) 279 bdev_conf = self.gen_spdk_bdev_conf(subsystems) 280 self.remote_call("echo '%s' > %s/bdev.conf" % (bdev_conf, self.spdk_dir)) 281 ioengine = "%s/examples/bdev/fio_plugin/fio_plugin" % self.spdk_dir 282 spdk_conf = "spdk_conf=%s/bdev.conf" % self.spdk_dir 283 filename_section = self.gen_fio_filename_conf(subsystems) 284 else: 285 ioengine = "libaio" 286 spdk_conf = "" 287 filename_section = self.gen_fio_filename_conf() 288 289 fio_config = fio_conf_template.format(ioengine=ioengine, spdk_conf=spdk_conf, 290 rw=rw, rwmixread=rwmixread, block_size=block_size, 291 io_depth=io_depth, ramp_time=ramp_time, run_time=run_time) 292 if num_jobs: 293 fio_config = fio_config + "numjobs=%s" % num_jobs 294 fio_config = fio_config + filename_section 295 296 fio_config_filename = "%s_%s_%s_m_%s" % (block_size, io_depth, rw, rwmixread) 297 if hasattr(self, "num_cores"): 298 fio_config_filename += "_%sCPU" % self.num_cores 299 fio_config_filename += ".fio" 300 301 self.remote_call("mkdir -p %s/nvmf_perf" % self.spdk_dir) 302 self.remote_call("echo '%s' > %s/nvmf_perf/%s" % (fio_config, self.spdk_dir, fio_config_filename)) 303 self.log_print("Created FIO Config:") 304 self.log_print(fio_config) 305 306 return os.path.join(self.spdk_dir, "nvmf_perf", fio_config_filename) 307 308 def run_fio(self, fio_config_file, run_num=None): 309 job_name, _ = os.path.splitext(fio_config_file) 310 self.log_print("Starting FIO run for job: %s" % job_name) 311 self.log_print("Using FIO: %s" % self.fio_bin) 312 if run_num: 313 for i in range(1, run_num + 1): 314 output_filename = job_name + "_run_" + str(i) + "_" + self.name + ".json" 315 cmd = "sudo %s %s --output-format=json --output=%s" % (self.fio_bin, fio_config_file, output_filename) 316 output, error = self.remote_call(cmd) 317 self.log_print(output) 318 self.log_print(error) 319 else: 320 output_filename = job_name + "_" + self.name + ".json" 321 cmd = "sudo %s %s --output-format=json --output=%s" % (self.fio_bin, fio_config_file, output_filename) 322 output, error = self.remote_call(cmd) 323 self.log_print(output) 324 self.log_print(error) 325 self.log_print("FIO run finished. Results in: %s" % output_filename) 326 327 328class KernelTarget(Target): 329 def __init__(self, name, username, password, mode, nic_ips, transport="rdma", 330 use_null_block=False, sar_settings=None, 331 nvmet_bin="nvmetcli", **kwargs): 332 333 super(KernelTarget, self).__init__(name, username, password, mode, nic_ips, transport, 334 use_null_block, sar_settings) 335 self.nvmet_bin = nvmet_bin 336 337 def __del__(self): 338 nvmet_command(self.nvmet_bin, "clear") 339 340 def kernel_tgt_gen_nullblock_conf(self, address): 341 nvmet_cfg = { 342 "ports": [], 343 "hosts": [], 344 "subsystems": [], 345 } 346 347 nvmet_cfg["subsystems"].append({ 348 "allowed_hosts": [], 349 "attr": { 350 "allow_any_host": "1", 351 "version": "1.3" 352 }, 353 "namespaces": [ 354 { 355 "device": { 356 "path": "/dev/nullb0", 357 "uuid": "%s" % uuid.uuid4() 358 }, 359 "enable": 1, 360 "nsid": 1 361 } 362 ], 363 "nqn": "nqn.2018-09.io.spdk:cnode1" 364 }) 365 366 nvmet_cfg["ports"].append({ 367 "addr": { 368 "adrfam": "ipv4", 369 "traddr": address, 370 "trsvcid": "4420", 371 "trtype": "%s" % self.transport, 372 }, 373 "portid": 1, 374 "referrals": [], 375 "subsystems": ["nqn.2018-09.io.spdk:cnode1"] 376 }) 377 with open("kernel.conf", 'w') as fh: 378 fh.write(json.dumps(nvmet_cfg, indent=2)) 379 380 def kernel_tgt_gen_subsystem_conf(self, nvme_list, address_list): 381 382 nvmet_cfg = { 383 "ports": [], 384 "hosts": [], 385 "subsystems": [], 386 } 387 388 # Split disks between NIC IP's 389 disks_per_ip = int(len(nvme_list) / len(address_list)) 390 disk_chunks = [nvme_list[i * disks_per_ip:disks_per_ip + disks_per_ip * i] for i in range(0, len(address_list))] 391 392 subsys_no = 1 393 port_no = 0 394 for ip, chunk in zip(address_list, disk_chunks): 395 for disk in chunk: 396 nvmet_cfg["subsystems"].append({ 397 "allowed_hosts": [], 398 "attr": { 399 "allow_any_host": "1", 400 "version": "1.3" 401 }, 402 "namespaces": [ 403 { 404 "device": { 405 "path": disk, 406 "uuid": "%s" % uuid.uuid4() 407 }, 408 "enable": 1, 409 "nsid": subsys_no 410 } 411 ], 412 "nqn": "nqn.2018-09.io.spdk:cnode%s" % subsys_no 413 }) 414 415 nvmet_cfg["ports"].append({ 416 "addr": { 417 "adrfam": "ipv4", 418 "traddr": ip, 419 "trsvcid": "%s" % (4420 + port_no), 420 "trtype": "%s" % self.transport 421 }, 422 "portid": subsys_no, 423 "referrals": [], 424 "subsystems": ["nqn.2018-09.io.spdk:cnode%s" % subsys_no] 425 }) 426 subsys_no += 1 427 port_no += 1 428 429 with open("kernel.conf", "w") as fh: 430 fh.write(json.dumps(nvmet_cfg, indent=2)) 431 pass 432 433 def tgt_start(self): 434 self.log_print("Configuring kernel NVMeOF Target") 435 436 if self.null_block: 437 print("Configuring with null block device.") 438 if len(self.nic_ips) > 1: 439 print("Testing with null block limited to single RDMA NIC.") 440 print("Please specify only 1 IP address.") 441 exit(1) 442 self.subsys_no = 1 443 self.kernel_tgt_gen_nullblock_conf(self.nic_ips[0]) 444 else: 445 print("Configuring with NVMe drives.") 446 nvme_list = get_nvme_devices() 447 self.kernel_tgt_gen_subsystem_conf(nvme_list, self.nic_ips) 448 self.subsys_no = len(nvme_list) 449 450 nvmet_command(self.nvmet_bin, "clear") 451 nvmet_command(self.nvmet_bin, "restore kernel.conf") 452 self.log_print("Done configuring kernel NVMeOF Target") 453 454 455class SPDKTarget(Target): 456 457 def __init__(self, name, username, password, mode, nic_ips, transport="rdma", 458 use_null_block=False, sar_settings=None, 459 num_shared_buffers=4096, num_cores=1, **kwargs): 460 461 super(SPDKTarget, self).__init__(name, username, password, mode, nic_ips, transport, 462 use_null_block, sar_settings) 463 self.num_cores = num_cores 464 self.num_shared_buffers = num_shared_buffers 465 466 def spdk_tgt_configure(self): 467 self.log_print("Configuring SPDK NVMeOF target via RPC") 468 numa_list = get_used_numa_nodes() 469 470 # Create RDMA transport layer 471 rpc.nvmf.nvmf_create_transport(self.client, trtype=self.transport, num_shared_buffers=self.num_shared_buffers) 472 self.log_print("SPDK NVMeOF transport layer:") 473 rpc.client.print_dict(rpc.nvmf.nvmf_get_transports(self.client)) 474 475 if self.null_block: 476 nvme_section = self.spdk_tgt_add_nullblock() 477 subsystems_section = self.spdk_tgt_add_subsystem_conf(self.nic_ips, req_num_disks=1) 478 else: 479 nvme_section = self.spdk_tgt_add_nvme_conf() 480 subsystems_section = self.spdk_tgt_add_subsystem_conf(self.nic_ips) 481 self.log_print("Done configuring SPDK NVMeOF Target") 482 483 def spdk_tgt_add_nullblock(self): 484 self.log_print("Adding null block bdev to config via RPC") 485 rpc.bdev.bdev_null_create(self.client, 102400, 4096, "Nvme0n1") 486 self.log_print("SPDK Bdevs configuration:") 487 rpc.client.print_dict(rpc.bdev.bdev_get_bdevs(self.client)) 488 489 def spdk_tgt_add_nvme_conf(self, req_num_disks=None): 490 self.log_print("Adding NVMe bdevs to config via RPC") 491 492 bdfs = get_nvme_devices_bdf() 493 bdfs = [b.replace(":", ".") for b in bdfs] 494 495 if req_num_disks: 496 if req_num_disks > len(bdfs): 497 self.log_print("ERROR: Requested number of disks is more than available %s" % len(bdfs)) 498 sys.exit(1) 499 else: 500 bdfs = bdfs[0:req_num_disks] 501 502 for i, bdf in enumerate(bdfs): 503 rpc.bdev.bdev_nvme_attach_controller(self.client, name="Nvme%s" % i, trtype="PCIe", traddr=bdf) 504 505 self.log_print("SPDK Bdevs configuration:") 506 rpc.client.print_dict(rpc.bdev.bdev_get_bdevs(self.client)) 507 508 def spdk_tgt_add_subsystem_conf(self, ips=None, req_num_disks=None): 509 self.log_print("Adding subsystems to config") 510 if not req_num_disks: 511 req_num_disks = get_nvme_devices_count() 512 513 # Distribute bdevs between provided NICs 514 num_disks = range(1, req_num_disks + 1) 515 disks_per_ip = int(len(num_disks) / len(ips)) 516 disk_chunks = [num_disks[i * disks_per_ip:disks_per_ip + disks_per_ip * i] for i in range(0, len(ips))] 517 518 # Create subsystems, add bdevs to namespaces, add listeners 519 for ip, chunk in zip(ips, disk_chunks): 520 for c in chunk: 521 nqn = "nqn.2018-09.io.spdk:cnode%s" % c 522 serial = "SPDK00%s" % c 523 bdev_name = "Nvme%sn1" % (c - 1) 524 rpc.nvmf.nvmf_create_subsystem(self.client, nqn, serial, 525 allow_any_host=True, max_namespaces=8) 526 rpc.nvmf.nvmf_subsystem_add_ns(self.client, nqn, bdev_name) 527 528 rpc.nvmf.nvmf_subsystem_add_listener(self.client, nqn, 529 trtype=self.transport, 530 traddr=ip, 531 trsvcid="4420", 532 adrfam="ipv4") 533 534 self.log_print("SPDK NVMeOF subsystem configuration:") 535 rpc.client.print_dict(rpc.nvmf.nvmf_get_subsystems(self.client)) 536 537 def tgt_start(self): 538 self.subsys_no = get_nvme_devices_count() 539 self.log_print("Starting SPDK NVMeOF Target process") 540 nvmf_app_path = os.path.join(self.spdk_dir, "app/nvmf_tgt/nvmf_tgt") 541 command = " ".join([nvmf_app_path, "-m", self.num_cores]) 542 proc = subprocess.Popen(command, shell=True) 543 self.pid = os.path.join(self.spdk_dir, "nvmf.pid") 544 545 with open(self.pid, "w") as fh: 546 fh.write(str(proc.pid)) 547 self.nvmf_proc = proc 548 self.log_print("SPDK NVMeOF Target PID=%s" % self.pid) 549 self.log_print("Waiting for spdk to initilize...") 550 while True: 551 if os.path.exists("/var/tmp/spdk.sock"): 552 break 553 time.sleep(1) 554 self.client = rpc.client.JSONRPCClient("/var/tmp/spdk.sock") 555 556 self.spdk_tgt_configure() 557 558 def __del__(self): 559 if hasattr(self, "nvmf_proc"): 560 try: 561 self.nvmf_proc.terminate() 562 self.nvmf_proc.wait() 563 except Exception as e: 564 self.log_print(e) 565 self.nvmf_proc.kill() 566 self.nvmf_proc.communicate() 567 568 569class KernelInitiator(Initiator): 570 def __init__(self, name, username, password, mode, nic_ips, ip, transport, 571 fio_bin="/usr/src/fio/fio", **kwargs): 572 573 super(KernelInitiator, self).__init__(name, username, password, mode, nic_ips, ip, transport, 574 fio_bin=fio_bin) 575 576 self.extra_params = "" 577 if kwargs["extra_params"]: 578 self.extra_params = kwargs["extra_params"] 579 580 def __del__(self): 581 self.ssh_connection.close() 582 583 def kernel_init_connect(self, address_list, subsys_no): 584 subsystems = self.discover_subsystems(address_list, subsys_no) 585 self.log_print("Below connection attempts may result in error messages, this is expected!") 586 for subsystem in subsystems: 587 self.log_print("Trying to connect %s %s %s" % subsystem) 588 self.remote_call("sudo %s connect -t %s -s %s -n %s -a %s %s" % (self.nvmecli_bin, 589 self.transport, 590 *subsystem, 591 self.extra_params)) 592 time.sleep(2) 593 594 def kernel_init_disconnect(self, address_list, subsys_no): 595 subsystems = self.discover_subsystems(address_list, subsys_no) 596 for subsystem in subsystems: 597 self.remote_call("sudo %s disconnect -n %s" % (self.nvmecli_bin, subsystem[1])) 598 time.sleep(1) 599 600 def gen_fio_filename_conf(self): 601 out, err = self.remote_call("lsblk -o NAME -nlp") 602 nvme_list = [x for x in out.split("\n") if "nvme" in x] 603 604 filename_section = "" 605 for i, nvme in enumerate(nvme_list): 606 filename_section = "\n".join([filename_section, 607 "[filename%s]" % i, 608 "filename=%s" % nvme]) 609 610 return filename_section 611 612 613class SPDKInitiator(Initiator): 614 def __init__(self, name, username, password, mode, nic_ips, ip, transport="rdma", 615 num_cores=1, fio_bin="/usr/src/fio/fio", **kwargs): 616 super(SPDKInitiator, self).__init__(name, username, password, mode, nic_ips, ip, transport, 617 fio_bin=fio_bin) 618 619 self.num_cores = num_cores 620 621 def install_spdk(self, local_spdk_zip): 622 self.put_file(local_spdk_zip, "/tmp/spdk_drop.zip") 623 self.log_print("Copied sources zip from target") 624 self.remote_call("unzip -qo /tmp/spdk_drop.zip -d %s" % self.spdk_dir) 625 626 self.log_print("Sources unpacked") 627 self.log_print("Using fio binary %s" % self.fio_bin) 628 self.remote_call("cd %s; git submodule update --init; ./configure --with-rdma --with-fio=%s;" 629 "make clean; make -j$(($(nproc)*2))" % (self.spdk_dir, os.path.dirname(self.fio_bin))) 630 631 self.log_print("SPDK built") 632 self.remote_call("sudo %s/scripts/setup.sh" % self.spdk_dir) 633 634 def gen_spdk_bdev_conf(self, remote_subsystem_list): 635 header = "[Nvme]" 636 row_template = """ TransportId "trtype:{transport} adrfam:IPv4 traddr:{ip} trsvcid:{svc} subnqn:{nqn}" Nvme{i}""" 637 638 bdev_rows = [row_template.format(transport=self.transport, 639 svc=x[0], 640 nqn=x[1], 641 ip=x[2], 642 i=i) for i, x in enumerate(remote_subsystem_list)] 643 bdev_rows = "\n".join(bdev_rows) 644 bdev_section = "\n".join([header, bdev_rows]) 645 return bdev_section 646 647 def gen_fio_filename_conf(self, remote_subsystem_list): 648 subsystems = [str(x) for x in range(0, len(remote_subsystem_list))] 649 650 # If num_cpus exists then limit FIO to this number of CPUs 651 # Otherwise - each connected subsystem gets its own CPU 652 if hasattr(self, 'num_cores'): 653 self.log_print("Limiting FIO workload execution to %s cores" % self.num_cores) 654 threads = range(0, int(self.num_cores)) 655 else: 656 threads = range(0, len(subsystems)) 657 658 n = int(len(subsystems) / len(threads)) 659 660 filename_section = "" 661 for t in threads: 662 header = "[filename%s]" % t 663 disks = "\n".join(["filename=Nvme%sn1" % x for x in subsystems[n * t:n + n * t]]) 664 filename_section = "\n".join([filename_section, header, disks]) 665 666 return filename_section 667 668 669if __name__ == "__main__": 670 spdk_zip_path = "/tmp/spdk.zip" 671 target_results_dir = "/tmp/results" 672 673 if (len(sys.argv) > 1): 674 config_file_path = sys.argv[1] 675 else: 676 script_full_dir = os.path.dirname(os.path.realpath(__file__)) 677 config_file_path = os.path.join(script_full_dir, "config.json") 678 679 print("Using config file: %s" % config_file_path) 680 with open(config_file_path, "r") as config: 681 data = json.load(config) 682 683 initiators = [] 684 fio_cases = [] 685 686 for k, v in data.items(): 687 if "target" in k: 688 if data[k]["mode"] == "spdk": 689 target_obj = SPDKTarget(name=k, **data["general"], **v) 690 elif data[k]["mode"] == "kernel": 691 target_obj = KernelTarget(name=k, **data["general"], **v) 692 elif "initiator" in k: 693 if data[k]["mode"] == "spdk": 694 init_obj = SPDKInitiator(name=k, **data["general"], **v) 695 elif data[k]["mode"] == "kernel": 696 init_obj = KernelInitiator(name=k, **data["general"], **v) 697 initiators.append(init_obj) 698 elif "fio" in k: 699 fio_workloads = itertools.product(data[k]["bs"], 700 data[k]["qd"], 701 data[k]["rw"]) 702 703 fio_run_time = data[k]["run_time"] 704 fio_ramp_time = data[k]["ramp_time"] 705 fio_rw_mix_read = data[k]["rwmixread"] 706 fio_run_num = data[k]["run_num"] if "run_num" in data[k].keys() else None 707 fio_num_jobs = data[k]["num_jobs"] if "num_jobs" in data[k].keys() else None 708 else: 709 continue 710 711 # Copy and install SPDK on remote initiators 712 target_obj.zip_spdk_sources(target_obj.spdk_dir, spdk_zip_path) 713 threads = [] 714 for i in initiators: 715 if i.mode == "spdk": 716 t = threading.Thread(target=i.install_spdk, args=(spdk_zip_path,)) 717 threads.append(t) 718 t.start() 719 for t in threads: 720 t.join() 721 722 target_obj.tgt_start() 723 724 # Poor mans threading 725 # Run FIO tests 726 for block_size, io_depth, rw in fio_workloads: 727 threads = [] 728 configs = [] 729 for i in initiators: 730 if i.mode == "kernel": 731 i.kernel_init_connect(i.nic_ips, target_obj.subsys_no) 732 733 cfg = i.gen_fio_config(rw, fio_rw_mix_read, block_size, io_depth, target_obj.subsys_no, 734 fio_num_jobs, fio_ramp_time, fio_run_time) 735 configs.append(cfg) 736 737 for i, cfg in zip(initiators, configs): 738 t = threading.Thread(target=i.run_fio, args=(cfg, fio_run_num)) 739 threads.append(t) 740 if target_obj.enable_sar: 741 sar_file_name = "_".join([str(block_size), str(rw), str(io_depth), "sar"]) 742 sar_file_name = ".".join([sar_file_name, "txt"]) 743 t = threading.Thread(target=target_obj.measure_sar, args=(target_results_dir, sar_file_name)) 744 threads.append(t) 745 746 for t in threads: 747 t.start() 748 for t in threads: 749 t.join() 750 751 for i in initiators: 752 if i.mode == "kernel": 753 i.kernel_init_disconnect(i.nic_ips, target_obj.subsys_no) 754 i.copy_result_files(target_results_dir) 755 756 target_obj.parse_results(target_results_dir) 757