root / tools / burnin @ ec5c88dc
History | View | Annotate | Download (17.3 kB)
| 1 |
#!/usr/bin/python |
|---|---|
| 2 |
# |
| 3 |
|
| 4 |
# Copyright (C) 2006, 2007 Google Inc. |
| 5 |
# |
| 6 |
# This program is free software; you can redistribute it and/or modify |
| 7 |
# it under the terms of the GNU General Public License as published by |
| 8 |
# the Free Software Foundation; either version 2 of the License, or |
| 9 |
# (at your option) any later version. |
| 10 |
# |
| 11 |
# This program is distributed in the hope that it will be useful, but |
| 12 |
# WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 |
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 |
# General Public License for more details. |
| 15 |
# |
| 16 |
# You should have received a copy of the GNU General Public License |
| 17 |
# along with this program; if not, write to the Free Software |
| 18 |
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA |
| 19 |
# 02110-1301, USA. |
| 20 |
|
| 21 |
|
| 22 |
"""Burnin program |
| 23 |
|
| 24 |
""" |
| 25 |
|
| 26 |
import os |
| 27 |
import sys |
| 28 |
import optparse |
| 29 |
import time |
| 30 |
from itertools import izip, islice, cycle |
| 31 |
from cStringIO import StringIO |
| 32 |
|
| 33 |
from ganeti import opcodes |
| 34 |
from ganeti import mcpu |
| 35 |
from ganeti import constants |
| 36 |
from ganeti import cli |
| 37 |
from ganeti import logger |
| 38 |
from ganeti import errors |
| 39 |
from ganeti import utils |
| 40 |
|
| 41 |
|
| 42 |
USAGE = ("\tburnin -o OS_NAME [options...] instance_name ...")
|
| 43 |
|
| 44 |
|
| 45 |
def Usage(): |
| 46 |
"""Shows program usage information and exits the program.""" |
| 47 |
|
| 48 |
print >> sys.stderr, "Usage:" |
| 49 |
print >> sys.stderr, USAGE |
| 50 |
sys.exit(2) |
| 51 |
|
| 52 |
|
| 53 |
def Log(msg): |
| 54 |
"""Simple function that prints out its argument. |
| 55 |
|
| 56 |
""" |
| 57 |
print msg |
| 58 |
sys.stdout.flush() |
| 59 |
|
| 60 |
|
| 61 |
class Burner(object): |
| 62 |
"""Burner class.""" |
| 63 |
|
| 64 |
def __init__(self): |
| 65 |
"""Constructor.""" |
| 66 |
logger.SetupLogging(constants.LOG_BURNIN, debug=False, stderr_logging=True) |
| 67 |
self._feed_buf = StringIO() |
| 68 |
self.nodes = [] |
| 69 |
self.instances = [] |
| 70 |
self.to_rem = [] |
| 71 |
self.opts = None |
| 72 |
self.cl = cli.GetClient() |
| 73 |
self.ParseOptions() |
| 74 |
self.GetState() |
| 75 |
|
| 76 |
def ClearFeedbackBuf(self): |
| 77 |
"""Clear the feedback buffer.""" |
| 78 |
self._feed_buf.truncate(0) |
| 79 |
|
| 80 |
def GetFeedbackBuf(self): |
| 81 |
"""Return the contents of the buffer.""" |
| 82 |
return self._feed_buf.getvalue() |
| 83 |
|
| 84 |
def Feedback(self, msg): |
| 85 |
"""Acumulate feedback in our buffer.""" |
| 86 |
self._feed_buf.write("%s %s\n" % (time.ctime(utils.MergeTime(msg[0])),
|
| 87 |
msg[2])) |
| 88 |
if self.opts.verbose: |
| 89 |
Log(msg) |
| 90 |
|
| 91 |
def ExecOp(self, op): |
| 92 |
"""Execute an opcode and manage the exec buffer.""" |
| 93 |
self.ClearFeedbackBuf() |
| 94 |
return cli.SubmitOpCode(op, feedback_fn=self.Feedback, cl=self.cl) |
| 95 |
|
| 96 |
def ExecJobSet(self, jobs): |
| 97 |
"""Execute a set of jobs and return once all are done. |
| 98 |
|
| 99 |
The method will return the list of results, if all jobs are |
| 100 |
successfull. Otherwise, OpExecError will be raised from within |
| 101 |
cli.py. |
| 102 |
|
| 103 |
""" |
| 104 |
self.ClearFeedbackBuf() |
| 105 |
job_ids = [cli.SendJob(job, cl=self.cl) for job in jobs] |
| 106 |
Log("- Submitted job IDs %s" % ", ".join(job_ids))
|
| 107 |
results = [] |
| 108 |
for jid in job_ids: |
| 109 |
Log("- Waiting for job %s" % jid)
|
| 110 |
results.append(cli.PollJob(jid, cl=self.cl, feedback_fn=self.Feedback)) |
| 111 |
|
| 112 |
return results |
| 113 |
|
| 114 |
def ParseOptions(self): |
| 115 |
"""Parses the command line options. |
| 116 |
|
| 117 |
In case of command line errors, it will show the usage and exit the |
| 118 |
program. |
| 119 |
|
| 120 |
""" |
| 121 |
|
| 122 |
parser = optparse.OptionParser(usage="\n%s" % USAGE, |
| 123 |
version="%%prog (ganeti) %s" % |
| 124 |
constants.RELEASE_VERSION, |
| 125 |
option_class=cli.CliOption) |
| 126 |
|
| 127 |
parser.add_option("-o", "--os", dest="os", default=None,
|
| 128 |
help="OS to use during burnin", |
| 129 |
metavar="<OS>") |
| 130 |
parser.add_option("--os-size", dest="os_size", help="Disk size",
|
| 131 |
default=4 * 1024, type="unit", metavar="<size>") |
| 132 |
parser.add_option("--swap-size", dest="swap_size", help="Swap size",
|
| 133 |
default=4 * 1024, type="unit", metavar="<size>") |
| 134 |
parser.add_option("--mem-size", dest="mem_size", help="Memory size",
|
| 135 |
default=128, type="unit", metavar="<size>") |
| 136 |
parser.add_option("-v", "--verbose",
|
| 137 |
action="store_true", dest="verbose", default=False, |
| 138 |
help="print command execution messages to stdout") |
| 139 |
parser.add_option("--no-replace1", dest="do_replace1",
|
| 140 |
help="Skip disk replacement with the same secondary", |
| 141 |
action="store_false", default=True) |
| 142 |
parser.add_option("--no-replace2", dest="do_replace2",
|
| 143 |
help="Skip disk replacement with a different secondary", |
| 144 |
action="store_false", default=True) |
| 145 |
parser.add_option("--no-failover", dest="do_failover",
|
| 146 |
help="Skip instance failovers", action="store_false", |
| 147 |
default=True) |
| 148 |
parser.add_option("--no-importexport", dest="do_importexport",
|
| 149 |
help="Skip instance export/import", action="store_false", |
| 150 |
default=True) |
| 151 |
parser.add_option("--no-startstop", dest="do_startstop",
|
| 152 |
help="Skip instance stop/start", action="store_false", |
| 153 |
default=True) |
| 154 |
parser.add_option("--rename", dest="rename", default=None,
|
| 155 |
help="Give one unused instance name which is taken" |
| 156 |
" to start the renaming sequence", |
| 157 |
metavar="<instance_name>") |
| 158 |
parser.add_option("-t", "--disk-template", dest="disk_template",
|
| 159 |
choices=("diskless", "file", "plain", "drbd"),
|
| 160 |
default="drbd", |
| 161 |
help="Disk template (diskless, file, plain or drbd)" |
| 162 |
" [drbd]") |
| 163 |
parser.add_option("-n", "--nodes", dest="nodes", default="",
|
| 164 |
help="Comma separated list of nodes to perform" |
| 165 |
" the burnin on (defaults to all nodes)") |
| 166 |
parser.add_option("--iallocator", dest="iallocator",
|
| 167 |
default=None, type="string", |
| 168 |
help="Perform the allocation using an iallocator" |
| 169 |
" instead of fixed node spread (node restrictions no" |
| 170 |
" longer apply, therefore -n/--nodes must not be used") |
| 171 |
parser.add_option("-p", "--parallel", default=False, action="store_true",
|
| 172 |
dest="parallel", |
| 173 |
help="Enable parallelization of some operations in" |
| 174 |
" order to speed burnin or to test granular locking") |
| 175 |
|
| 176 |
options, args = parser.parse_args() |
| 177 |
if len(args) < 1 or options.os is None: |
| 178 |
Usage() |
| 179 |
|
| 180 |
supported_disk_templates = (constants.DT_DISKLESS, |
| 181 |
constants.DT_FILE, |
| 182 |
constants.DT_PLAIN, |
| 183 |
constants.DT_DRBD8) |
| 184 |
if options.disk_template not in supported_disk_templates: |
| 185 |
Log("Unknown disk template '%s'" % options.disk_template)
|
| 186 |
sys.exit(1) |
| 187 |
|
| 188 |
if options.nodes and options.iallocator: |
| 189 |
Log("Give either the nodes option or the iallocator option, not both")
|
| 190 |
sys.exit(1) |
| 191 |
|
| 192 |
self.opts = options |
| 193 |
self.instances = args |
| 194 |
|
| 195 |
def GetState(self): |
| 196 |
"""Read the cluster state from the config.""" |
| 197 |
if self.opts.nodes: |
| 198 |
names = self.opts.nodes.split(",")
|
| 199 |
else: |
| 200 |
names = [] |
| 201 |
try: |
| 202 |
op = opcodes.OpQueryNodes(output_fields=["name"], names=names) |
| 203 |
result = self.ExecOp(op) |
| 204 |
except errors.GenericError, err: |
| 205 |
err_code, msg = cli.FormatError(err) |
| 206 |
Log(msg) |
| 207 |
sys.exit(err_code) |
| 208 |
self.nodes = [data[0] for data in result] |
| 209 |
|
| 210 |
result = self.ExecOp(opcodes.OpDiagnoseOS(output_fields=["name", "valid"], |
| 211 |
names=[])) |
| 212 |
|
| 213 |
if not result: |
| 214 |
Log("Can't get the OS list")
|
| 215 |
sys.exit(1) |
| 216 |
|
| 217 |
# filter non-valid OS-es |
| 218 |
os_set = [val[0] for val in result if val[1]] |
| 219 |
|
| 220 |
if self.opts.os not in os_set: |
| 221 |
Log("OS '%s' not found" % self.opts.os)
|
| 222 |
sys.exit(1) |
| 223 |
|
| 224 |
def CreateInstances(self): |
| 225 |
"""Create the given instances. |
| 226 |
|
| 227 |
""" |
| 228 |
self.to_rem = [] |
| 229 |
mytor = izip(cycle(self.nodes), |
| 230 |
islice(cycle(self.nodes), 1, None), |
| 231 |
self.instances) |
| 232 |
jobset = [] |
| 233 |
for pnode, snode, instance in mytor: |
| 234 |
if self.opts.iallocator: |
| 235 |
pnode = snode = None |
| 236 |
Log("- Add instance %s (iallocator: %s)" %
|
| 237 |
(instance, self.opts.iallocator)) |
| 238 |
elif self.opts.disk_template not in constants.DTS_NET_MIRROR: |
| 239 |
snode = None |
| 240 |
Log("- Add instance %s on node %s" % (instance, pnode))
|
| 241 |
else: |
| 242 |
Log("- Add instance %s on nodes %s/%s" % (instance, pnode, snode))
|
| 243 |
|
| 244 |
op = opcodes.OpCreateInstance(instance_name=instance, |
| 245 |
mem_size=self.opts.mem_size, |
| 246 |
disk_size=self.opts.os_size, |
| 247 |
swap_size=self.opts.swap_size, |
| 248 |
disk_template=self.opts.disk_template, |
| 249 |
mode=constants.INSTANCE_CREATE, |
| 250 |
os_type=self.opts.os, |
| 251 |
pnode=pnode, |
| 252 |
snode=snode, |
| 253 |
vcpus=1, |
| 254 |
start=True, |
| 255 |
ip_check=True, |
| 256 |
wait_for_sync=True, |
| 257 |
mac="auto", |
| 258 |
kernel_path=None, |
| 259 |
initrd_path=None, |
| 260 |
hvm_boot_order=None, |
| 261 |
file_driver="loop", |
| 262 |
file_storage_dir=None, |
| 263 |
iallocator=self.opts.iallocator, |
| 264 |
hvm_nic_type=constants.HT_HVM_NIC_RTL8139, |
| 265 |
hvm_disk_type=constants.HT_HVM_DEV_IOEMU) |
| 266 |
|
| 267 |
if self.opts.parallel: |
| 268 |
jobset.append([op]) |
| 269 |
# FIXME: here we should not append to to_rem uncoditionally, |
| 270 |
# but only when the job is successful |
| 271 |
self.to_rem.append(instance) |
| 272 |
else: |
| 273 |
self.ExecOp(op) |
| 274 |
self.to_rem.append(instance) |
| 275 |
if self.opts.parallel: |
| 276 |
self.ExecJobSet(jobset) |
| 277 |
|
| 278 |
def ReplaceDisks1D8(self): |
| 279 |
"""Replace disks on primary and secondary for drbd8.""" |
| 280 |
for instance in self.instances: |
| 281 |
for mode in constants.REPLACE_DISK_SEC, constants.REPLACE_DISK_PRI: |
| 282 |
op = opcodes.OpReplaceDisks(instance_name=instance, |
| 283 |
mode=mode, |
| 284 |
disks=["sda", "sdb"]) |
| 285 |
Log("- Replace disks (%s) for instance %s" % (mode, instance))
|
| 286 |
self.ExecOp(op) |
| 287 |
|
| 288 |
def ReplaceDisks2(self): |
| 289 |
"""Replace secondary node.""" |
| 290 |
mode = constants.REPLACE_DISK_SEC |
| 291 |
|
| 292 |
mytor = izip(islice(cycle(self.nodes), 2, None), |
| 293 |
self.instances) |
| 294 |
for tnode, instance in mytor: |
| 295 |
if self.opts.iallocator: |
| 296 |
tnode = None |
| 297 |
op = opcodes.OpReplaceDisks(instance_name=instance, |
| 298 |
mode=mode, |
| 299 |
remote_node=tnode, |
| 300 |
iallocator=self.opts.iallocator, |
| 301 |
disks=["sda", "sdb"]) |
| 302 |
Log("- Replace secondary (%s) for instance %s" % (mode, instance))
|
| 303 |
self.ExecOp(op) |
| 304 |
|
| 305 |
def Failover(self): |
| 306 |
"""Failover the instances.""" |
| 307 |
|
| 308 |
for instance in self.instances: |
| 309 |
op = opcodes.OpFailoverInstance(instance_name=instance, |
| 310 |
ignore_consistency=False) |
| 311 |
|
| 312 |
Log("- Failover instance %s" % (instance))
|
| 313 |
self.ExecOp(op) |
| 314 |
|
| 315 |
def ImportExport(self): |
| 316 |
"""Export the instance, delete it, and import it back. |
| 317 |
|
| 318 |
""" |
| 319 |
|
| 320 |
mytor = izip(cycle(self.nodes), |
| 321 |
islice(cycle(self.nodes), 1, None), |
| 322 |
islice(cycle(self.nodes), 2, None), |
| 323 |
self.instances) |
| 324 |
|
| 325 |
for pnode, snode, enode, instance in mytor: |
| 326 |
|
| 327 |
if self.opts.iallocator: |
| 328 |
pnode = snode = None |
| 329 |
import_log_msg = ("- Import instance %s from node %s (iallocator: %s)" %
|
| 330 |
(instance, enode, self.opts.iallocator)) |
| 331 |
elif self.opts.disk_template not in constants.DTS_NET_MIRROR: |
| 332 |
snode = None |
| 333 |
import_log_msg = ("- Import instance %s from node %s to node %s" %
|
| 334 |
(instance, enode, pnode)) |
| 335 |
else: |
| 336 |
import_log_msg = ("- Import instance %s from node %s to nodes %s/%s" %
|
| 337 |
(instance, enode, pnode, snode)) |
| 338 |
|
| 339 |
exp_op = opcodes.OpExportInstance(instance_name=instance, |
| 340 |
target_node=enode, |
| 341 |
shutdown=True) |
| 342 |
rem_op = opcodes.OpRemoveInstance(instance_name=instance, |
| 343 |
ignore_failures=True) |
| 344 |
nam_op = opcodes.OpQueryInstances(output_fields=["name"], |
| 345 |
names=[instance]) |
| 346 |
full_name = self.ExecOp(nam_op)[0][0] |
| 347 |
imp_dir = os.path.join(constants.EXPORT_DIR, full_name) |
| 348 |
imp_op = opcodes.OpCreateInstance(instance_name=instance, |
| 349 |
mem_size=128, |
| 350 |
disk_size=self.opts.os_size, |
| 351 |
swap_size=self.opts.swap_size, |
| 352 |
disk_template=self.opts.disk_template, |
| 353 |
mode=constants.INSTANCE_IMPORT, |
| 354 |
src_node=enode, |
| 355 |
src_path=imp_dir, |
| 356 |
pnode=pnode, |
| 357 |
snode=snode, |
| 358 |
vcpus=1, |
| 359 |
start=True, |
| 360 |
ip_check=True, |
| 361 |
wait_for_sync=True, |
| 362 |
mac="auto", |
| 363 |
file_storage_dir=None, |
| 364 |
file_driver=None, |
| 365 |
iallocator=self.opts.iallocator, |
| 366 |
hvm_nic_type= |
| 367 |
constants.HT_HVM_NIC_RTL8139, |
| 368 |
hvm_disk_type= |
| 369 |
constants.HT_HVM_DEV_IOEMU) |
| 370 |
|
| 371 |
erem_op = opcodes.OpRemoveExport(instance_name=instance) |
| 372 |
|
| 373 |
Log("- Export instance %s to node %s" % (instance, enode))
|
| 374 |
self.ExecOp(exp_op) |
| 375 |
Log("- Remove instance %s" % (instance))
|
| 376 |
self.ExecOp(rem_op) |
| 377 |
self.to_rem.remove(instance) |
| 378 |
Log(import_log_msg) |
| 379 |
self.ExecOp(imp_op) |
| 380 |
Log("- Remove export of instance %s" % (instance))
|
| 381 |
self.ExecOp(erem_op) |
| 382 |
|
| 383 |
self.to_rem.append(instance) |
| 384 |
|
| 385 |
def StopInstance(self, instance): |
| 386 |
"""Stop given instance.""" |
| 387 |
op = opcodes.OpShutdownInstance(instance_name=instance) |
| 388 |
Log("- Shutdown instance %s" % instance)
|
| 389 |
self.ExecOp(op) |
| 390 |
|
| 391 |
def StartInstance(self, instance): |
| 392 |
"""Start given instance.""" |
| 393 |
op = opcodes.OpStartupInstance(instance_name=instance, force=False) |
| 394 |
Log("- Start instance %s" % instance)
|
| 395 |
self.ExecOp(op) |
| 396 |
|
| 397 |
def RenameInstance(self, instance, instance_new): |
| 398 |
"""Rename instance.""" |
| 399 |
op = opcodes.OpRenameInstance(instance_name=instance, |
| 400 |
new_name=instance_new) |
| 401 |
Log("- Rename instance %s to %s" % (instance, instance_new))
|
| 402 |
self.ExecOp(op) |
| 403 |
|
| 404 |
def StopStart(self): |
| 405 |
"""Stop/start the instances.""" |
| 406 |
for instance in self.instances: |
| 407 |
self.StopInstance(instance) |
| 408 |
self.StartInstance(instance) |
| 409 |
|
| 410 |
def Remove(self): |
| 411 |
"""Remove the instances.""" |
| 412 |
for instance in self.to_rem: |
| 413 |
op = opcodes.OpRemoveInstance(instance_name=instance, |
| 414 |
ignore_failures=True) |
| 415 |
Log("- Remove instance %s" % instance)
|
| 416 |
self.ExecOp(op) |
| 417 |
|
| 418 |
|
| 419 |
def Rename(self): |
| 420 |
"""Rename the instances.""" |
| 421 |
rename = self.opts.rename |
| 422 |
for instance in self.instances: |
| 423 |
self.StopInstance(instance) |
| 424 |
self.RenameInstance(instance, rename) |
| 425 |
self.StartInstance(rename) |
| 426 |
self.StopInstance(rename) |
| 427 |
self.RenameInstance(rename, instance) |
| 428 |
self.StartInstance(instance) |
| 429 |
|
| 430 |
def BurninCluster(self): |
| 431 |
"""Test a cluster intensively. |
| 432 |
|
| 433 |
This will create instances and then start/stop/failover them. |
| 434 |
It is safe for existing instances but could impact performance. |
| 435 |
|
| 436 |
""" |
| 437 |
|
| 438 |
opts = self.opts |
| 439 |
|
| 440 |
Log("- Testing global parameters")
|
| 441 |
|
| 442 |
if (len(self.nodes) == 1 and |
| 443 |
opts.disk_template not in (constants.DT_DISKLESS, constants.DT_PLAIN, |
| 444 |
constants.DT_FILE)): |
| 445 |
Log("When one node is available/selected the disk template must"
|
| 446 |
" be 'diskless', 'file' or 'plain'") |
| 447 |
sys.exit(1) |
| 448 |
|
| 449 |
has_err = True |
| 450 |
try: |
| 451 |
self.CreateInstances() |
| 452 |
if opts.do_replace1 and opts.disk_template in constants.DTS_NET_MIRROR: |
| 453 |
self.ReplaceDisks1D8() |
| 454 |
if (opts.do_replace2 and len(self.nodes) > 2 and |
| 455 |
opts.disk_template in constants.DTS_NET_MIRROR) : |
| 456 |
self.ReplaceDisks2() |
| 457 |
|
| 458 |
if opts.do_failover and opts.disk_template in constants.DTS_NET_MIRROR: |
| 459 |
self.Failover() |
| 460 |
|
| 461 |
if opts.do_importexport: |
| 462 |
self.ImportExport() |
| 463 |
|
| 464 |
if opts.do_startstop: |
| 465 |
self.StopStart() |
| 466 |
|
| 467 |
if opts.rename: |
| 468 |
self.Rename() |
| 469 |
|
| 470 |
has_err = False |
| 471 |
finally: |
| 472 |
if has_err: |
| 473 |
Log("Error detected: opcode buffer follows:\n\n")
|
| 474 |
Log(self.GetFeedbackBuf()) |
| 475 |
Log("\n\n")
|
| 476 |
self.Remove() |
| 477 |
|
| 478 |
return 0 |
| 479 |
|
| 480 |
|
| 481 |
def main(): |
| 482 |
"""Main function""" |
| 483 |
|
| 484 |
burner = Burner() |
| 485 |
return burner.BurninCluster() |
| 486 |
|
| 487 |
|
| 488 |
if __name__ == "__main__": |
| 489 |
main() |