def config_controller(stream, config_file=None, password='******'):
"""
Configure controller-0 using optional arguments
"""
args = ''
if config_file:
args += '--config-file ' + config_file + ' '
serial.send_bytes(stream,
"sudo config_controller {}".format(args),
expect_prompt=False)
host_helper.check_password(stream, password=password)
ret = serial.expect_bytes(stream,
"unlock controller to proceed.",
timeout=HostTimeout.LAB_CONFIG)
if ret != 0:
LOG.info("Configuration failed. Exiting installer.")
raise Exception("Configcontroller failed")
def config_controller(stream, default=True, release='R6', config_file=None, backup=None, clone_iso=None,
restore_system=None, restore_images=None, remote_host=None, password='******'):
"""
Configure controller-0 using optional arguments
Args:
stream(stream): stream to controller-0
default(bool): Use default settings
config_file(str): Config file to use.
backup(str):
clone_iso(str):
release(str): Release version
restore_system(str):
restore_images(str):
remote_host(str): Host to retrieve licence from if necessary
Steps:
- Checks for license file
- Runs config_controller with default argument or with config-file if requested
"""
# TODO:Currently only uses config_file and default as optional arguments
args = ''
if config_file:
args += '--config-file ' + config_file
if release != 'R6':
ret = serial.send_bytes(stream, "ls | grep license.lic", prompt='license.lic', fail_ok=True, timeout=10)
if ret != 0:
LOG.info("License file not found. Please retrieve license and press enter to continue.")
input()
if (release == 'R4' or release == 'R3') and not config_file:
LOG.info("Configuration fails for R4/3 when using --default. "
"Please configure manually before continuing the installation")
return 1
LOG.info("Configuring controller-0")
start = time.time()
serial.send_bytes(stream, "sudo config_controller {}".format(args), expect_prompt=False)
host_helper.check_password(stream, password=password)
ret = serial.expect_bytes(stream, "Configuration was applied", timeout=HostTimeout.LAB_CONFIG)
if ret != 0:
LOG.info("Configuration failed. Exiting installer.")
exit()
kpi.CONFIGTIME = time.time() - start
LOG.info("Configuration time: {} minutes".format(kpi.CONFIGTIME/60))
def setup_networking(stream, release, ctrlr0_ip, hostadapter_ip, password='******'):
"""
Setup initial networking so we can transfer files.
"""
ip = ctrlr0_ip
host_ip = hostadapter_ip
if release == "R2":
interface = "eth0"
else:
interface = "enp0s3"
ret = serial.send_bytes(stream, "/sbin/ip address list", prompt=ctrlr0_ip, fail_ok=True, timeout=10)
if ret != 0:
LOG.info("Setting networking up.")
else:
LOG.info("Skipping networking setup")
return
LOG.info("{} being set up with ip {}".format(interface, ip))
serial.send_bytes(stream, "sudo /sbin/ip addr add {}/24 dev {}".format(ip, interface), expect_prompt=False)
host_helper.check_password(stream, password=password)
time.sleep(2)
serial.send_bytes(stream, "sudo /sbin/ip link set {} up".format(interface), expect_prompt=False)
host_helper.check_password(stream, password=password)
time.sleep(2)
serial.send_bytes(stream, "sudo route add default gw {}".format(host_ip), expect_prompt=False)
host_helper.check_password(stream, password=password)
NETWORKING_TIME = 60
LOG.info("Wait a minute for networking to be established")
time.sleep(NETWORKING_TIME)
def run_install_scripts(stream, host_list, aio_type=None, storage=False, release='R6', socks=None, streams=None,
labname=None, username='******', password='******', conf_files={}):
"""
Runs lab install.sh iterations. Currently does not support Simplex systems
Args:
stream: Stream to controller-0
host_list: list of hosts, used when running aio scripts to install controller-1 at the appropriate time
release: Release that is installed.
aio_type: Option to run the script for aio setup
storage: Option to run the script for storage setup
streams: Dictionary of streams to nodes
Steps:
- Checks for lab_setup files
- Runs lab_setup iterations
- Unlocks nodes
"""
LOG.info("Starting to run the second round of lab_setup script. ")
if release == 'R6' or release == 'R5' or release == 'R4':
ret = serial.send_bytes(stream, '/bin/ls /home/' + username + '/images/ | grep tis-centos-guest.img',
prompt="tis-centos-guest.img", fail_ok=True,
timeout=10)
else:
ret = serial.send_bytes(stream, '/bin/ls /home/' + username + '/images/ | grep cgcs-guest.img',
prompt="cgcs-guest.img", fail_ok=True,
timeout=10)
if ret != 0:
LOG.info("Guest image not found. Please transfer the file before continuing. "
"Press enter once guest image is obtained.")
input()
time.sleep(5)
serial.send_bytes(stream, "chmod +x *.sh", timeout=20)
ret = serial.send_bytes(stream, '/bin/ls /home/' + username + '/ | grep lab_setup.sh', prompt="lab_setup.sh",
fail_ok=True, timeout=10)
if ret != 0:
LOG.info("Lab_setup.sh not found. Please transfer the "
"required files before continuing. Press enter once files are obtained.")
input()
conf_str = ""
for file in conf_files:
conf_str = conf_str + " -f {}".format(file)
LOG.info("lab_setup.sh will be run with options: {}".format(conf_str))
start = time.time()
if aio_type:
## FOR AIO
serial.send_bytes(stream, "source /etc/nova/openrc", prompt='keystone')
if release != 'R6':
serial.send_bytes(stream, "./lab_setup.sh {}".format(conf_str), expect_prompt=False, fail_ok=True)
host_helper.check_password(stream, password=password)
ret = serial.expect_bytes(stream, "topping after", timeout=HostTimeout.LAB_INSTALL, fail_ok=True)
if ret != 0:
LOG.info("Lab_setup.sh failed. Pausing to allow for debugging. "
"Please re-run the iteration before continuing. Press enter to continue.")
input()
LOG.info("Running system compute-config-complete, "
"installation will resume once controller-0 reboots and services are active")
serial.send_bytes(stream, "source /etc/nova/openrc", prompt='keystone')
serial.send_bytes(stream, "system compute-config-complete", expect_prompt=False)
serial.expect_bytes(stream, "login:"******"Installing {} ".format(ctrlr1))
host_helper.install_host(stream, ctrlr1, host_type='controller', host_id=2)
# Now wait for controller-1 to come up. Look for "login" on the serial port.
# Close the socket if we are done
try:
serial.expect_bytes(cont1_stream, "ontroller-1 login:"******"{} installation complete".format(ctrlr1))
except Exception as e:
LOG.info("Connection failed for {} with {}.".format(ctrlr1, e))
## Sometimes we get UnicodeDecodeError exception due to the output
## of installation on the serial port, so ignore the exception and try one more time.
if HostTimeout.HOST_INSTALL > (time.time()-start_time):
LOG.info("Ignore the exception and wait for {} to be installed.".format(ctrlr1))
serial.expect_bytes(cont1_stream, "ontroller-1 login:"******"./lab_setup.sh {}".format(conf_str), expect_prompt=False)
host_helper.check_password(stream, password=password)
ret = serial.expect_bytes(stream, "topping after", timeout=HostTimeout.LAB_INSTALL, fail_ok=True)
if ret != 0:
LOG.info("Lab_setup.sh failed. Pausing to allow for debugging."
" Please re-run the iteration before continuing. Press enter to continue.")
input()
## Unlock controller-1
now = time.time()
ret = host_helper.unlock_host(stream, ctrlr1)
if ret == 1:
LOG.info("Cannot unlock controller-1, pausing to allow for debugging. "
"Please unlock before continuing. Press enter to continue.")
input()
retry = 0
while retry < 5:
serial.send_bytes(cont1_stream, '\n', expect_prompt=False)
try:
ret = serial.expect_bytes(cont1_stream, "ontroller-1 login:"******"Unlocking controller-1 timed-out. pausing to allow for debugging. "
"Please unlock before continuing. Press enter to continue.")
input()
break
else:
LOG.info("Unlocking controller-1 time (mins): {}".format((time.time() - now)/60))
if (time.time() - now)/60 < 15.0:
LOG.info("login is found right after host-unlock. Wait and try again.")
now = time.time()
time.sleep(10)
retry += 1
else:
break
except Exception as e:
LOG.info("Unlocking controller-1 failed with {}".format(e))
LOG.info("Pausing to allow for debugging. "
"Please unlock before continuing. Press enter to continue.")
input()
break
serial.disconnect(socks[labname + '-' + ctrlr1])
serial.send_bytes(stream, "./lab_setup.sh {}".format(conf_str), expect_prompt=False)
host_helper.check_password(stream, password=password)
ret = serial.expect_bytes(stream, "topping after", timeout=HostTimeout.LAB_INSTALL, fail_ok=True)
if ret != 0:
LOG.info("Lab_setup.sh failed. Pausing to allow for debugging. "
"Please re-run the iteration before continuing. Press enter to continue.")
input()
LOG.info("Completed install successfully.")
else:
## FOR NON-AIO cases
serial.send_bytes(stream, "source /etc/nova/openrc", prompt='keystone')
# TODO (WEI): double check this
# Why only if not R5, we run lab_setup.sh here?
#if release != 'R5':
serial.send_bytes(stream, "./lab_setup.sh {}".format(conf_str), expect_prompt=False)
host_helper.check_password(stream, password=password)
ret = serial.expect_bytes(stream, "topping after", timeout=HostTimeout.LAB_INSTALL, fail_ok=True)
if ret != 0:
LOG.info("Lab_setup.sh failed. Pausing to allow for debugging. "
"Please re-run the iteration before continuing. Press enter to continue.")
input()
## TODO:
## 1. Unlock sometimes won't bring a node into enabled state. Need to reboot/reset it.
## 2. When unlocking a node, sometimes the node will reboot again after the "login"
## prompt is seen on the serial port, so seeing "login" doesn't mean the node becomes
## "unlocked/enabled". The workaround can be:
## 1) time.sleep(10) and repeat the wait for "login" logic again.
## 2) After seeing "login" prompt, send to controller-0
## "system host-list | grep node" to look for "enabled".
## Unlock controller-1
now = time.time()
ret = host_helper.unlock_host(stream, "controller-1")
if ret == 1:
LOG.info("Cannot unlock controller-1, pausing to allow for debugging. "
"Please unlock before continuing. Press enter to continue.")
input()
## Wait for controller-1 to come up
retry = 0
ctrlr_1 = labname + "-controller-1"
while retry < 5:
serial.send_bytes(streams[ctrlr_1], '\n', expect_prompt=False)
try:
ret = serial.expect_bytes(streams[ctrlr_1], "ontroller-1 login:"******"Unlocking controller-1 timed-out. pausing to allow for debugging. Press enter to continue.")
input()
break
else:
LOG.info("Unlock controller-1 time (mins): {}".format((time.time() - now)/60))
## Sometimes "login" is found right after "host-unlock" is issued, so we need to wait and try again.
## Also sometimes the node will go thru reboot again after "login" is detected on the serial port.
## As the node unlock time is usually longer than 15 mins, as a temperary workaround
## let's wait at least for 15 mins now.
if (time.time() - now)/60 < 15.0:
LOG.info("Found controller-1 login right away. Need to wait and try again.")
now = time.time()
time.sleep(10)
retry += 1
else:
break
except Exception as e:
LOG.info("Unlock controller-1 failed with {} pausing to allow for debugging. Press enter to continue.".format(e))
input()
break
serial.disconnect(socks[ctrlr_1])
host_list.remove(ctrlr_1)
## TODO: Well, there is a chance that controller-1 is not "unlocked/enabled" yet.
## See comment #2 above.
## If it is a storage lab
if storage:
LOG.info("Re-running lab_setup.sh to configure storage nodes.")
serial.send_bytes(stream, "./lab_setup.sh {}".format(conf_str), expect_prompt=False)
host_helper.check_password(stream, password=password)
ret = serial.expect_bytes(stream, "topping after", timeout=HostTimeout.LAB_INSTALL, fail_ok=True)
if ret != 0:
LOG.info("Lab_setup.sh failed. Pausing to allow for debugging. "
"Please re-run the iteration before continuing. Press enter to continue.")
input()
now = time.time()
for host in host_list:
host = host[len(labname)+1:]
if host.startswith('storage'):
LOG.info("Unlock {}".format(host))
ret = host_helper.unlock_host(stream, host)
if ret == 1:
LOG.info("Cannot unlock {}, pausing to allow for debugging. "
"Please unlock before continuing. Press enter to continue.".format(host))
input()
time.sleep(20)
## TODO: Need the retry strategy here too.
for host in host_list:
if 'storage' in host:
serial.send_bytes(streams[host], '\n', expect_prompt=False)
try:
ret = serial.expect_bytes(streams[host], "{} login:"******"Unlocking {} timed-out. Pause for debugging. Press enter to continue.".format(host))
input()
else:
LOG.info("Unlocking {} time (mins): {}".format(host, (time.time() - now)/60))
if (time.time() - now)/60 < 15.0:
LOG.info("Found {} login too soon. Make sure it is up. Press enter to continue.".format(host))
input()
except Exception as e:
LOG.info("Unlock {} failed with {}. Pause for debugging. Press enter to continue.".format(host, e))
input()
serial.disconnect(socks[host])
host_list.remove(host)
LOG.info("Completed storage node unlock")
LOG.info("Re-running lab_setup.sh before unlocking compute nodes.")
serial.send_bytes(stream, "./lab_setup.sh {}".format(conf_str), expect_prompt=False)
host_helper.check_password(stream, password=password)
ret = serial.expect_bytes(stream, "topping after", timeout=HostTimeout.LAB_INSTALL, fail_ok=True)
if ret != 0:
LOG.info("Lab_setup.sh failed. Pausing to allow for debugging. "
"Please re-run the iteration before continuing. Press enter to continue.")
input()
# unlock compute nodes
now = time.time()
for host in host_list:
host = host[len(labname)+1:]
ret = host_helper.unlock_host(stream, host)
if ret == 1:
LOG.info("Cannot unlock {}, pausing to allow for debugging. "
"Please unlock before continuing. Press enter to continue.".format(host))
input()
time.sleep(20)
LOG.info("Waiting for {} to unlock.".format(host_list))
## Check unlocking status
failed_nodes = []
for host in host_list:
# TODO Fix it! 'ogin:' is always found immediately after unlock
# WEI: fixed it by retrying
retry = 0
while retry < 5:
serial.send_bytes(streams[host], '\n', expect_prompt=False)
try:
ret = serial.expect_bytes(streams[host], "{} login:"******"Unlock {} timed-out.".format(host))
failed_nodes.append(host)
break
else:
LOG.info("Unlock {} time (mins): {}".format(host, (time.time() - now)/60))
if (time.time() - now)/60 < 15.0:
LOG.info("login is found right after host-unlock. Wait and try again.")
now = time.time()
time.sleep(10)
retry += 1
else:
break
except Exception as e:
LOG.info("Unlock {} failed with {}".format(host, e))
failed_nodes.append(host)
break
serial.disconnect(socks[host])
## Let's reset the VMs that failed to unlock
if failed_nodes:
vboxmanage.vboxmanage_controlvms(failed_nodes, action="reset")
time.sleep(10)
tmp_streams = {}
tmp_socks = {}
LOG.info(failed_nodes)
port = 10001
for host in failed_nodes:
tmp_sock = serial.connect('{}'.format(host), port)
tmp_stream = streamexpect.wrap(tmp_sock, echo=True, close_stream=False)
time.sleep(10)
tmp_socks[host] = tmp_sock
tmp_streams[host] = tmp_stream
port += 1
host_failed = False
for host in failed_nodes:
serial.send_bytes(tmp_streams[host], '\n', expect_prompt=False)
try:
ret = serial.expect_bytes(tmp_streams[host], "{} login:"******"{} timed-out to become unlocked/available after reset.".format(host))
host_failed = True
else:
LOG.info("{} became unlocked/available after reset. time (mins): {}".format(host, (time.time() - now)/60))
except Exception as e:
LOG.info("{} failed to become unlocked/available after reset with {}".format(host, e))
host_failed = True
serial.disconnect(tmp_socks[host])
if host_failed:
LOG.info("Not all the nodes are unlocked successfully. Pausing to allow for debugging. "
"Once they all become unlocked/enabled/available, press enter to continue.")
input()
serial.send_bytes(stream, "./lab_setup.sh {}".format(conf_str), expect_prompt=False)
host_helper.check_password(stream, password=password)
ret = serial.expect_bytes(stream, "Done", timeout=HostTimeout.LAB_INSTALL, fail_ok=True)
if ret != 0:
LOG.info("Lab_setup.sh failed. Pausing to allow for debugging. "
"Please re-run the iteration before continuing."
" Press enter to continue.")
input()
LOG.info("Completed lab install.")
kpi.LABTIME = time.time()-start
LOG.info("Lab install time: {}".format(kpi.LABTIME/60))
def lab_setup_controller_0_locked(stream, username='******', password='******', conf_files={}):
"""
Runs initial lab_setup when controller-0 is locked.
This is for R5 only.
Args:
stream: Stream to controller-0
Steps:
- Checks if controller-0 is locked
- Checks for lab_setup files
- Runs first lab_setup iteration
- Unlocks controller-0
"""
time.sleep(10)
serial.send_bytes(stream, "source /etc/nova/openrc", prompt='keystone')
serial.send_bytes(stream, "system host-list", expect_prompt=False)
try:
serial.expect_bytes(stream, "locked")
except streamexpect.ExpectTimeout:
LOG.info("Controller should be locked when configuration is completed.")
return 1
ret = serial.send_bytes(stream, '/bin/ls /home/' + username + '/images/ | grep tis-centos-guest.img',
prompt="tis-centos-guest.img", fail_ok=True, timeout=10)
if ret != 0:
LOG.info("Guest image not found. Please transfer the "
"required files before continuing. Press enter once files are obtained.")
input()
time.sleep(5)
ret = serial.send_bytes(stream, '/bin/ls /home/' + username + '/ | grep lab_setup.sh', prompt="lab_setup.sh",
fail_ok=True, timeout=10)
if ret != 0:
LOG.info("Lab_setup.sh not found. Please transfer the "
"required files before continuing. Press enter once files are obtained.")
input()
time.sleep(5)
conf_str = ""
for file in conf_files:
conf_str = conf_str + " -f {}".format(file)
serial.send_bytes(stream, "sh lab_setup.sh {}".format(conf_str), timeout=HostTimeout.LAB_INSTALL, expect_prompt=False)
host_helper.check_password(stream, password=password)
ret = serial.expect_bytes(stream, "topping after", timeout=3500, fail_ok=True)
if ret != 0:
LOG.info("Lab_setup.sh failed. Pausing to allow for debugging. "
"Please re-run the iteration before continuing. Press enter to continue.")
input()
start = time.time()
host_helper.unlock_host(stream, 'controller-0')
try:
ret = serial.expect_bytes(stream, 'login:', timeout=HostTimeout.CONTROLLER_UNLOCK)
if ret != 0:
LOG.info("Controller-0 not unlocked,Pausing to allow for debugging. "
"Please re-run the iteration before continuing. Press enter to continue.")
input()
except Exception as e:
LOG.info("Unlock controller-0 failed with {}. ".format(e))
## Sometimes we get UnicodeDecodeError exception due to the output
## on the serial port, so ignore that and try one more time.
LOG.info("Ignore the exception and wait for controller-0 to be unlocked again.")
if HostTimeout.CONTROLLER_UNLOCK > (time.time()-start):
serial.expect_bytes(stream, "login:"******"Controller-0 unlock time: {} minutes".format(end))
LOG.info("Waiting for services to activate.")
time.sleep(60)
LOG.info("Length of reboot {} minutes".format(kpi.REBOOTTIME/60))
host_helper.login(stream, username=username, password=password)
## TODO (WEI): Remove it
def enable_lvm(stream, release, password='******'):
"""
Enables LVM backend
Args:
stream: stream to controller-0.
release: Release version installed.
"""
if release != 'R6':
LOG.info("Storage backends configured in config_controller for non R6 releases.")
return
serial.send_bytes(stream, "NODE=controller-0;DEVICE=/dev/sdb;SIZE=10237")
serial.send_bytes(stream, "sudo parted -s $DEVICE mktable gpt", expect_prompt=False)
host_helper.check_password(stream, password=password)
serial.send_bytes(stream, "system host-disk-list 1 | grep /dev/sdb", prompt="10237")
serial.send_bytes(stream, "DISK=$(system host-disk-list $NODE | grep $DEVICE | awk '{print $2}')")
serial.send_bytes(stream, "system host-disk-partition-add $NODE $DISK $SIZE -t lvm_phys_vol")
serial.send_bytes(stream, "system host-lvg-add $NODE cinder-volumes")
serial.send_bytes(stream, "while true; do system host-disk-partition-list $NODE --nowrap | grep $DEVICE | "
"grep Ready; if [ $? -eq 0 ]; then break; fi; sleep 1; done")
serial.send_bytes(stream, "PARTITION=$(system host-disk-partition-list $NODE --disk $DISK --nowrap | grep "
"part1 | awk '{print $2}')")
serial.send_bytes(stream, "system host-pv-add $NODE cinder-volumes $PARTITION")
serial.send_bytes(stream, "system storage-backend-add lvm -s cinder --confirmed")
serial.send_bytes(stream, "while true; do system storage-backend-list | grep lvm | grep configured; if "
"[ $? -eq 0 ]; then break; else sleep 10; fi; done")