def check(self, task):
node = task.get_obj('Node')
conn = node.libvirt_conn()
for vm in node.vm_set.filter(state__in=['running', 'starting']):
try:
libvirt_vm = conn.lookupByName(vm.libvirt_name)
except Exception as e:
vm.set_state('stopped')
vm.save()
log(msg='Failed to find VM %s at node %s' %
(vm.id, vm.node.address),
exception=e,
tags=('agent', 'node', 'error'),
context=task.logger_ctx)
if libvirt_vm.state() == libvirt.VIR_DOMAIN_RUNNING:
vm.set_state('running')
vm.save()
else:
vm.set_state('stopped')
vm.save()
conn.close()
node.state = 'ok'
node.save()
def suspend(self, task):
"""
Suspend node to RAM for defined in config seconds. After this time + NODE_WAKEUP_TIME
node is suspended again, unles it's state is not wake up. Available only
in admin site or through plugins.
"""
node = task.get_obj('Node')
if VM.objects.filter(node=node).exclude(state='closed').count() > 0:
task.comment = "Node is in use. Aborting suspend"
task.save()
return
node.set_state('suspend')
node.save()
log(msg="suspending node %s" % node.address,
tags=('agent', 'node', 'info'),
context=task.logger_ctx)
system.call(['ping', '-c', '1', node.address])
arp = open('/proc/net/arp', 'r').readlines()
for line in arp:
fields = line.split()
if fields[0] == node.address:
node.set_prop('mac', fields[3])
node.save()
conn = node.libvirt_conn()
conn.suspendForDuration(libvirt.VIR_NODE_SUSPEND_TARGET_MEM,
config.get('core', 'NODE_SUSPEND_DURATION'))
conn.close()
def upload_url(self, task):
'''
Download datq from url and put its contents into given image. Operation.data
should contains:
- action
- url
- size
'''
image = task.get_obj('Image')
if image.attached_to != None:
raise TaskError('image_attached')
image.set_state('downloading')
image.save()
try:
volume = open(image.storage.path + '/' + image.libvirt_name, 'r+')
except Exception as e:
raise TaskFatalError('libvirt_image_not_found', exception=e)
try:
remote = urllib2.urlopen(task.get_prop('url'))
except Exception as e:
raise TaskError('url_not_found', exception=e)
bytes = 0
while bytes < int(task.get_prop('size')):
data = remote.read(1024 * 250)
if len(data) == 0:
break
volume.write(data)
bytes += len(data)
image = task.get_obj('Image')
image.set_prop('progress',
float(bytes) / float(task.get_prop('size')))
image.save()
remote.close()
volume.close()
log(msg="Rebasing image to no backend",
tags=('agent', 'image', 'info'),
context=task.logger_ctx)
if image.format in ['qcow2', 'qed']:
r = system.call([
'sudo', 'qemu-img', 'rebase', '-u', '-f', image.format, '-u',
'-b', '', image.storage.path + '/' + image.libvirt_name
],
stderr=None,
stdout=None)
if r != 0:
image.set_state('failed')
image.save()
return
image = task.get_obj('Image')
image.size = os.stat(image.storage.path + '/' +
image.libvirt_name).st_size
image.set_state('ok')
image.save()
def _sheepdog_startup(self):
for i in range(30):
r = system.call('dog node list', shell=True)
if r == 0:
break
else:
log(msg="sheepdog_startup: Restarting shepdog service", tags=('system', 'info'))
system.call('service sheepdog restart', shell=True)
time.sleep(i)
def upload_data(self, task):
'''
Put file given in operation.data['filename'] into given image (operation.image)
at offset. The file can extend existing image. Operation.data should contain:
- action
- offset
- filename
'''
image = task.get_obj('Image')
if image.attached_to != None:
raise TaskError('image_attached')
image.set_state('downloading')
image.save()
try:
volume = open(image.storage.path + '/' + image.libvirt_name, 'r+')
except Exception as e:
raise TaskFatalError('image_not_found', exception=e)
data_chunk = DataChunk(cache_key=task.get_prop('chunk_id'))
data = base64.b64decode(data_chunk.data)
volume.seek(data_chunk.offset)
volume.write(data)
volume.close()
data_chunk.delete()
log(msg="Rebasing image to no backend",
tags=('agent', 'image', 'info'),
context=task.logger_ctx)
if image.format in ['qcow2', 'qed']:
r = system.call([
'sudo', 'qemu-img', 'rebase', '-u', '-f', image.format, '-u',
'-b', '', image.storage.path + '/' + image.libvirt_name
],
stderr=None,
stdout=None)
if r != 0:
image = task.get_obj('Image')
image.set_state('failed')
image.save()
return
image = task.get_obj('Image')
image.size = os.stat(image.storage.path + '/' +
image.libvirt_name).st_size
image.set_state('ok')
image.save()
def start(self):
super(Hook, self).finish()
network = self.task.get_obj('Subnet')
try:
dnsmasq_procs = [int(x) for x in subprocess.check_output(['pgrep', 'dnsmasq']).splitlines()]
ns_procs = [int(x) for x in subprocess.check_output(['sudo', 'ip', 'netns', 'pids', network.netns_name]).splitlines()]
for pid in ns_procs:
if pid in dnsmasq_procs:
try:
#os.kill(pid, 15)
subprocess.call(['sudo', 'ip', 'netns', 'exec', network.netns_name, 'kill', '-15', str(pid)])
except:
pass
except Exception, e:
log(msg="Failed to kill DHCP process: " + str(e), tags=('agent', 'dhcp', 'network'))
def delete(self, task):
vpn = task.get_obj('VPN')
vpn.set_state('removing')
vpn.save()
try:
pid = int(open('/var/lib/cloudOver/coreVpn/%s.pid' % vpn.id, 'rb').read(1024))
system.call(['sudo', 'kill', '-15', str(pid)])
except Exception as e:
log(msg='Failed to kill openvpn process', exception=e, tags=('corevpn', 'error'))
system.call('rm -rf /var/lib/cloudOver/coreVpn/certs/%s' % vpn.id, shell=True)
try:
os.remove('/var/lib/cloudOver/coreVpn/%s.pid' % vpn.id)
except:
log(msg='Failed to remove pid file', tags=('error'))
vpn.set_state('removed')
vpn.save()
def get_user_data(request, auth_hash, auth_seed, vm_ip):
log(msg='Serving cloudinit for %s' % vm_ip, tags=('coretalk', 'debug'))
try:
lease = resolve_ip(auth_hash, auth_seed, vm_ip)
except:
return HttpResponse('none')
vm = lease.vm
if vm == None:
return HttpResponse('none')
try:
user_data_id = vm.get_prop('coretalk_user_data')
except:
return HttpResponse('none')
try:
user_data = UserData.objects.get(id=user_data_id)
return HttpResponse(user_data.data)
except:
return HttpResponse('none')
def upload_url(self, task):
'''
Download datq from url and put its contents into given image. Operation.data
should contains:
- action
- url
- size
'''
image = task.get_obj('Image')
if image.attached_to is not None:
raise TaskError('image_attached')
image.set_state('downloading')
image.save()
try:
remote = urllib2.urlopen(task.get_prop('url'))
except Exception as e:
raise TaskError('url_not_found', exception=e)
# Create temporary object
system.call('dog vdi create %s_tmp %s' %
(image.libvirt_name, task.get_prop('size')),
shell=True)
bytes = 0
while bytes < int(task.get_prop('size')):
data = remote.read(1024 * 1024)
if len(data) == 0:
log(msg='End of data',
context=task.logger_ctx,
tags=('agent', 'image', 'info'))
break
try:
p = subprocess.Popen([
'dog', 'vdi', 'write', image.libvirt_name + '_tmp',
str(bytes)
],
stdin=subprocess.PIPE)
p.stdin.write(data)
p.stdin.close()
p.wait()
except Exception as e:
log(msg='Failed to finish download at %d bytes' % bytes,
exception=e,
context=task.logger_ctx,
tags=('agent', 'image', 'error'))
break
bytes += len(data)
image.set_prop('progress',
float(bytes) / float(task.get_prop('size')))
image.save()
log(msg='Converting image %s to no-backend' % (image.libvirt_name),
context=task.logger_ctx,
tags=('agent', 'network', 'debug'))
system.call('dog vdi delete %s' % image.libvirt_name, shell=True)
r = system.call([
'qemu-img', 'convert', '-f', image.format,
'sheepdog:%s_tmp' % image.libvirt_name, '-O', 'raw',
'sheepdog:%s' % image.libvirt_name
])
if r != 0:
image.set_state('failed')
image.save()
return
system.call('dog vdi delete %s_tmp' % image.libvirt_name, shell=True)
image.size = int(task.get_prop('size'))
image.set_state('ok')
image.save()
remote.close()