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