root / branches / 1.1 / src / haizea / cli / commands.py @ 707
| 1 |
# -------------------------------------------------------------------------- #
|
|---|---|
| 2 |
# Copyright 2006-2009, University of Chicago #
|
| 3 |
# Copyright 2008-2009, Distributed Systems Architecture Group, Universidad #
|
| 4 |
# Complutense de Madrid (dsa-research.org) #
|
| 5 |
# #
|
| 6 |
# Licensed under the Apache License, Version 2.0 (the "License"); you may #
|
| 7 |
# not use this file except in compliance with the License. You may obtain #
|
| 8 |
# a copy of the License at #
|
| 9 |
# #
|
| 10 |
# http://www.apache.org/licenses/LICENSE-2.0 #
|
| 11 |
# #
|
| 12 |
# Unless required by applicable law or agreed to in writing, software #
|
| 13 |
# distributed under the License is distributed on an "AS IS" BASIS, #
|
| 14 |
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. #
|
| 15 |
# See the License for the specific language governing permissions and #
|
| 16 |
# limitations under the License. #
|
| 17 |
# -------------------------------------------------------------------------- #
|
| 18 |
|
| 19 |
from haizea.core.manager import Manager |
| 20 |
from haizea.common.utils import generate_config_name, unpickle |
| 21 |
from haizea.core.configfile import HaizeaConfig, HaizeaMultiConfig |
| 22 |
from haizea.core.accounting import AccountingDataCollection |
| 23 |
from haizea.common.config import ConfigException |
| 24 |
from haizea.common.stats import percentile |
| 25 |
from haizea.cli.optionparser import Option |
| 26 |
from haizea.cli import Command |
| 27 |
from mx.DateTime import TimeDelta, Parser |
| 28 |
import haizea.common.defaults as defaults |
| 29 |
import sys |
| 30 |
import os |
| 31 |
import errno |
| 32 |
import signal |
| 33 |
from time import sleep |
| 34 |
|
| 35 |
try:
|
| 36 |
import xml.etree.ElementTree as ET |
| 37 |
except ImportError: |
| 38 |
# Compatibility with Python <=2.4
|
| 39 |
import elementtree.ElementTree as ET |
| 40 |
|
| 41 |
|
| 42 |
class haizea(Command): |
| 43 |
"""
|
| 44 |
This is the main Haizea command. By default, it will start Haizea as a daemon, which
|
| 45 |
can receive requests via RPC or interact with other components such as OpenNebula. It can
|
| 46 |
also start as a foreground process, and write all log messages to the console. All
|
| 47 |
Haizea options are specified through the configuration file."""
|
| 48 |
|
| 49 |
name = "haizea"
|
| 50 |
|
| 51 |
def __init__(self, argv): |
| 52 |
Command.__init__(self, argv)
|
| 53 |
|
| 54 |
self.optparser.add_option(Option("-c", "--conf", action="store", type="string", dest="conf", |
| 55 |
help = """
|
| 56 |
The location of the Haizea configuration file. If not
|
| 57 |
specified, Haizea will first look for it in
|
| 58 |
/etc/haizea/haizea.conf and then in ~/.haizea/haizea.conf.
|
| 59 |
"""))
|
| 60 |
self.optparser.add_option(Option("-f", "--fg", action="store_true", dest="foreground", |
| 61 |
help = """
|
| 62 |
Runs Haizea in the foreground.
|
| 63 |
"""))
|
| 64 |
self.optparser.add_option(Option("--stop", action="store_true", dest="stop", |
| 65 |
help = """
|
| 66 |
Stops the Haizea daemon.
|
| 67 |
"""))
|
| 68 |
|
| 69 |
def run(self): |
| 70 |
self.parse_options()
|
| 71 |
|
| 72 |
pidfile = defaults.DAEMON_PIDFILE # TODO: Make configurable
|
| 73 |
|
| 74 |
if self.opt.stop == None: |
| 75 |
# Start Haizea
|
| 76 |
|
| 77 |
# Check if a daemon is already running
|
| 78 |
if os.path.exists(pidfile):
|
| 79 |
pf = file(pidfile,'r') |
| 80 |
pid = int(pf.read().strip())
|
| 81 |
pf.close() |
| 82 |
|
| 83 |
try:
|
| 84 |
os.kill(pid, signal.SIG_DFL) |
| 85 |
except OSError, (err, msg): |
| 86 |
if err == errno.ESRCH:
|
| 87 |
# Pidfile is stale. Remove it.
|
| 88 |
os.remove(pidfile) |
| 89 |
else:
|
| 90 |
msg = "Unexpected error when checking pid file '%s'.\n%s\n" %(pidfile, msg)
|
| 91 |
sys.stderr.write(msg) |
| 92 |
sys.exit(1)
|
| 93 |
else:
|
| 94 |
msg = "Haizea seems to be already running (pid %i)\n" % pid
|
| 95 |
sys.stderr.write(msg) |
| 96 |
sys.exit(1)
|
| 97 |
|
| 98 |
try:
|
| 99 |
configfile=self.opt.conf
|
| 100 |
if configfile == None: |
| 101 |
# Look for config file in default locations
|
| 102 |
for loc in defaults.CONFIG_LOCATIONS: |
| 103 |
if os.path.exists(loc):
|
| 104 |
config = HaizeaConfig.from_file(loc) |
| 105 |
break
|
| 106 |
else:
|
| 107 |
print >> sys.stdout, "No configuration file specified, and none found at default locations." |
| 108 |
print >> sys.stdout, "Make sure a config file exists at:\n -> %s" % "\n -> ".join(defaults.CONFIG_LOCATIONS) |
| 109 |
print >> sys.stdout, "Or specify a configuration file with the --conf option." |
| 110 |
exit(1) |
| 111 |
else:
|
| 112 |
config = HaizeaConfig.from_file(configfile) |
| 113 |
except ConfigException, msg:
|
| 114 |
print >> sys.stderr, "Error in configuration file:" |
| 115 |
print >> sys.stderr, msg
|
| 116 |
exit(1) |
| 117 |
|
| 118 |
daemon = not self.opt.foreground |
| 119 |
|
| 120 |
manager = Manager(config, daemon, pidfile) |
| 121 |
|
| 122 |
manager.start() |
| 123 |
elif self.opt.stop: # Stop Haizea |
| 124 |
# Based on code in: http://aspn.activestate.com/ASPN/Cookbook/Python/Recipe/66012
|
| 125 |
try:
|
| 126 |
pf = file(pidfile,'r') |
| 127 |
pid = int(pf.read().strip())
|
| 128 |
pf.close() |
| 129 |
except IOError: |
| 130 |
msg = "Could not stop, pid file '%s' missing.\n"
|
| 131 |
sys.stderr.write(msg % pidfile) |
| 132 |
sys.exit(1)
|
| 133 |
try:
|
| 134 |
while 1: |
| 135 |
os.kill(pid, signal.SIGTERM) |
| 136 |
sleep(1)
|
| 137 |
except OSError, err: |
| 138 |
err = str(err)
|
| 139 |
if err.find("No such process") > 0: |
| 140 |
os.remove(pidfile) |
| 141 |
else:
|
| 142 |
print str(err) |
| 143 |
sys.exit(1)
|
| 144 |
|
| 145 |
class haizea_generate_configs(Command): |
| 146 |
"""
|
| 147 |
Takes an Haizea multiconfiguration file and generates the individual
|
| 148 |
configuration files. See the Haizea manual for more details on multiconfiguration
|
| 149 |
files."""
|
| 150 |
|
| 151 |
name = "haizea-generate-configs"
|
| 152 |
|
| 153 |
def __init__(self, argv): |
| 154 |
Command.__init__(self, argv)
|
| 155 |
|
| 156 |
self.optparser.add_option(Option("-c", "--conf", action="store", type="string", dest="conf", required=True, |
| 157 |
help = """
|
| 158 |
Multiconfiguration file.
|
| 159 |
"""))
|
| 160 |
self.optparser.add_option(Option("-d", "--dir", action="store", type="string", dest="dir", required=True, |
| 161 |
help = """
|
| 162 |
Directory where the individual configuration files
|
| 163 |
must be created.
|
| 164 |
"""))
|
| 165 |
|
| 166 |
def run(self): |
| 167 |
self.parse_options()
|
| 168 |
|
| 169 |
configfile=self.opt.conf
|
| 170 |
multiconfig = HaizeaMultiConfig.from_file(configfile) |
| 171 |
|
| 172 |
etcdir = self.opt.dir
|
| 173 |
|
| 174 |
configs = multiconfig.get_configs() |
| 175 |
|
| 176 |
etcdir = os.path.abspath(etcdir) |
| 177 |
if not os.path.exists(etcdir): |
| 178 |
os.makedirs(etcdir) |
| 179 |
|
| 180 |
for c in configs: |
| 181 |
profile = c.get_attr("profile")
|
| 182 |
tracefile = c.get("tracefile")
|
| 183 |
injfile = c.get("injectionfile")
|
| 184 |
annotationfile = c.get("annotationfile")
|
| 185 |
configname = generate_config_name(profile, tracefile, annotationfile, injfile) |
| 186 |
configfile = etcdir + "/%s.conf" % configname
|
| 187 |
fc = open(configfile, "w") |
| 188 |
c.config.write(fc) |
| 189 |
fc.close() |
| 190 |
|
| 191 |
class haizea_generate_scripts(Command): |
| 192 |
"""
|
| 193 |
Generates a script, based on a script template, to run all the individual
|
| 194 |
configuration files generated by haizea-generate-configs. This command
|
| 195 |
requires Mako Templates for Python (http://www.makotemplates.org/)."""
|
| 196 |
|
| 197 |
name = "haizea-generate-scripts"
|
| 198 |
|
| 199 |
def __init__(self, argv): |
| 200 |
Command.__init__(self, argv)
|
| 201 |
|
| 202 |
self.optparser.add_option(Option("-c", "--conf", action="store", type="string", dest="conf", required=True, |
| 203 |
help = """
|
| 204 |
Multiconfiguration file used in haizea-generate-configs.
|
| 205 |
"""))
|
| 206 |
self.optparser.add_option(Option("-d", "--confdir", action="store", type="string", dest="confdir", required=True, |
| 207 |
help = """
|
| 208 |
Directory containing the individual configuration files.
|
| 209 |
"""))
|
| 210 |
self.optparser.add_option(Option("-t", "--template", action="store", type="string", dest="template", required=True, |
| 211 |
help = """
|
| 212 |
Script template (sample templates are included in /usr/share/haizea/etc)
|
| 213 |
"""))
|
| 214 |
self.optparser.add_option(Option("-m", "--only-missing", action="store_true", dest="onlymissing", |
| 215 |
help = """
|
| 216 |
If specified, the generated script will only run the configurations
|
| 217 |
that have not already produced a datafile. This is useful when some simulations
|
| 218 |
fail, and you don't want to have to rerun them all.
|
| 219 |
"""))
|
| 220 |
|
| 221 |
def run(self): |
| 222 |
self.parse_options()
|
| 223 |
|
| 224 |
configfile=self.opt.conf
|
| 225 |
multiconfig = HaizeaMultiConfig.from_file(configfile) |
| 226 |
|
| 227 |
try:
|
| 228 |
from mako.template import Template |
| 229 |
except Exception, e: |
| 230 |
print "You need Mako Templates for Python to run this command." |
| 231 |
print "You can download them at http://www.makotemplates.org/" |
| 232 |
exit(1) |
| 233 |
|
| 234 |
configs = multiconfig.get_configs() |
| 235 |
|
| 236 |
etcdir = os.path.abspath(self.opt.confdir)
|
| 237 |
if not os.path.exists(etcdir): |
| 238 |
os.makedirs(etcdir) |
| 239 |
|
| 240 |
templatedata = [] |
| 241 |
for c in configs: |
| 242 |
profile = c.get_attr("profile")
|
| 243 |
tracefile = c.get("tracefile")
|
| 244 |
injfile = c.get("injectionfile")
|
| 245 |
datafile = c.get("datafile")
|
| 246 |
annotationfile = c.get("annotationfile")
|
| 247 |
configname = generate_config_name(profile, tracefile, annotationfile, injfile) |
| 248 |
if not self.opt.onlymissing or not os.path.exists(datafile): |
| 249 |
configfile = etcdir + "/%s.conf" % configname
|
| 250 |
templatedata.append((configname, configfile)) |
| 251 |
|
| 252 |
template = Template(filename=self.opt.template)
|
| 253 |
print template.render(configs=templatedata, etcdir=etcdir)
|
| 254 |
|
| 255 |
|
| 256 |
class haizea_convert_data(Command): |
| 257 |
"""
|
| 258 |
Converts Haizea datafiles into another (easier to process) format.
|
| 259 |
"""
|
| 260 |
|
| 261 |
name = "haizea-convert-data"
|
| 262 |
|
| 263 |
def __init__(self, argv): |
| 264 |
Command.__init__(self, argv)
|
| 265 |
|
| 266 |
self.optparser.add_option(Option("-t", "--type", action="store", dest="type", |
| 267 |
choices = ["per-run", "per-lease", "counter"], |
| 268 |
help = """
|
| 269 |
Type of data to produce.
|
| 270 |
"""))
|
| 271 |
self.optparser.add_option(Option("-c", "--counter", action="store", dest="counter", |
| 272 |
help = """
|
| 273 |
Counter to print out when using '--type counter'.
|
| 274 |
"""))
|
| 275 |
self.optparser.add_option(Option("-f", "--format", action="store", type="string", dest="format", |
| 276 |
help = """
|
| 277 |
Output format. Currently supported: csv
|
| 278 |
"""))
|
| 279 |
self.optparser.add_option(Option("-l", "--list-counters", action="store_true", dest="list_counters", |
| 280 |
help = """
|
| 281 |
If specified, the command will just print out the names of counters
|
| 282 |
stored in the data file and then exit, regardless of other parameters.
|
| 283 |
"""))
|
| 284 |
|
| 285 |
def run(self): |
| 286 |
self.parse_options()
|
| 287 |
|
| 288 |
datafiles=self.args[1:] |
| 289 |
if len(datafiles) == 0: |
| 290 |
print "Please specify at least one datafile to convert" |
| 291 |
exit(1) |
| 292 |
|
| 293 |
datafile1 = unpickle(datafiles[0])
|
| 294 |
|
| 295 |
counter_names = datafile1.counters.keys() |
| 296 |
attr_names = datafile1.attrs.keys() |
| 297 |
lease_stats_names = datafile1.lease_stats_names |
| 298 |
stats_names = datafile1.stats_names |
| 299 |
|
| 300 |
if self.opt.list_counters: |
| 301 |
for counter in counter_names: |
| 302 |
print counter
|
| 303 |
exit(0) |
| 304 |
|
| 305 |
if self.opt.type == "per-run": |
| 306 |
header_fields = attr_names + stats_names |
| 307 |
elif self.opt.type == "per-lease": |
| 308 |
header_fields = attr_names + ["lease_id"] + lease_stats_names
|
| 309 |
elif self.opt.type == "counter": |
| 310 |
counter = self.opt.counter
|
| 311 |
if not datafile1.counters.has_key(counter): |
| 312 |
print "The specified datafile does not have a counter called '%s'" % counter |
| 313 |
exit(1) |
| 314 |
header_fields = attr_names + ["time", "value"] |
| 315 |
if datafile1.counter_avg_type[counter] != AccountingDataCollection.AVERAGE_NONE:
|
| 316 |
header_fields.append("average")
|
| 317 |
|
| 318 |
header = ",".join(header_fields)
|
| 319 |
|
| 320 |
print header
|
| 321 |
|
| 322 |
for datafile in datafiles: |
| 323 |
data = unpickle(datafile) |
| 324 |
|
| 325 |
attrs = [data.attrs[attr_name] for attr_name in attr_names] |
| 326 |
|
| 327 |
if self.opt.type == "per-run": |
| 328 |
fields = attrs + [`data.stats[stats_name]` for stats_name in stats_names] |
| 329 |
print ",".join(fields) |
| 330 |
elif self.opt.type == "per-lease": |
| 331 |
leases = data.lease_stats |
| 332 |
for lease_id, lease_stat in leases.items(): |
| 333 |
fields = attrs + [`lease_id`] + [`lease_stat.get(lease_stat_name,"")` for lease_stat_name in lease_stats_names] |
| 334 |
print ",".join(fields) |
| 335 |
elif self.opt.type == "counter": |
| 336 |
for (time, lease_id, value, avg) in data.counters[counter]: |
| 337 |
fields = attrs + [`time`, `value`] |
| 338 |
if data.counter_avg_type[counter] != AccountingDataCollection.AVERAGE_NONE:
|
| 339 |
fields.append(`avg`) |
| 340 |
print ",".join(fields) |
| 341 |
|
| 342 |
|
| 343 |
|
| 344 |
class haizea_lwf2xml(Command): |
| 345 |
"""
|
| 346 |
Converts old Haizea LWF file into new XML-based LWF format
|
| 347 |
"""
|
| 348 |
|
| 349 |
name = "haizea-lwf2xml"
|
| 350 |
|
| 351 |
def __init__(self, argv): |
| 352 |
Command.__init__(self, argv)
|
| 353 |
|
| 354 |
self.optparser.add_option(Option("-i", "--in", action="store", type="string", dest="inf", |
| 355 |
help = """
|
| 356 |
Input file
|
| 357 |
"""))
|
| 358 |
self.optparser.add_option(Option("-o", "--out", action="store", type="string", dest="outf", |
| 359 |
help = """
|
| 360 |
Output file
|
| 361 |
"""))
|
| 362 |
|
| 363 |
def run(self): |
| 364 |
self.parse_options()
|
| 365 |
|
| 366 |
infile = self.opt.inf
|
| 367 |
outfile = self.opt.outf
|
| 368 |
|
| 369 |
root = ET.Element("lease-workload")
|
| 370 |
root.set("name", infile)
|
| 371 |
description = ET.SubElement(root, "description")
|
| 372 |
time = TimeDelta(seconds=0)
|
| 373 |
lease_id = 1
|
| 374 |
requests = ET.SubElement(root, "lease-requests")
|
| 375 |
|
| 376 |
|
| 377 |
infile = open(infile, "r") |
| 378 |
for line in infile: |
| 379 |
if line[0]!='#' and len(line.strip()) != 0: |
| 380 |
fields = line.split() |
| 381 |
submit_time = int(fields[0]) |
| 382 |
start_time = int(fields[1]) |
| 383 |
duration = int(fields[2]) |
| 384 |
real_duration = int(fields[3]) |
| 385 |
num_nodes = int(fields[4]) |
| 386 |
cpu = int(fields[5]) |
| 387 |
mem = int(fields[6]) |
| 388 |
disk = int(fields[7]) |
| 389 |
vm_image = fields[8]
|
| 390 |
vm_imagesize = int(fields[9]) |
| 391 |
|
| 392 |
|
| 393 |
|
| 394 |
lease_request = ET.SubElement(requests, "lease-request")
|
| 395 |
lease_request.set("arrival", str(TimeDelta(seconds=submit_time))) |
| 396 |
if real_duration != duration:
|
| 397 |
realduration = ET.SubElement(lease_request, "realduration")
|
| 398 |
realduration.set("time", str(TimeDelta(seconds=real_duration))) |
| 399 |
|
| 400 |
lease = ET.SubElement(lease_request, "lease")
|
| 401 |
lease.set("id", `lease_id`)
|
| 402 |
|
| 403 |
|
| 404 |
nodes = ET.SubElement(lease, "nodes")
|
| 405 |
node_set = ET.SubElement(nodes, "node-set")
|
| 406 |
node_set.set("numnodes", `num_nodes`)
|
| 407 |
res = ET.SubElement(node_set, "res")
|
| 408 |
res.set("type", "CPU") |
| 409 |
if cpu == 1: |
| 410 |
res.set("amount", "100") |
| 411 |
else:
|
| 412 |
pass
|
| 413 |
res = ET.SubElement(node_set, "res")
|
| 414 |
res.set("type", "Memory") |
| 415 |
res.set("amount", `mem`)
|
| 416 |
|
| 417 |
start = ET.SubElement(lease, "start")
|
| 418 |
if start_time == -1: |
| 419 |
lease.set("preemptible", "true") |
| 420 |
else:
|
| 421 |
lease.set("preemptible", "false") |
| 422 |
exact = ET.SubElement(start, "exact")
|
| 423 |
exact.set("time", str(TimeDelta(seconds=start_time))) |
| 424 |
|
| 425 |
duration_elem = ET.SubElement(lease, "duration")
|
| 426 |
duration_elem.set("time", str(TimeDelta(seconds=duration))) |
| 427 |
|
| 428 |
software = ET.SubElement(lease, "software")
|
| 429 |
diskimage = ET.SubElement(software, "disk-image")
|
| 430 |
diskimage.set("id", vm_image)
|
| 431 |
diskimage.set("size", `vm_imagesize`)
|
| 432 |
|
| 433 |
|
| 434 |
lease_id += 1
|
| 435 |
tree = ET.ElementTree(root) |
| 436 |
print ET.tostring(root)
|
| 437 |
|
| 438 |
|
| 439 |
|
| 440 |
class haizea_swf2lwf(Command): |
| 441 |
"""
|
| 442 |
Converts Standard Workload Format (SWF, used in the Parallel Workloads Archive at
|
| 443 |
http://www.cs.huji.ac.il/labs/parallel/workload/) to Lease Workload Format
|
| 444 |
"""
|
| 445 |
|
| 446 |
name = "haizea-swf2lwf"
|
| 447 |
|
| 448 |
def __init__(self, argv): |
| 449 |
Command.__init__(self, argv)
|
| 450 |
|
| 451 |
self.optparser.add_option(Option("-i", "--in", action="store", type="string", dest="inf", required=True, |
| 452 |
help = """
|
| 453 |
Input file
|
| 454 |
"""))
|
| 455 |
self.optparser.add_option(Option("-o", "--out", action="store", type="string", dest="outf", required=True, |
| 456 |
help = """
|
| 457 |
Output file
|
| 458 |
"""))
|
| 459 |
self.optparser.add_option(Option("-p", "--preemptible", action="store", type="string", dest="preemptible", required=True, |
| 460 |
help = """
|
| 461 |
Should the leases be preemptable or not?
|
| 462 |
"""))
|
| 463 |
self.optparser.add_option(Option("-f", "--from", action="store", type="string", dest="from_time", default="00:00:00:00", |
| 464 |
help = """
|
| 465 |
This parameter, together with the --amount parameter, allows converting just
|
| 466 |
an interval of the SWF file.
|
| 467 |
|
| 468 |
This parameter must be a timestamp of the format DD:HH:MM:SS. Only jobs sumitted
|
| 469 |
at or after that time will be converted.
|
| 470 |
|
| 471 |
Default is 00:00:00:00
|
| 472 |
"""))
|
| 473 |
self.optparser.add_option(Option("-l", "--interval-length", action="store", type="string", dest="interval_length", |
| 474 |
help = """
|
| 475 |
Length of the interval in format DD:HH:MM:SS. Default is to convert jobs until the
|
| 476 |
end of the SWF file.
|
| 477 |
"""))
|
| 478 |
self.optparser.add_option(Option("-q", "--queues", action="store", type="string", dest="queues", |
| 479 |
help = """
|
| 480 |
Only convert jobs from the specified queues
|
| 481 |
"""))
|
| 482 |
self.optparser.add_option(Option("-m", "--memory", action="store", type="string", dest="mem", |
| 483 |
help = """
|
| 484 |
Memory requested by jobs.
|
| 485 |
"""))
|
| 486 |
self.optparser.add_option(Option("-s", "--scale", action="store", type="string", dest="scale", |
| 487 |
help = """
|
| 488 |
Scale number of processors by 1/SCALE.
|
| 489 |
"""))
|
| 490 |
|
| 491 |
def run(self): |
| 492 |
self.parse_options()
|
| 493 |
|
| 494 |
infile = self.opt.inf
|
| 495 |
outfile = self.opt.outf
|
| 496 |
|
| 497 |
from_time = Parser.DateTimeDeltaFromString(self.opt.from_time)
|
| 498 |
if self.opt.interval_length == None: |
| 499 |
to_time = None
|
| 500 |
else:
|
| 501 |
to_time = from_time + Parser.DateTimeDeltaFromString(self.opt.interval_length)
|
| 502 |
|
| 503 |
root = ET.Element("lease-workload")
|
| 504 |
root.set("name", infile)
|
| 505 |
description = ET.SubElement(root, "description")
|
| 506 |
description.text = "Created with haizea-swf2lwf %s" % " ".join(self.argv[1:]) |
| 507 |
time = TimeDelta(seconds=0)
|
| 508 |
requests = ET.SubElement(root, "lease-requests")
|
| 509 |
|
| 510 |
slowdowns = [] |
| 511 |
users = set()
|
| 512 |
|
| 513 |
infile = open(infile, "r") |
| 514 |
for line in infile: |
| 515 |
if line[0]!=';' and len(line.strip()) != 0: |
| 516 |
fields = line.split() |
| 517 |
|
| 518 |
# Unpack the job's attributes. The description of each field is
|
| 519 |
# taken from the SWF documentation at
|
| 520 |
# http://www.cs.huji.ac.il/labs/parallel/workload/swf.html
|
| 521 |
|
| 522 |
# Job Number -- a counter field, starting from 1.
|
| 523 |
job_number = int(fields[0]) |
| 524 |
|
| 525 |
# Submit Time -- in seconds. The earliest time the log refers to is zero,
|
| 526 |
# and is the submittal time the of the first job. The lines in the log are
|
| 527 |
# sorted by ascending submittal times. It makes sense for jobs to also be
|
| 528 |
# numbered in this order.
|
| 529 |
submit_time = int(fields[1]) |
| 530 |
|
| 531 |
# Wait Time -- in seconds. The difference between the job's submit time
|
| 532 |
# and the time at which it actually began to run. Naturally, this is only
|
| 533 |
# relevant to real logs, not to models.
|
| 534 |
wait_time = int(fields[2]) |
| 535 |
|
| 536 |
# Run Time -- in seconds. The wall clock time the job was running (end
|
| 537 |
# time minus start time).
|
| 538 |
# We decided to use ``wait time'' and ``run time'' instead of the equivalent
|
| 539 |
# ``start time'' and ``end time'' because they are directly attributable to
|
| 540 |
# the scheduler and application, and are more suitable for models where only
|
| 541 |
# the run time is relevant.
|
| 542 |
# Note that when values are rounded to an integral number of seconds (as
|
| 543 |
# often happens in logs) a run time of 0 is possible and means the job ran
|
| 544 |
# for less than 0.5 seconds. On the other hand it is permissable to use
|
| 545 |
# floating point values for time fields.
|
| 546 |
run_time = int(fields[3]) |
| 547 |
|
| 548 |
# Number of Allocated Processors -- an integer. In most cases this is also
|
| 549 |
# the number of processors the job uses; if the job does not use all of them,
|
| 550 |
# we typically don't know about it.
|
| 551 |
num_processors_allocated = int(fields[4]) |
| 552 |
|
| 553 |
# Average CPU Time Used -- both user and system, in seconds. This is the
|
| 554 |
# average over all processors of the CPU time used, and may therefore be
|
| 555 |
# smaller than the wall clock runtime. If a log contains the total CPU time
|
| 556 |
# used by all the processors, it is divided by the number of allocated
|
| 557 |
# processors to derive the average.
|
| 558 |
avg_cpu_time = float(fields[5]) |
| 559 |
|
| 560 |
# Used Memory -- in kilobytes. This is again the average per processor.
|
| 561 |
used_memory = int(fields[6]) |
| 562 |
|
| 563 |
# Requested Number of Processors.
|
| 564 |
num_processors_requested = int(fields[7]) |
| 565 |
|
| 566 |
# Requested Time. This can be either runtime (measured in wallclock seconds),
|
| 567 |
# or average CPU time per processor (also in seconds) -- the exact meaning
|
| 568 |
# is determined by a header comment. In many logs this field is used for
|
| 569 |
# the user runtime estimate (or upper bound) used in backfilling. If a log
|
| 570 |
# contains a request for total CPU time, it is divided by the number of
|
| 571 |
# requested processors.
|
| 572 |
time_requested = int(fields[8]) |
| 573 |
|
| 574 |
# Requested Memory (again kilobytes per processor).
|
| 575 |
mem_requested = int(fields[9]) |
| 576 |
|
| 577 |
# Status 1 if the job was completed, 0 if it failed, and 5 if cancelled.
|
| 578 |
# If information about chekcpointing or swapping is included, other values
|
| 579 |
# are also possible. See usage note below. This field is meaningless for
|
| 580 |
# models, so would be -1.
|
| 581 |
status = int(fields[10]) |
| 582 |
|
| 583 |
# User ID -- a natural number, between one and the number of different users.
|
| 584 |
user_id = int(fields[11]) |
| 585 |
|
| 586 |
# Group ID -- a natural number, between one and the number of different groups.
|
| 587 |
# Some systems control resource usage by groups rather than by individual users.
|
| 588 |
group_id = int(fields[12]) |
| 589 |
|
| 590 |
# Executable (Application) Number -- a natural number, between one and the number
|
| 591 |
# of different applications appearing in the workload. in some logs, this might
|
| 592 |
# represent a script file used to run jobs rather than the executable directly;
|
| 593 |
# this should be noted in a header comment.
|
| 594 |
exec_number = int(fields[13]) |
| 595 |
|
| 596 |
# Queue Number -- a natural number, between one and the number of different
|
| 597 |
# queues in the system. The nature of the system's queues should be explained
|
| 598 |
# in a header comment. This field is where batch and interactive jobs should
|
| 599 |
# be differentiated: we suggest the convention of denoting interactive jobs by 0.
|
| 600 |
queue = int(fields[14]) |
| 601 |
|
| 602 |
# Partition Number -- a natural number, between one and the number of different
|
| 603 |
# partitions in the systems. The nature of the system's partitions should be
|
| 604 |
# explained in a header comment. For example, it is possible to use partition
|
| 605 |
# numbers to identify which machine in a cluster was used.
|
| 606 |
partition = int(fields[15]) |
| 607 |
|
| 608 |
# Preceding Job Number -- this is the number of a previous job in the workload,
|
| 609 |
# such that the current job can only start after the termination of this preceding
|
| 610 |
# job. Together with the next field, this allows the workload to include feedback
|
| 611 |
# as described below.
|
| 612 |
prec_job = int(fields[16]) |
| 613 |
|
| 614 |
# Think Time from Preceding Job -- this is the number of seconds that should elapse
|
| 615 |
# between the termination of the preceding job and the submittal of this one.
|
| 616 |
prec_job_thinktime = int(fields[17]) |
| 617 |
|
| 618 |
|
| 619 |
# Check if we have to skip this job
|
| 620 |
|
| 621 |
submit_time = TimeDelta(seconds=submit_time) |
| 622 |
|
| 623 |
if submit_time < from_time:
|
| 624 |
continue
|
| 625 |
|
| 626 |
if to_time != None and submit_time > to_time: |
| 627 |
break
|
| 628 |
|
| 629 |
if run_time < 0 and status==5: |
| 630 |
# This is a job that got cancelled while waiting in the queue
|
| 631 |
continue
|
| 632 |
|
| 633 |
if self.opt.queues != None: |
| 634 |
queues = [int(q) for q in self.opt.queues.split(",")] |
| 635 |
if queue not in queues: |
| 636 |
# Job was submitted to a queue we're filtering out
|
| 637 |
continue
|
| 638 |
|
| 639 |
if self.opt.scale != None: |
| 640 |
num_processors_requested = int(num_processors_requested/int(self.opt.scale)) |
| 641 |
|
| 642 |
# Make submission time relative to starting time of trace
|
| 643 |
submit_time = submit_time - from_time |
| 644 |
|
| 645 |
lease_request = ET.SubElement(requests, "lease-request")
|
| 646 |
lease_request.set("arrival", str(submit_time)) |
| 647 |
|
| 648 |
if run_time == 0: |
| 649 |
# As specified in the SWF documentation, a runtime of 0 means
|
| 650 |
# the job ran for less than a second, so we round up to 1.
|
| 651 |
run_time = 1
|
| 652 |
realduration = ET.SubElement(lease_request, "realduration")
|
| 653 |
realduration.set("time", str(TimeDelta(seconds=run_time))) |
| 654 |
|
| 655 |
lease = ET.SubElement(lease_request, "lease")
|
| 656 |
lease.set("id", `job_number`)
|
| 657 |
|
| 658 |
|
| 659 |
nodes = ET.SubElement(lease, "nodes")
|
| 660 |
node_set = ET.SubElement(nodes, "node-set")
|
| 661 |
node_set.set("numnodes", `num_processors_requested`)
|
| 662 |
res = ET.SubElement(node_set, "res")
|
| 663 |
res.set("type", "CPU") |
| 664 |
res.set("amount", "100") |
| 665 |
|
| 666 |
res = ET.SubElement(node_set, "res")
|
| 667 |
res.set("type", "Memory") |
| 668 |
if self.opt.mem != None: |
| 669 |
res.set("amount", self.opt.mem) |
| 670 |
elif mem_requested != -1: |
| 671 |
res.set("amount", `mem_requested / 1024`) |
| 672 |
else:
|
| 673 |
print "Cannot convert this file. Job #%i does not specify requested memory, and --memory parameter not specified" % job_number |
| 674 |
exit(-1) |
| 675 |
|
| 676 |
if run_time < 10: |
| 677 |
run_time2 = 10
|
| 678 |
else:
|
| 679 |
run_time2 = run_time |
| 680 |
slowdown = wait_time + run_time2 / float(run_time2)
|
| 681 |
slowdowns.append(slowdown) |
| 682 |
|
| 683 |
if not user_id in users: |
| 684 |
users.add(user_id) |
| 685 |
|
| 686 |
start = ET.SubElement(lease, "start")
|
| 687 |
#lease.set("preemptible", self.opt.preemptible)
|
| 688 |
lease.set("user", `user_id`)
|
| 689 |
|
| 690 |
duration_elem = ET.SubElement(lease, "duration")
|
| 691 |
duration_elem.set("time", str(TimeDelta(seconds=time_requested))) |
| 692 |
|
| 693 |
# No software environment specified. The annotator would have to be used to
|
| 694 |
# add one (or an image file when running a simulation).
|
| 695 |
software = ET.SubElement(lease, "software")
|
| 696 |
diskimage = ET.SubElement(software, "none")
|
| 697 |
|
| 698 |
# Add unused SWF attributes to the extra section, for future reference.
|
| 699 |
extra = ET.SubElement(lease, "extra")
|
| 700 |
attr = ET.SubElement(extra, "attr")
|
| 701 |
attr.set("name", "SWF_waittime") |
| 702 |
attr.set("value", `wait_time`)
|
| 703 |
attr = ET.SubElement(extra, "attr")
|
| 704 |
attr.set("name", "SWF_avgcputime") |
| 705 |
attr.set("value", `avg_cpu_time`)
|
| 706 |
attr = ET.SubElement(extra, "attr")
|
| 707 |
attr.set("name", "SWF_queue") |
| 708 |
attr.set("value", `queue`)
|
| 709 |
attr = ET.SubElement(extra, "attr")
|
| 710 |
attr.set("name", "SWF_group") |
| 711 |
attr.set("value", `group_id`)
|
| 712 |
attr = ET.SubElement(extra, "attr")
|
| 713 |
attr.set("name", "SWF_execnumber") |
| 714 |
attr.set("value", `exec_number`)
|
| 715 |
|
| 716 |
tree = ET.ElementTree(root) |
| 717 |
|
| 718 |
outfile = open(outfile, "w") |
| 719 |
tree.write(outfile) |
| 720 |
|
| 721 |
infile.close() |
| 722 |
outfile.close() |
| 723 |
|
| 724 |
slowdowns.sort() |
| 725 |
|
| 726 |
print "SLOWDOWNS" |
| 727 |
print "---------" |
| 728 |
print "min: %.2f" % slowdowns[0] |
| 729 |
print "10p: %.2f" % percentile(slowdowns, 0.1) |
| 730 |
print "25p: %.2f" % percentile(slowdowns, 0.25) |
| 731 |
print "med: %.2f" % percentile(slowdowns, 0.5) |
| 732 |
print "75p: %.2f" % percentile(slowdowns, 0.75) |
| 733 |
print "90p: %.2f" % percentile(slowdowns, 0.9) |
| 734 |
print "max: %.2f" % slowdowns[-1] |
| 735 |
print
|
| 736 |
print "USERS" |
| 737 |
print "-----" |
| 738 |
print "Number of users: %i" % len(users) |
| 739 |
|