def icinga_cancel_downtime(host):
icinga_host = lib.resolve_dns('icinga.wikimedia.org', 'CNAME')
cmd = 'echo "[{now}] DEL_DOWNTIME_BY_HOST_NAME;{hostname}" > {commandfile}'.format(
now=int(time.time()),
hostname=host.split('.')[0],
commandfile=get_commandfile())
return run_cumin(icinga_host, [cmd], timeout=300)
def main():
"""Downtime a single host on Icinga."""
args = parse_args()
user = lib.get_running_user()
if args.debug:
logger.setLevel(logging.DEBUG)
if args.sleep:
lib.print_line('Sleeping for {s} seconds'.format(s=args.sleep),
host=args.host)
time.sleep(args.sleep)
lib.print_line('Running Puppet on the Icinga server', host=args.host)
try:
if args.debug:
lib.run_puppet([lib.resolve_dns(lib.ICINGA_DOMAIN, 'CNAME')],
no_raise=True)
lib.icinga_downtime(args.host,
user,
args.phab_task_id,
title='wmf-downtime-host')
else:
# This is needed due to a bug in tqdm and a limitation in Cumin
with open(os.devnull, 'w', 1) as cumin_output:
stderr = sys.stderr
stdout = sys.stdout
sys.stderr = cumin_output
sys.stdout = cumin_output
lib.run_puppet([lib.resolve_dns(lib.ICINGA_DOMAIN, 'CNAME')],
no_raise=True)
lib.icinga_downtime(args.host,
user,
args.phab_task_id,
title='wmf-downtime-host')
retcode = 0
except BaseException as e:
message = 'Unable to run wmf-downtime-host'
lib.print_line('{message}: {error}'.format(message=message, error=e),
host=args.host)
logger.exception(message)
retcode = 2
finally:
if not args.debug:
sys.stderr = stderr
sys.stdout = stdout
return retcode
def run(args, user, log_path):
"""Run the WMF auto reimage according to command line arguments.
Arguments:
args -- parsed command line arguments
user -- the user that launched the script, for auditing purposes
log_path -- the path of the logfile
"""
previous = None # Previous state in conftool
rename_from = None # In case of host rename, hold the previous hostname
# Validate hosts have a signed Puppet certificate
if not args.new:
lib.validate_hosts([args.host], no_raise=args.no_verify)
# Set Icinga downtime
if not args.new and not args.no_downtime:
lib.icinga_downtime(args.host, user, args.phab_task_id)
# Depool via conftool
if args.conftool and not args.new:
previous = lib.conftool_depool(args.host, pooled=args.conftool_value)
lib.print_line('Waiting 3 minutes to let the host drain',
host=args.host)
time.sleep(180)
if args.no_pxe:
lib.print_line('Skipping PXE reboot', host=args.host)
if (not lib.validate_hosts([args.host], no_raise=True)
and lib.puppet_check_cert_to_sign(args.host) == 1):
# There is no signed or pending signing certificate for the host
lib.puppet_generate_cert(args.host)
else:
lib.puppet_remove_host(args.host) # Cleanup Puppet
# Reboot into PXE mode to start the reimage
lib.set_pxe_boot(args.host, args.mgmt)
status = lib.ipmitool_command(args.mgmt,
['chassis', 'power', 'status'])
if status.startswith('Chassis Power is off'):
lib.print_line('Current power status is off, powering on',
host=args.host)
ipmi_command = ['chassis', 'power', 'on']
else:
lib.print_line('Power cycling', host=args.host)
ipmi_command = ['chassis', 'power', 'cycle']
lib.print_line(lib.ipmitool_command(args.mgmt,
ipmi_command).rstrip('\n'),
host=args.host)
# If the host is renamed, swap the hostnames now
if args.rename is not None:
rename_from = args.host
args.host = args.rename
args.mgmt = args.rename_mgmt
# Wait that the host is booting into the installer using Cumin's direct backend
lib.wait_reboot(args.host,
start=datetime.utcnow(),
installer_key=True,
debian_installer=True)
# Wait for the reboot into the new system
time.sleep(
30
) # Avoid race conditions, the host is in the d-i, need to wait anyway
lib.wait_reboot(args.host, start=datetime.utcnow(), installer_key=True)
# Generate the Puppet certificate and signing request
if lib.detect_init_system(args.host) == 'systemd':
lib.puppet_generate_cert(args.host)
# Sign the new Puppet certificate
if lib.puppet_wait_cert_and_sign(args.host):
if args.mask: # Mask systemd services
lib.mask_systemd_services(args.host, args.mask)
lib.puppet_first_run(args.host)
# Ensure the host is in Icinga
lib.run_puppet([lib.resolve_dns(lib.ICINGA_DOMAIN, 'CNAME')],
no_raise=True)
lib.icinga_downtime(args.host, user, args.phab_task_id)
lib.check_bios_bootparams(args.host, args.mgmt)
# Issue a reboot and wait for it and also for Puppet to complete
if not args.no_reboot:
reboot_time = datetime.utcnow()
# Ensure the host is in the known hosts
lib.run_puppet([socket.getfqdn()], no_raise=True)
lib.reboot_host(args.host)
boot_time = datetime.utcnow()
lib.wait_reboot(args.host, start=reboot_time)
lib.wait_puppet_run(args.host, start=boot_time)
# Run Apache fast test
if args.apache:
lib.run_apache_fast_test(args.host)
# The unmask is *not* done automatically, the commands to unmask are printed and logged
if args.mask:
lib.print_unmask_message(args.host, args.mask)
# The repool is *not* done automatically, the command to repool is printed and logged
if args.conftool:
lib.print_repool_message(previous,
rename_from=rename_from,
rename_to=args.rename)
lib.print_line('Reimage completed', host=args.host)
def icinga_downtime(host, reason, seconds):
icinga_host = lib.resolve_dns('icinga.wikimedia.org', 'CNAME')
cmd = 'icinga-downtime -h {host} -d {seconds} -r "{reason}"'.format(
host=host.split('.')[0], seconds=seconds, reason=reason)
return run_cumin(icinga_host, [cmd], timeout=60)
