Revision 5178f1bc tools/burnin
| b/tools/burnin | ||
|---|---|---|
| 27 | 27 |
import sys |
| 28 | 28 |
import optparse |
| 29 | 29 |
import time |
| 30 |
import socket |
|
| 31 |
import urllib2 |
|
| 32 |
import errno |
|
| 30 | 33 |
from itertools import izip, islice, cycle |
| 31 | 34 |
from cStringIO import StringIO |
| 32 | 35 |
|
| ... | ... | |
| 41 | 44 |
USAGE = ("\tburnin -o OS_NAME [options...] instance_name ...")
|
| 42 | 45 |
|
| 43 | 46 |
|
| 47 |
class InstanceDown(Exception): |
|
| 48 |
"""The checked instance was not up""" |
|
| 49 |
|
|
| 50 |
|
|
| 44 | 51 |
def Usage(): |
| 45 | 52 |
"""Shows program usage information and exits the program.""" |
| 46 | 53 |
|
| ... | ... | |
| 190 | 197 |
dest="parallel", |
| 191 | 198 |
help="Enable parallelization of some operations in" |
| 192 | 199 |
" order to speed burnin or to test granular locking") |
| 200 |
parser.add_option("--net-timeout", default=15, type="int",
|
|
| 201 |
dest="net_timeout", |
|
| 202 |
help="The instance check network timeout in seconds" |
|
| 203 |
" (defaults to 15 seconds)") |
|
| 204 |
parser.add_option("-C", "--http-check", default=False, action="store_true",
|
|
| 205 |
dest="http_check", |
|
| 206 |
help="Enable checking of instance status via http," |
|
| 207 |
" looking for /hostname.txt that should contain the" |
|
| 208 |
" name of the instance") |
|
| 209 |
|
|
| 193 | 210 |
|
| 194 | 211 |
options, args = parser.parse_args() |
| 195 | 212 |
if len(args) < 1 or options.os is None: |
| ... | ... | |
| 205 | 222 |
|
| 206 | 223 |
if options.disk_template == constants.DT_DISKLESS: |
| 207 | 224 |
disk_size = disk_growth = [] |
| 208 |
opts.do_addremove_disks = False |
|
| 225 |
options.do_addremove_disks = False
|
|
| 209 | 226 |
else: |
| 210 | 227 |
disk_size = [utils.ParseUnit(v) for v in options.disk_size.split(",")]
|
| 211 | 228 |
disk_growth = [utils.ParseUnit(v) |
| ... | ... | |
| 234 | 251 |
} |
| 235 | 252 |
self.hvp = {}
|
| 236 | 253 |
|
| 254 |
socket.setdefaulttimeout(options.net_timeout) |
|
| 255 |
|
|
| 237 | 256 |
def GetState(self): |
| 238 | 257 |
"""Read the cluster state from the config.""" |
| 239 | 258 |
if self.opts.nodes: |
| ... | ... | |
| 314 | 333 |
if self.opts.parallel: |
| 315 | 334 |
self.ExecJobSet(jobset) |
| 316 | 335 |
|
| 336 |
for instance in self.instances: |
|
| 337 |
self._CheckInstanceAlive(instance) |
|
| 338 |
|
|
| 317 | 339 |
def GrowDisks(self): |
| 318 | 340 |
"""Grow both the os and the swap disks by the requested amount, if any.""" |
| 319 | 341 |
for instance in self.instances: |
| ... | ... | |
| 360 | 382 |
|
| 361 | 383 |
Log("- Failover instance %s" % (instance))
|
| 362 | 384 |
self.ExecOp(op) |
| 385 |
for instance in self.instances: |
|
| 386 |
self._CheckInstanceAlive(instance) |
|
| 363 | 387 |
|
| 364 | 388 |
def ImportExport(self): |
| 365 | 389 |
"""Export the instance, delete it, and import it back. |
| ... | ... | |
| 429 | 453 |
|
| 430 | 454 |
self.to_rem.append(instance) |
| 431 | 455 |
|
| 456 |
for instance in self.instances: |
|
| 457 |
self._CheckInstanceAlive(instance) |
|
| 458 |
|
|
| 432 | 459 |
def StopInstance(self, instance): |
| 433 | 460 |
"""Stop given instance.""" |
| 434 | 461 |
op = opcodes.OpShutdownInstance(instance_name=instance) |
| ... | ... | |
| 454 | 481 |
self.StopInstance(instance) |
| 455 | 482 |
self.StartInstance(instance) |
| 456 | 483 |
|
| 484 |
for instance in self.instances: |
|
| 485 |
self._CheckInstanceAlive(instance) |
|
| 486 |
|
|
| 457 | 487 |
def Remove(self): |
| 458 | 488 |
"""Remove the instances.""" |
| 459 | 489 |
for instance in self.to_rem: |
| ... | ... | |
| 469 | 499 |
self.StopInstance(instance) |
| 470 | 500 |
self.RenameInstance(instance, rename) |
| 471 | 501 |
self.StartInstance(rename) |
| 502 |
self._CheckInstanceAlive(rename) |
|
| 472 | 503 |
self.StopInstance(rename) |
| 473 | 504 |
self.RenameInstance(rename, instance) |
| 474 | 505 |
self.StartInstance(instance) |
| 475 | 506 |
|
| 507 |
for instance in self.instances: |
|
| 508 |
self._CheckInstanceAlive(instance) |
|
| 509 |
|
|
| 476 | 510 |
def Reinstall(self): |
| 477 | 511 |
"""Reinstall the instances.""" |
| 478 | 512 |
for instance in self.instances: |
| ... | ... | |
| 485 | 519 |
Log("- Reinstall instance %s specifying the OS" % (instance,))
|
| 486 | 520 |
self.ExecOp(op) |
| 487 | 521 |
self.StartInstance(instance) |
| 522 |
for instance in self.instances: |
|
| 523 |
self._CheckInstanceAlive(instance) |
|
| 488 | 524 |
|
| 489 | 525 |
def Reboot(self): |
| 490 | 526 |
"""Reinstall the instances.""" |
| ... | ... | |
| 495 | 531 |
ignore_secondaries=False) |
| 496 | 532 |
Log("- Reboot instance %s with type '%s'" % (instance, reboot_type))
|
| 497 | 533 |
self.ExecOp(op) |
| 534 |
self._CheckInstanceAlive(instance) |
|
| 498 | 535 |
|
| 499 | 536 |
def ActivateDisks(self): |
| 500 | 537 |
"""Activate and deactivate disks of the instances.""" |
| ... | ... | |
| 509 | 546 |
Log("- Deactivate disks of offline instance %s" % (instance,))
|
| 510 | 547 |
self.ExecOp(op_deact) |
| 511 | 548 |
self.StartInstance(instance) |
| 549 |
for instance in self.instances: |
|
| 550 |
self._CheckInstanceAlive(instance) |
|
| 512 | 551 |
|
| 513 | 552 |
def AddRemoveDisks(self): |
| 514 | 553 |
"""Add and remove an extra disk for the instances.""" |
| ... | ... | |
| 524 | 563 |
Log("- Removing the last disk of instance %s" % (instance,))
|
| 525 | 564 |
self.ExecOp(op_rem) |
| 526 | 565 |
self.StartInstance(instance) |
| 566 |
for instance in self.instances: |
|
| 567 |
self._CheckInstanceAlive(instance) |
|
| 527 | 568 |
|
| 528 | 569 |
def AddRemoveNICs(self): |
| 529 | 570 |
"""Add and remove an extra NIC for the instances.""" |
| ... | ... | |
| 537 | 578 |
Log("- Removing the last NIC of instance %s" % (instance,))
|
| 538 | 579 |
self.ExecOp(op_rem) |
| 539 | 580 |
|
| 581 |
def _CheckInstanceAlive(self, instance): |
|
| 582 |
"""Check if an instance is alive by doing http checks. |
|
| 583 |
|
|
| 584 |
This will try to retrieve the url on the instance /hostname.txt |
|
| 585 |
and check that it contains the hostname of the instance. In case |
|
| 586 |
we get ECONNREFUSED, we retry up to the net timeout seconds, for |
|
| 587 |
any other error we abort. |
|
| 588 |
|
|
| 589 |
""" |
|
| 590 |
if not self.opts.http_check: |
|
| 591 |
return |
|
| 592 |
try: |
|
| 593 |
for retries in range(self.opts.net_timeout): |
|
| 594 |
try: |
|
| 595 |
url = urllib2.urlopen("http://%s/hostname.txt" % instance)
|
|
| 596 |
except urllib2.URLError, err: |
|
| 597 |
if err.args[0][0] == errno.ECONNREFUSED: |
|
| 598 |
time.sleep(1) |
|
| 599 |
continue |
|
| 600 |
raise |
|
| 601 |
except urllib2.URLError, err: |
|
| 602 |
raise InstanceDown(instance, str(err)) |
|
| 603 |
hostname = url.read().strip() |
|
| 604 |
if hostname != instance: |
|
| 605 |
raise InstanceDown(instance, ("Hostname mismatch, expected %s, got %s" %
|
|
| 606 |
(instance, hostname))) |
|
| 607 |
|
|
| 540 | 608 |
def BurninCluster(self): |
| 541 | 609 |
"""Test a cluster intensively. |
| 542 | 610 |
|
Also available in: Unified diff