ceph/qa/tasks/qemu.py

   1 """
   2 Qemu task
   3 """
   4 from cStringIO import StringIO
   5
   6 import contextlib
   7 import logging
   8 import os
   9 import yaml
  10
  11 from teuthology import misc as teuthology
  12 from teuthology import contextutil
  13 from tasks import rbd
  14 from teuthology.orchestra import run
  15 from teuthology.config import config as teuth_config
  16
  17 log = logging.getLogger(__name__)
  18
  19 DEFAULT_NUM_DISKS = 2
  20 DEFAULT_IMAGE_URL = 'http://download.ceph.com/qa/ubuntu-12.04.qcow2'
  21 DEFAULT_IMAGE_SIZE = 10240 # in megabytes
  22 DEFAULT_CPUS = 1
  23 DEFAULT_MEM = 4096 # in megabytes
  24
  25 def create_images(ctx, config, managers):
  26     for client, client_config in config.iteritems():
  27         disks = client_config.get('disks', DEFAULT_NUM_DISKS)
  28         if not isinstance(disks, list):
  29             disks = [{} for n in range(int(disks))]
  30         clone = client_config.get('clone', False)
  31         assert disks, 'at least one rbd device must be used'
  32         for i, disk in enumerate(disks[1:]):
  33             create_config = {
  34                 client: {
  35                     'image_name': '{client}.{num}'.format(client=client,
  36                                                           num=i + 1),
  37                     'image_format': 2 if clone else 1,
  38                     'image_size': (disk or {}).get('image_size',
  39                                                    DEFAULT_IMAGE_SIZE),
  40                     }
  41                 }
  42             managers.append(
  43                 lambda create_config=create_config:
  44                 rbd.create_image(ctx=ctx, config=create_config)
  45                 )
  46
  47 def create_clones(ctx, config, managers):
  48     for client, client_config in config.iteritems():
  49         clone = client_config.get('clone', False)
  50         if clone:
  51             num_disks = client_config.get('disks', DEFAULT_NUM_DISKS)
  52             if isinstance(num_disks, list):
  53                 num_disks = len(num_disks)
  54             for i in xrange(num_disks):
  55                 create_config = {
  56                     client: {
  57                         'image_name':
  58                         '{client}.{num}-clone'.format(client=client, num=i),
  59                         'parent_name':
  60                         '{client}.{num}'.format(client=client, num=i),
  61                         }
  62                     }
  63                 managers.append(
  64                     lambda create_config=create_config:
  65                     rbd.clone_image(ctx=ctx, config=create_config)
  66                     )
  67
  68 @contextlib.contextmanager
  69 def create_dirs(ctx, config):
  70     """
  71     Handle directory creation and cleanup
  72     """
  73     testdir = teuthology.get_testdir(ctx)
  74     for client, client_config in config.iteritems():
  75         assert 'test' in client_config, 'You must specify a test to run'
  76         (remote,) = ctx.cluster.only(client).remotes.keys()
  77         remote.run(
  78             args=[
  79                 'install', '-d', '-m0755', '--',
  80                 '{tdir}/qemu'.format(tdir=testdir),
  81                 '{tdir}/archive/qemu'.format(tdir=testdir),
  82                 ]
  83             )
  84     try:
  85         yield
  86     finally:
  87         for client, client_config in config.iteritems():
  88             assert 'test' in client_config, 'You must specify a test to run'
  89             (remote,) = ctx.cluster.only(client).remotes.keys()
  90             remote.run(
  91                 args=[
  92                     'rmdir', '{tdir}/qemu'.format(tdir=testdir), run.Raw('||'), 'true',
  93                     ]
  94                 )
  95
  96 @contextlib.contextmanager
  97 def generate_iso(ctx, config):
  98     """Execute system commands to generate iso"""
  99     log.info('generating iso...')
 100     testdir = teuthology.get_testdir(ctx)
 101
 102     # use ctx.config instead of config, because config has been
 103     # through teuthology.replace_all_with_clients()
 104     refspec = ctx.config.get('branch')
 105     if refspec is None:
 106         refspec = ctx.config.get('tag')
 107     if refspec is None:
 108         refspec = ctx.config.get('sha1')
 109     if refspec is None:
 110         refspec = 'HEAD'
 111
 112     # hack: the git_url is always ceph-ci or ceph
 113     git_url = teuth_config.get_ceph_git_url()
 114     repo_name = 'ceph.git'
 115     if git_url.count('ceph-ci'):
 116         repo_name = 'ceph-ci.git'
 117
 118     for client, client_config in config.iteritems():
 119         assert 'test' in client_config, 'You must specify a test to run'
 120         test_url = client_config['test'].format(repo=repo_name, branch=refspec)
 121         (remote,) = ctx.cluster.only(client).remotes.keys()
 122         src_dir = os.path.dirname(__file__)
 123         userdata_path = os.path.join(testdir, 'qemu', 'userdata.' + client)
 124         metadata_path = os.path.join(testdir, 'qemu', 'metadata.' + client)
 125
 126         with file(os.path.join(src_dir, 'userdata_setup.yaml'), 'rb') as f:
 127             test_setup = ''.join(f.readlines())
 128             # configuring the commands to setup the nfs mount
 129             mnt_dir = "/export/{client}".format(client=client)
 130             test_setup = test_setup.format(
 131                 mnt_dir=mnt_dir
 132             )
 133
 134         with file(os.path.join(src_dir, 'userdata_teardown.yaml'), 'rb') as f:
 135             test_teardown = ''.join(f.readlines())
 136
 137         user_data = test_setup
 138         if client_config.get('type', 'filesystem') == 'filesystem':
 139             num_disks = client_config.get('disks', DEFAULT_NUM_DISKS)
 140             if isinstance(num_disks, list):
 141                 num_disks = len(num_disks)
 142             for i in xrange(1, num_disks):
 143                 dev_letter = chr(ord('a') + i)
 144                 user_data += """
 145 - |
 146   #!/bin/bash
 147   mkdir /mnt/test_{dev_letter}
 148   mkfs -t xfs /dev/vd{dev_letter}
 149   mount -t xfs /dev/vd{dev_letter} /mnt/test_{dev_letter}
 150 """.format(dev_letter=dev_letter)
 151
 152         user_data += """
 153 - |
 154   #!/bin/bash
 155   test -d /etc/ceph || mkdir /etc/ceph
 156   cp /mnt/cdrom/ceph.* /etc/ceph/
 157 """
 158
 159         cloud_config_archive = client_config.get('cloud_config_archive', [])
 160         if cloud_config_archive:
 161           user_data += yaml.safe_dump(cloud_config_archive, default_style='|',
 162                                       default_flow_style=False)
 163
 164         # this may change later to pass the directories as args to the
 165         # script or something. xfstests needs that.
 166         user_data += """
 167 - |
 168   #!/bin/bash
 169   test -d /mnt/test_b && cd /mnt/test_b
 170   /mnt/cdrom/test.sh > /mnt/log/test.log 2>&1 && touch /mnt/log/success
 171 """ + test_teardown
 172
 173         user_data = user_data.format(
 174             ceph_branch=ctx.config.get('branch'),
 175             ceph_sha1=ctx.config.get('sha1'))
 176         teuthology.write_file(remote, userdata_path, StringIO(user_data))
 177
 178         with file(os.path.join(src_dir, 'metadata.yaml'), 'rb') as f:
 179             teuthology.write_file(remote, metadata_path, f)
 180
 181         test_file = '{tdir}/qemu/{client}.test.sh'.format(tdir=testdir, client=client)
 182
 183         log.info('fetching test %s for %s', test_url, client)
 184         remote.run(
 185             args=[
 186                 'wget', '-nv', '-O', test_file,
 187                 test_url,
 188                 run.Raw('&&'),
 189                 'chmod', '755', test_file,
 190                 ],
 191             )
 192         remote.run(
 193             args=[
 194                 'genisoimage', '-quiet', '-input-charset', 'utf-8',
 195                 '-volid', 'cidata', '-joliet', '-rock',
 196                 '-o', '{tdir}/qemu/{client}.iso'.format(tdir=testdir, client=client),
 197                 '-graft-points',
 198                 'user-data={userdata}'.format(userdata=userdata_path),
 199                 'meta-data={metadata}'.format(metadata=metadata_path),
 200                 'ceph.conf=/etc/ceph/ceph.conf',
 201                 'ceph.keyring=/etc/ceph/ceph.keyring',
 202                 'test.sh={file}'.format(file=test_file),
 203                 ],
 204             )
 205     try:
 206         yield
 207     finally:
 208         for client in config.iterkeys():
 209             (remote,) = ctx.cluster.only(client).remotes.keys()
 210             remote.run(
 211                 args=[
 212                     'rm', '-f',
 213                     '{tdir}/qemu/{client}.iso'.format(tdir=testdir, client=client),
 214                     os.path.join(testdir, 'qemu', 'userdata.' + client),
 215                     os.path.join(testdir, 'qemu', 'metadata.' + client),
 216                     '{tdir}/qemu/{client}.test.sh'.format(tdir=testdir, client=client),
 217                     ],
 218                 )
 219
 220 @contextlib.contextmanager
 221 def download_image(ctx, config):
 222     """Downland base image, remove image file when done"""
 223     log.info('downloading base image')
 224     testdir = teuthology.get_testdir(ctx)
 225     for client, client_config in config.iteritems():
 226         (remote,) = ctx.cluster.only(client).remotes.keys()
 227         base_file = '{tdir}/qemu/base.{client}.qcow2'.format(tdir=testdir, client=client)
 228         image_url = client_config.get('image_url', DEFAULT_IMAGE_URL)
 229         remote.run(
 230             args=[
 231                 'wget', '-nv', '-O', base_file, image_url,
 232                 ]
 233             )
 234
 235         disks = client_config.get('disks', None)
 236         if not isinstance(disks, list):
 237             disks = [{}]
 238         image_name = '{client}.0'.format(client=client)
 239         image_size = (disks[0] or {}).get('image_size', DEFAULT_IMAGE_SIZE)
 240         remote.run(
 241             args=[
 242                 'qemu-img', 'convert', '-f', 'qcow2', '-O', 'raw',
 243                 base_file, 'rbd:rbd/{image_name}'.format(image_name=image_name)
 244                 ]
 245             )
 246         remote.run(
 247             args=[
 248                 'rbd', 'resize',
 249                 '--size={image_size}M'.format(image_size=image_size),
 250                 image_name,
 251                 ]
 252             )
 253     try:
 254         yield
 255     finally:
 256         log.debug('cleaning up base image files')
 257         for client in config.iterkeys():
 258             base_file = '{tdir}/qemu/base.{client}.qcow2'.format(
 259                 tdir=testdir,
 260                 client=client,
 261                 )
 262             (remote,) = ctx.cluster.only(client).remotes.keys()
 263             remote.run(
 264                 args=[
 265                     'rm', '-f', base_file,
 266                     ],
 267                 )
 268
 269
 270 def _setup_nfs_mount(remote, client, mount_dir):
 271     """
 272     Sets up an nfs mount on the remote that the guest can use to
 273     store logs. This nfs mount is also used to touch a file
 274     at the end of the test to indiciate if the test was successful
 275     or not.
 276     """
 277     export_dir = "/export/{client}".format(client=client)
 278     log.info("Creating the nfs export directory...")
 279     remote.run(args=[
 280         'sudo', 'mkdir', '-p', export_dir,
 281     ])
 282     log.info("Mounting the test directory...")
 283     remote.run(args=[
 284         'sudo', 'mount', '--bind', mount_dir, export_dir,
 285     ])
 286     log.info("Adding mount to /etc/exports...")
 287     export = "{dir} *(rw,no_root_squash,no_subtree_check,insecure)".format(
 288         dir=export_dir
 289     )
 290     remote.run(args=[
 291         'sudo', 'sed', '-i', '/^\/export\//d', "/etc/exports",
 292     ])
 293     remote.run(args=[
 294         'echo', export, run.Raw("|"),
 295         'sudo', 'tee', '-a', "/etc/exports",
 296     ])
 297     log.info("Restarting NFS...")
 298     if remote.os.package_type == "deb":
 299         remote.run(args=['sudo', 'service', 'nfs-kernel-server', 'restart'])
 300     else:
 301         remote.run(args=['sudo', 'systemctl', 'restart', 'nfs'])
 302
 303
 304 def _teardown_nfs_mount(remote, client):
 305     """
 306     Tears down the nfs mount on the remote used for logging and reporting the
 307     status of the tests being ran in the guest.
 308     """
 309     log.info("Tearing down the nfs mount for {remote}".format(remote=remote))
 310     export_dir = "/export/{client}".format(client=client)
 311     log.info("Stopping NFS...")
 312     if remote.os.package_type == "deb":
 313         remote.run(args=[
 314             'sudo', 'service', 'nfs-kernel-server', 'stop'
 315         ])
 316     else:
 317         remote.run(args=[
 318             'sudo', 'systemctl', 'stop', 'nfs'
 319         ])
 320     log.info("Unmounting exported directory...")
 321     remote.run(args=[
 322         'sudo', 'umount', export_dir
 323     ])
 324     log.info("Deleting exported directory...")
 325     remote.run(args=[
 326         'sudo', 'rm', '-r', '/export'
 327     ])
 328     log.info("Deleting export from /etc/exports...")
 329     remote.run(args=[
 330         'sudo', 'sed', '-i', '$ d', '/etc/exports'
 331     ])
 332     log.info("Starting NFS...")
 333     if remote.os.package_type == "deb":
 334         remote.run(args=[
 335             'sudo', 'service', 'nfs-kernel-server', 'start'
 336         ])
 337     else:
 338         remote.run(args=[
 339             'sudo', 'systemctl', 'start', 'nfs'
 340         ])
 341
 342
 343 @contextlib.contextmanager
 344 def run_qemu(ctx, config):
 345     """Setup kvm environment and start qemu"""
 346     procs = []
 347     testdir = teuthology.get_testdir(ctx)
 348     for client, client_config in config.iteritems():
 349         (remote,) = ctx.cluster.only(client).remotes.keys()
 350         log_dir = '{tdir}/archive/qemu/{client}'.format(tdir=testdir, client=client)
 351         remote.run(
 352             args=[
 353                 'mkdir', log_dir, run.Raw('&&'),
 354                 'sudo', 'modprobe', 'kvm',
 355                 ]
 356             )
 357
 358         # make an nfs mount to use for logging and to
 359         # allow to test to tell teuthology the tests outcome
 360         _setup_nfs_mount(remote, client, log_dir)
 361
 362         # Hack to make sure /dev/kvm permissions are set correctly
 363         # See http://tracker.ceph.com/issues/17977 and
 364         # https://bugzilla.redhat.com/show_bug.cgi?id=1333159
 365         remote.run(args='sudo udevadm control --reload')
 366         remote.run(args='sudo udevadm trigger /dev/kvm')
 367         remote.run(args='ls -l /dev/kvm')
 368
 369         qemu_cmd = 'qemu-system-x86_64'
 370         if remote.os.package_type == "rpm":
 371             qemu_cmd = "/usr/libexec/qemu-kvm"
 372         args=[
 373             'adjust-ulimits',
 374             'ceph-coverage',
 375             '{tdir}/archive/coverage'.format(tdir=testdir),
 376             'daemon-helper',
 377             'term',
 378             qemu_cmd, '-enable-kvm', '-nographic', '-cpu', 'host',
 379             '-smp', str(client_config.get('cpus', DEFAULT_CPUS)),
 380             '-m', str(client_config.get('memory', DEFAULT_MEM)),
 381             # cd holding metadata for cloud-init
 382             '-cdrom', '{tdir}/qemu/{client}.iso'.format(tdir=testdir, client=client),
 383             ]
 384
 385         cachemode = 'none'
 386         ceph_config = ctx.ceph['ceph'].conf.get('global', {})
 387         ceph_config.update(ctx.ceph['ceph'].conf.get('client', {}))
 388         ceph_config.update(ctx.ceph['ceph'].conf.get(client, {}))
 389         if ceph_config.get('rbd cache', True):
 390             if ceph_config.get('rbd cache max dirty', 1) > 0:
 391                 cachemode = 'writeback'
 392             else:
 393                 cachemode = 'writethrough'
 394
 395         clone = client_config.get('clone', False)
 396         num_disks = client_config.get('disks', DEFAULT_NUM_DISKS)
 397         if isinstance(num_disks, list):
 398             num_disks = len(num_disks)
 399         for i in xrange(num_disks):
 400             suffix = '-clone' if clone else ''
 401             args.extend([
 402                 '-drive',
 403                 'file=rbd:rbd/{img}:id={id},format=raw,if=virtio,cache={cachemode}'.format(
 404                     img='{client}.{num}{suffix}'.format(client=client, num=i,
 405                                                         suffix=suffix),
 406                     id=client[len('client.'):],
 407                     cachemode=cachemode,
 408                     ),
 409                 ])
 410
 411         log.info('starting qemu...')
 412         procs.append(
 413             remote.run(
 414                 args=args,
 415                 logger=log.getChild(client),
 416                 stdin=run.PIPE,
 417                 wait=False,
 418                 )
 419             )
 420
 421     try:
 422         yield
 423     finally:
 424         log.info('waiting for qemu tests to finish...')
 425         run.wait(procs)
 426
 427         log.debug('checking that qemu tests succeeded...')
 428         for client in config.iterkeys():
 429             (remote,) = ctx.cluster.only(client).remotes.keys()
 430
 431             # ensure we have permissions to all the logs
 432             log_dir = '{tdir}/archive/qemu/{client}'.format(tdir=testdir,
 433                                                             client=client)
 434             remote.run(
 435                 args=[
 436                     'sudo', 'chmod', 'a+rw', '-R', log_dir
 437                     ]
 438                 )
 439
 440             # teardown nfs mount
 441             _teardown_nfs_mount(remote, client)
 442             # check for test status
 443             remote.run(
 444                 args=[
 445                     'test', '-f',
 446                     '{tdir}/archive/qemu/{client}/success'.format(
 447                         tdir=testdir,
 448                         client=client
 449                         ),
 450                     ],
 451                 )
 452
 453
 454 @contextlib.contextmanager
 455 def task(ctx, config):
 456     """
 457     Run a test inside of QEMU on top of rbd. Only one test
 458     is supported per client.
 459
 460     For example, you can specify which clients to run on::
 461
 462         tasks:
 463         - ceph:
 464         - qemu:
 465             client.0:
 466               test: http://download.ceph.com/qa/test.sh
 467             client.1:
 468               test: http://download.ceph.com/qa/test2.sh
 469
 470     Or use the same settings on all clients:
 471
 472         tasks:
 473         - ceph:
 474         - qemu:
 475             all:
 476               test: http://download.ceph.com/qa/test.sh
 477
 478     For tests that don't need a filesystem, set type to block::
 479
 480         tasks:
 481         - ceph:
 482         - qemu:
 483             client.0:
 484               test: http://download.ceph.com/qa/test.sh
 485               type: block
 486
 487     The test should be configured to run on /dev/vdb and later
 488     devices.
 489
 490     If you want to run a test that uses more than one rbd image,
 491     specify how many images to use::
 492
 493         tasks:
 494         - ceph:
 495         - qemu:
 496             client.0:
 497               test: http://download.ceph.com/qa/test.sh
 498               type: block
 499               disks: 2
 500
 501     - or -
 502
 503         tasks:
 504         - ceph:
 505         - qemu:
 506             client.0:
 507               test: http://ceph.com/qa/test.sh
 508               type: block
 509               disks:
 510                 - image_size: 1024
 511                 - image_size: 2048
 512
 513     You can set the amount of CPUs and memory the VM has (default is 1 CPU and
 514     4096 MB)::
 515
 516         tasks:
 517         - ceph:
 518         - qemu:
 519             client.0:
 520               test: http://download.ceph.com/qa/test.sh
 521               cpus: 4
 522               memory: 512 # megabytes
 523
 524     If you want to run a test against a cloned rbd image, set clone to true::
 525
 526         tasks:
 527         - ceph:
 528         - qemu:
 529             client.0:
 530               test: http://download.ceph.com/qa/test.sh
 531               clone: true
 532
 533     If you need to configure additional cloud-config options, set cloud_config
 534     to the required data set::
 535
 536         tasks:
 537         - ceph
 538         - qemu:
 539             client.0:
 540                 test: http://ceph.com/qa/test.sh
 541                 cloud_config_archive:
 542                     - |
 543                       #/bin/bash
 544                       touch foo1
 545                     - content: |
 546                         test data
 547                       type: text/plain
 548                       filename: /tmp/data
 549
 550     If you need to override the default cloud image, set image_url:
 551
 552         tasks:
 553         - ceph
 554         - qemu:
 555             client.0:
 556                 test: http://ceph.com/qa/test.sh
 557                 image_url: https://cloud-images.ubuntu.com/releases/16.04/release/ubuntu-16.04-server-cloudimg-amd64-disk1.img
 558     """
 559     assert isinstance(config, dict), \
 560            "task qemu only supports a dictionary for configuration"
 561
 562     config = teuthology.replace_all_with_clients(ctx.cluster, config)
 563
 564     managers = []
 565     create_images(ctx=ctx, config=config, managers=managers)
 566     managers.extend([
 567         lambda: create_dirs(ctx=ctx, config=config),
 568         lambda: generate_iso(ctx=ctx, config=config),
 569         lambda: download_image(ctx=ctx, config=config),
 570         ])
 571     create_clones(ctx=ctx, config=config, managers=managers)
 572     managers.append(
 573         lambda: run_qemu(ctx=ctx, config=config),
 574         )
 575
 576     with contextutil.nested(*managers):
 577         yield