def main(argv=None):
if argv is None:
argv = sys.argv
parser = argparse.ArgumentParser(
description='Fixes issues with orphaned resource providers.')
mysqlargs = MySqlArgs({
'user': '******',
'password': '',
'host': 'localhost',
'port': 3306,
})
mysqlargs.inject(parser)
parser.add_argument(
'-q',
'--quiet',
action='store_true',
help='Quiet mode. No output if there was nothing to do.')
parser.add_argument(
'--slack',
type=str,
help=
'JSON file with Slack webhook information to send a notification to')
parser.add_argument('action',
choices=['info', 'update'],
help='Just display info or actually update them?')
args = parser.parse_args(argv[1:])
mysqlargs.extract(args)
if args.slack:
slack = Slackbot(args.slack, script_name='orphan-resource-providers')
else:
slack = None
db = mysqlargs.connect()
kwargs = {
'db': db,
'describe': args.action == 'info',
'quiet': args.quiet,
}
if slack:
with slack: # log exceptions
update_count = resource_providers_fixer(**kwargs)
else:
update_count = resource_providers_fixer(**kwargs)
if slack and (args.action == 'update') and ((not args.quiet)
or update_count):
if update_count > 0:
message = ('Commanded update of *{} resource providers*'.format(
update_count))
color = '#000000'
else:
message = ('No resource providers to update')
color = '#cccccc'
slack.post('orphan-resource-providers', message, color=color)
def main(argv=None):
if argv is None:
argv = sys.argv
parser = base_parser(
'Strange things, as per someone\'s definition of "strange".')
parser.add_argument('-v', '--verbose', action='store_true')
args = parser.parse_args(argv[1:])
if args.slack:
slack = Slackbot(args.slack)
else:
slack = None
os_vars = {
k: os.environ[k]
for k in os.environ if k.startswith(OS_ENV_PREFIX)
}
if args.osrc:
os_vars.update(osapi.load_osrc(args.osrc))
missing_os_vars = set(osapi.Auth.required_os_vars) - set(os_vars)
if missing_os_vars:
print('Missing required OS values in env/rcfile: {}'.format(
', '.join(missing_os_vars)),
file=sys.stderr)
return -1
auth = osapi.Auth(os_vars)
nodes = osrest.ironic_nodes(auth, details=True)
# hypervisors = osrest.nova_hypervisors(auth, details=True)
errored_nodes = [
n for n in nodes.values()
if n['provision_state'] == 'error' and not n['maintenance']
]
if not errored_nodes:
if args.verbose:
print('All good.')
return
message = ['Ironic nodes in "error" provision state, not in maintenance']
message.extend(
'• `{}`, last error: {}'.format(n['uuid'], n.get('last_error'))
for n in errored_nodes)
message = '\n'.join(message)
print(message.replace('•', '*'))
if slack:
slack.post(SUBCOMMAND, message, color='xkcd:red')
def main(argv=None):
script = 'clean-old-aggregates'
slack = Slackbot(args.slack, script_name=script) if args.slack else None
try:
term_leases = [
lease for lease in leases.values() if is_terminated(lease)
]
old_aggregates = [
aggs for aggs in (aggregates_for_lease(lease)
for lease in term_leases) if aggs != None
]
aggregate_list = list(itertools.chain(*old_aggregates))
errors, reports = clear_aggregates(aggregate_list)
orphan_list = orphan_find(aggregates)
for orphan in orphan_list:
destiny = has_active_allocation(orphan)
host = osrest.blazar.host(auth, orphan)
if destiny is None:
reports.append(
"Error identifying allocation for orphan host {}.".format(
orphan))
elif destiny is False:
reports.append(
"Returning orphan host {} to freepool.".format(orphan) +
"\n")
osrest.nova.aggregate_add_host(auth, 1,
host['hypervisor_hostname'])
else:
destination_agg = [
aggr['id'] for aggr in aggregates.values()
if aggr['name'] == destiny
][0]
reports.append(
"Moving orphan host {} to destined aggregate {}.".format(
orphan, destination_agg))
osrest.nova.aggregate_add_host(auth, destination_agg,
host['hypervisor_hostname'])
old_allocations = query.blazar_find_old_host_alloc(conn)
for alloc in old_allocations:
hostname, lease_id = del_expired_alloc(conn, alloc)
reports.append(
"Deleted host_allocation for host {} matching expired lease {}."
.format(hostname, lease_id))
conn.db.commit()
if reports:
str_report = '\n'.join(reports)
if slack:
if errors:
slack.error(str_report)
else:
slack.message(str_report)
except:
if slack:
slack.exception()
raise
def main(argv=None):
if argv is None:
argv = sys.argv
parser = argparse.ArgumentParser(description='Remove orphan ports in '
'Neutron referring to an inactive Ironic instance')
parser.add_argument('mode', choices=['info', 'delete'],
help='Just display data on the conflict ports or delete them')
parser.add_argument('--ignore-subnet', type=str,
help='Ignore Neutron ports in this subnet (UUID). Must provide either '
'this or --ignore-from-ironic-conf. This overrides the conf.')
parser.add_argument('-c', '--ignore-from-ironic-conf', type=str,
help='Ignore Neutron ports in the subnet(s) under the '
'"provisioning_network" network in the "neutron" section of '
'this configuration file.')
parser.add_argument('--slack', type=str,
help='JSON file with Slack webhook information to send a notification to')
parser.add_argument('--osrc', type=str,
help='Connection parameter file. Should include password. envars used '
'if not provided by this file.')
parser.add_argument('-v', '--verbose', action='store_true')
parser.add_argument('--force-sane', action='store_true',
help='Disable sanity checking (i.e. things really are that bad)')
args = parser.parse_args(argv[1:])
## Validate args
slack = Slackbot(args.slack, SUBCOMMAND) if args.slack else None
auth = Auth.from_env_or_args(args=args)
if args.ignore_subnet:
ignore_subnets = [args.ignore_subnet]
elif args.ignore_from_ironic_conf:
ironic_config = configparser.ConfigParser()
ironic_config.read(args.ignore_from_ironic_conf)
net_id = ironic_config['neutron']['provisioning_network']
network = osrest.neutron.network(auth, net_id)
ignore_subnets = network['subnets']
else:
print('Must provide --ignore-subnet or --ignore-from-ironic-conf',
file=sys.stderr)
return -1
# Do actual work
with slack or nullcontext():
conflict_macs = find_conflicts(auth, ignore_subnets, slack=slack)
if args.mode == 'info':
show_info(conflict_macs, slack=slack)
elif args.mode == 'delete':
delete(auth, conflict_macs,
safe=not args.force_sane, slack=slack, verbose=args.verbose)
else:
print('unknown command', file=sys.stderr)
return -1
def main(argv=None):
if argv is None:
argv = sys.argv
parser = argparse.ArgumentParser(
description='Floating IP and port reclaimer.')
mysqlargs = MySqlArgs({
'user': '******',
'password': '',
'host': 'localhost',
'port': 3306,
})
mysqlargs.inject(parser)
osapi.add_arguments(parser)
parser.add_argument(
'-q',
'--quiet',
action='store_true',
help='Quiet mode. No output to Slack if there was nothing to do.')
parser.add_argument(
'--slack',
type=str,
help=
'JSON file with Slack webhook information to send a notification to')
parser.add_argument('action',
choices=['info', 'clean'],
help='Just display info or actually fix them?')
args = parser.parse_args(argv[1:])
mysqlargs.extract(args)
auth = osapi.Auth.from_env_or_args(args=args)
if args.slack:
slack = Slackbot(args.slack, script_name='dirty-ports')
else:
slack = None
db = mysqlargs.connect()
kwargs = {
'db': db,
'auth': auth,
'take_action': args.action == 'clean',
'quiet': args.quiet,
'slack': slack,
}
if slack:
with slack: # log exceptions
cleaner(**kwargs)
else:
cleaner(**kwargs)
def main(argv=None):
if argv is None:
argv = sys.argv
parser = base_parser('Fixes issues with orphaned resource providers.')
mysqlargs = MySqlArgs({
'user': '******',
'password': '',
'host': 'localhost',
'port': 3306,
})
mysqlargs.inject(parser)
parser.add_argument(
'-q',
'--quiet',
action='store_true',
help='Quiet mode. No output if there was nothing to do.')
parser.add_argument('action',
choices=['info', 'update'],
help='Just display info or actually update them?')
args = parser.parse_args(argv[1:])
mysqlargs.extract(args)
slack = Slackbot(
args.slack,
script_name='orphan-resource-providers') if args.slack else None
try:
db = mysqlargs.connect()
update_count = resource_providers_fixer(db=db,
describe=args.action == 'info',
quiet=args.quiet)
if args.action == 'update':
if update_count > 0:
message = ('Commanded update of *{} resource providers*'.
format(update_count))
print(message)
slack.message(message)
elif not args.quiet:
print('No resource providers to delete')
except:
if slack:
slack.exception()
raise
def main(argv=None):
if argv is None:
argv = sys.argv
parser = base_parser('Retired node state enforcer.')
mysqlargs = MySqlArgs({
'user': '******',
'password': '',
'host': 'localhost',
'port': 3306,
})
mysqlargs.inject(parser)
parser.add_argument('--dryrun', help='dryrun mode', action='store_true')
args = parser.parse_args(argv[1:])
mysqlargs.extract(args)
conn = mysqlargs.connect()
slack = Slackbot(args.slack,
script_name='enforce-retirement') if args.slack else None
# Find retired nodes and ensure they are non reservable in blazar
correct_state(conn, slack, dryrun=args.dryrun)
def main(argv=None):
if argv is None:
argv = sys.argv
parser = argparse.ArgumentParser(description='Kick Ironic nodes that '
'are in an common/known error state')
parser.add_argument('mode', choices=['info', 'reset'],
help='Just display data on the stuck nodes or reset their states')
parser.add_argument('--slack', type=str,
help='JSON file with Slack webhook information to send a notification to')
parser.add_argument('--osrc', type=str,
help='Connection parameter file. Should include password. envars used '
'if not provided by this file.')
parser.add_argument('-v', '--verbose', action='store_true')
parser.add_argument('--dry-run', action='store_true',
help='Dry run, don\'t actually do anything')
args = parser.parse_args(argv[1:])
if args.slack:
slack = Slackbot(args.slack)
else:
slack = None
os_vars = {k: os.environ[k] for k in os.environ if k.startswith(OS_ENV_PREFIX)}
if args.osrc:
os_vars.update(load_osrc(args.osrc))
missing_os_vars = set(Auth.required_os_vars) - set(os_vars)
if missing_os_vars:
print(
'Missing required OS values in env/rcfile: {}'
.format(', '.join(missing_os_vars)),
file=sys.stderr
)
return -1
auth = Auth(os_vars)
nodes = osrest.ironic_nodes(auth, details=True)
cureable = cureable_nodes(nodes)
if args.mode == 'info':
print('{} node(s) in a state that we can treat'.format(len(cureable)))
for nid in cureable:
print('-' * 40)
print('\n'.join(
'{:<25s} {}'.format(key, nodes[nid].get(key))
for key
in [
'uuid',
'provision_updated_at',
'provision_state',
'last_error',
'instance_uuid',
'extra',
'maintenance',
]
))
if slack:
if cureable:
message = ('{} nodes in correctable error states (no action '
'taken)'.format(len(cureable)))
color = 'xkcd:orange red'
else:
error_nodes = sum(1
for (nid, n)
in nodes.items()
if (
not n['maintenance'] and
n['provision_state'] == 'error'
)
)
if error_nodes:
message = ('No nodes in correctable error states ({} other'
' nodes in error state)').format(error_nodes)
color = 'xkcd:yellow'
else:
message = 'No nodes in correctable error states'
color = 'xkcd:green'
slack.post(SUBCOMMAND, message, color=color)
return
if len(cureable) == 0:
if args.verbose:
message = 'Nothing to do.'
print(message)
if slack:
slack.post(SUBCOMMAND, message, color='xkcd:light grey')
return
print('To correct: {}'.format(repr(cureable)))
if slack:
message = ['Ironic nodes in correctable error states']
for nid in cureable:
message.append(' • `{}`: "{}"'.format(nid, nodes[nid]['last_error']))
message = '\n'.join(message)
slack.post(SUBCOMMAND, message, color='xkcd:darkish red')
reset_ok = []
too_many = []
for nid in cureable:
resetter = NodeResetter(auth, nid, dry_run=args.dry_run)
try:
resetter.reset()
except TooManyResets as e:
too_many.append(nid)
except Exception as e:
if slack:
error = '{}; check logs'.format(str(e))
slack.post(SUBCOMMAND, error, color='xkcd:red')
raise
else:
reset_ok.append((nid, resetter.tracker.count()))
print('Attempted to fix: {}'.format(repr(reset_ok)))
print('Refused to fix: {}'.format(repr(too_many)))
if slack:
message = []
if reset_ok:
message.append('Attempted reset of nodes')
message.extend(' • `{}`: {} resets'.format(*r) for r in reset_ok)
if too_many:
message.append('\nAbstained (already at limit)')
message.extend(' • `{}`'.format(*r) for r in reset_ok)
color = 'xkcd:chartreuse'
if args.dry_run:
color = 'xkcd:yellow'
message.append('dry run, no changes actually made.')
if too_many:
color = 'xkcd:orange'
message = '\n'.join(message)
slack.post(SUBCOMMAND, message, color=color)
def main(argv=None):
if argv is None:
argv = sys.argv
parser = base_parser('Remove orphan ports in Neutron referring to an '
'inactive Ironic instance')
parser.add_argument('mode', choices=['info', 'delete'],
help='Just display data on the conflict ports or delete them')
parser.add_argument('--ignore-subnet', type=str,
help='Ignore Neutron ports in this subnet (UUID). Must provide either '
'this or --ignore-from-ironic-conf. This overrides the conf.')
parser.add_argument('-c', '--ignore-from-ironic-conf', type=str,
help='Ignore Neutron ports in the subnet(s) under the '
'"provisioning_network" network in the "neutron" section of '
'this configuration file.')
parser.add_argument('-v', '--verbose', action='store_true')
parser.add_argument('--force-sane', action='store_true',
help='Disable sanity checking (i.e. things really are that bad)')
args = parser.parse_args(argv[1:])
# Validate args
slack = Slackbot(args.slack, SUBCOMMAND) if args.slack else None
auth = Auth.from_env_or_args(args=args)
if args.ignore_subnet:
ignore_subnets = [args.ignore_subnet]
elif args.ignore_from_ironic_conf:
ironic_config = configparser.ConfigParser()
ironic_config.read(args.ignore_from_ironic_conf)
net_id = ironic_config['neutron']['provisioning_network']
network = osrest.neutron.network(auth, net_id)
ignore_subnets = network['subnets']
else:
print('Must provide --ignore-subnet or --ignore-from-ironic-conf',
file=sys.stderr)
return -1
# Do actual work
try:
conflict_macs = find_conflicts(auth, ignore_subnets)
if args.mode == 'info':
show_info(conflict_macs)
elif args.mode == 'delete':
if (not args.force_sane) and len(conflict_macs) > 10:
raise RuntimeError('(in)sanity check: thinks there are {} conflicting MACs'.format(len(conflict_macs)))
for mac in conflict_macs.values():
osrest.neutron_port_delete(auth, mac['neutron_port_id'])
if slack:
message = 'Fixed Ironic/Neutron MAC conflicts\n{}'.format(
'\n'.join(
' • Neutron Port `{neutron_port_id}` → `{mac}` ← Ironic Node `{ironic_node_id}` (Port `{ironic_port}`)'
.format(**m) for m in conflict_macs.values()
)
)
slack.success(message)
else:
print('unknown command', file=sys.stderr)
return -1
except:
if slack:
slack.exception()
raise
def main(argv=None):
if argv is None:
argv = sys.argv
parser = base_parser(
'Kick Ironic nodes that are in an common/known error state')
parser.add_argument(
'mode',
choices=['info', 'reset'],
help='Just display data on the stuck nodes or reset their states')
parser.add_argument('-v', '--verbose', action='store_true')
parser.add_argument('--dry-run',
action='store_true',
help='Dry run, don\'t actually do anything')
args = parser.parse_args(argv[1:])
slack = Slackbot(
args.slack,
script_name='ironic-error-resetter') if args.slack else None
os_vars = {
k: os.environ[k]
for k in os.environ if k.startswith(OS_ENV_PREFIX)
}
if args.osrc:
os_vars.update(load_osrc(args.osrc))
missing_os_vars = set(Auth.required_os_vars) - set(os_vars)
if missing_os_vars:
print('Missing required OS values in env/rcfile: {}'.format(
', '.join(missing_os_vars)),
file=sys.stderr)
return -1
auth = Auth(os_vars)
try:
nodes = osrest.ironic_nodes(auth, details=True)
cureable = cureable_nodes(nodes)
if args.mode == 'info':
print('{} node(s) in a state that we can treat'.format(
len(cureable)))
for nid in cureable:
print('-' * 40)
print('\n'.join('{:<25s} {}'.format(key, nodes[nid].get(key))
for key in [
'uuid',
'provision_updated_at',
'provision_state',
'last_error',
'instance_uuid',
'extra',
'maintenance',
]))
return
if len(cureable) == 0:
if args.verbose:
print('Nothing to do.')
return
print('To correct: {}'.format(repr(cureable)))
reset_ok = []
too_many = []
for nid in cureable:
resetter = NodeResetter(auth, nid, dry_run=args.dry_run)
resetter.reset()
reset_ok.append((nid, resetter.tracker.count()))
message_lines = []
if reset_ok:
message_lines.append('Performed reset of nodes')
message_lines.extend(' • `{}`: {} resets'.format(*r)
for r in reset_ok)
if too_many:
message_lines.append('Skipped (already at limit)')
message_lines.extend(' • `{}`'.format(r) for r in too_many)
if args.dry_run:
message_lines.append('dry run, no changes actually made.')
message = '\n'.join(message_lines)
print(message)
if slack and (not args.dry_run):
slack.success(message)
except:
if slack:
slack.exception()
raise
def main(argv=None):
if argv is None:
argv = sys.argv
parser = base_parser('Floating IP and port reclaimer.')
mysqlargs = MySqlArgs({
'user': '******',
'password': '',
'host': 'localhost',
'port': 3306,
})
mysqlargs.inject(parser)
parser.add_argument('-w', '--whitelist', type=str,
help='File of project/tenant IDs/names to ignore, one per line. '
'Ignores case and dashes.')
parser.add_argument('action', choices=['info', 'delete'],
help='Just display info or actually delete them?')
parser.add_argument('type', choices=list(RESOURCE_QUERY),
help='Grab floating IPs or ports?')
parser.add_argument('idle_days', type=float,
help='Number of days since last active instance in project was '
'deleted to consider it idle.')
args = parser.parse_args(argv[1:])
mysqlargs.extract(args)
auth = osapi.Authv2.from_env_or_args(args=args)
slack = Slackbot(args.slack, script_name='neutron-reaper') if args.slack else None
whitelist = set()
if args.whitelist:
with open(args.whitelist) as f:
whitelist = {line for line in f}
db = mysqlargs.connect()
db.version = query.ROCKY
try:
to_delete = find_reapable_resources(db=db, auth=auth, type_=args.type, idle_days=args.idle_days, whitelist=whitelist)
thing = '{}{}'.format(
{'ip': 'floating IP', 'port': 'port'}[args.type],
('' if len(to_delete) == 1 else 's'),
)
if to_delete:
if args.action == 'delete':
for resource_id in to_delete:
RESOURCE_DELETE_COMMAND[args.type](auth, resource_id)
message = (
'Commanded deletion of *{} {}* ({:.0f} day grace-period)'
.format(len(to_delete), thing, args.idle_days)
)
print(message)
if slack:
slack.message(message)
else:
print((
'Found *{} {}* to delete ({:.0f} day grace-period):\n{}'
.format(len(to_delete), thing, args.idle_days, to_delete)
))
else:
print('No {} to delete ({:.0f} day grace-period)'.format(thing, args.idle_days))
except:
if slack:
slack.exception()
raise
def main(argv=None):
if argv is None:
argv = sys.argv
parser = base_parser('Reserve nodes for maintenance')
append_global_identity_args(parser, argv)
mysqlargs = MySqlArgs({
'user': '******',
'password': '',
'host': 'localhost',
'port': 3306,
})
mysqlargs.inject(parser)
parser.add_argument('--operator',
type=str,
required=True,
help='Chameleon account username of the operator')
parser.add_argument('--nodes',
type=str,
required=True,
help='node ids or node names; comma separated')
parser.add_argument('--reason',
type=str,
required=True,
help='maintenance reasons')
parser.add_argument('--dry-run',
action="store_true",
help='perform a trial run without making reservations')
parser.add_argument(
'--start-time',
type=valid_date,
default=None,
help=
'lease start time (YYYY-mm-DD HH:MM:SS); if not given, start at the earliest possible datetime'
)
parser.add_argument(
'--estimate-hours',
type=int,
default=168,
help=
'estimated hours required for maintenance; default is 168 hours (1 week)'
)
args = parser.parse_args(argv[1:])
slack = Slackbot(
args.slack,
script_name='maintenance-reservation') if args.slack else None
# connect to database
mysqlargs.extract(args)
db = mysqlargs.connect()
# keystone authentication
auth_args = {
'auth_url': args.os_auth_url,
'username': args.os_username,
'password': args.os_password,
'project_name': args.os_project_name,
'region_name': args.os_region_name,
'interface': 'public'
}
if args.os_user_domain_name:
auth_args['user_domain_name'] = args.os_user_domain_name
if args.os_project_domain_name:
auth_args['project_domain_name'] = args.os_project_domain_name
# get admin session for node information
admin_sess = get_session(**auth_args)
# get maint session for creating lease
auth_args['project_name'] = 'maintenance'
maint_sess = get_session(**auth_args)
try:
# get node details
nodes = get_nodes(admin_sess, args.nodes.split(','))
report_info = {}
for node in nodes:
lease_start_time = args.start_time
if not lease_start_time:
# find the earliest reservation time for the node
lease_start_time = get_node_earliest_reserve_time(
db, node.uuid, args.estimate_hours)
else:
# convert to utc
lease_start_time = lease_start_time.replace(
tzinfo=tz.tzlocal()).astimezone(tz.gettz('UTC'))
# reserve
reserve_args = {
'sess': maint_sess,
'node': node,
'start_time': lease_start_time,
'requested_hours': args.estimate_hours,
'reason': args.reason,
'operator': args.operator,
'dryrun': args.dry_run
}
start_time_str, end_time_str = reserve(**reserve_args)
report_info[node.name] = (start_time_str, end_time_str)
# summary
report_lines = [('Node {node_name} at {region} is under maintenance '
'from {start_time} to {end_time}').format(
node_name=key,
region=args.os_region_name,
start_time=value[0],
end_time=value[1])
for key, value in report_info.items()]
if report_lines:
report = '\n'.join(report_lines)
print(report)
if slack:
slack.message(report)
else:
print('nothing reserved!')
except:
if slack:
slack.exception()
raise
def main(argv=None):
if argv is None:
argv = sys.argv
parser = argparse.ArgumentParser(description='Kick Ironic nodes that '
'refer to a deleted/nonexistant Nova instance')
parser.add_argument('mode', choices=['info', 'delete'],
help='Just display data on the bound nodes or delete them')
parser.add_argument('--slack', type=str,
help='JSON file with Slack webhook information to send a notification to')
parser.add_argument('--osrc', type=str,
help='Connection parameter file. Should include password. envars used '
'if not provided by this file.')
parser.add_argument('-v', '--verbose', action='store_true')
parser.add_argument('--force-sane', action='store_true',
help='Disable sanity checking (i.e. things really are that bad)')
parser.add_argument('--force-insane', action='store_true',
help=argparse.SUPPRESS) # for testing
args = parser.parse_args(argv[1:])
if args.slack:
slack = Slackbot(args.slack)
else:
slack = None
os_vars = {k: os.environ[k] for k in os.environ if k.startswith(OS_ENV_PREFIX)}
if args.osrc:
os_vars.update(load_osrc(args.osrc))
missing_os_vars = set(Auth.required_os_vars) - set(os_vars)
if missing_os_vars:
print(
'Missing required OS values in env/rcfile: {}'
.format(', '.join(missing_os_vars)),
file=sys.stderr
)
return -1
auth = Auth(os_vars)
nodes = osrest.ironic_nodes(auth)
instances = osrest.nova_instances(auth)
node_instance_map = {
n['instance_uuid']: n
for n
in nodes.values()
if n['instance_uuid'] is not None
}
node_instance_ids = set(node_instance_map)
instance_ids = set(instances)
unbound_instances = node_instance_ids - instance_ids
if args.mode == 'info':
# no-op
if unbound_instances:
print('ZOMBIE INSTANCES ON NODES')
else:
print('No zombies currently.')
for inst_id in unbound_instances:
node = node_instance_map[inst_id]
assert inst_id not in instances, 'contradiction, this should be impossible'
print('-----')
print('Ironic Node\n'
' ID: {}'.format(node['uuid']))
print(' Instance: {}'.format(node['instance_uuid']))
print(' State: {}'.format(node['provision_state']))
if slack:
if unbound_instances:
message = ('{} nodes with dead instances (no action taken)'
.format(len(unbound_instances)))
color = 'xkcd:orange red'
else:
message = 'No nodes with dead instances.'
color = 'xkcd:green'
slack.post(SUBCOMMAND, message, color=color)
elif args.mode == 'delete':
if not args.force_sane or args.force_insane:
# sanity check(s) to avoid doing something stupid
if len(instance_ids) == 0 and len(unbound_instances) != 0:
_thats_crazy('(in)sanity check: 0 running instances(?!)', slack)
ubi_limit = 20 if not args.force_insane else -1
if len(unbound_instances) > ubi_limit:
_thats_crazy(
'(in)sanity check: it thinks there are {} unbound instances'
.format(len(unbound_instances)),
slack,
)
if slack:
if unbound_instances:
message = 'Possible Ironic nodes with nonexistant instances:\n{}'.format(
'\n'.join(
' • node `{}` → instance `{}`'.format(
node_instance_map[i]['uuid'],
node_instance_map[i]['instance_uuid'])
for i in unbound_instances
)
)
color = 'xkcd:darkish red'
elif args.verbose:
message = 'No Ironic nodes visibly clinging to dead instances'
color = 'xkcd:light grey'
else:
message = None
if message:
slack.post(SUBCOMMAND, message, color=color)
try:
for inst_id in unbound_instances:
node = node_instance_map[inst_id]
node_id = node['uuid']
if node['provision_state'] == 'available':
clear_node_instance_data(auth, node_id)
else:
osrest.ironic_node_set_state(auth, node_id, 'deleted')
except Exception as e:
if slack:
error = '{} while trying to clean instances; check logs for traceback'.format(str(e))
slack.post(SUBCOMMAND, error, color='xkcd:red')
raise
else:
if unbound_instances and slack:
ok_message = (
'Cleaned {} instance(s).'
.format(len(unbound_instances))
)
slack.post(SUBCOMMAND, ok_message, color='xkcd:chartreuse')
def main(argv=None):
if argv is None:
argv = sys.argv
parser = base_parser('Detects orphan leases and remove them.')
mysqlargs = MySqlArgs({
'user': '******',
'password': '',
'host': 'localhost',
'port': 3306,
})
mysqlargs.inject(parser)
parser.add_argument(
'-d',
'--dbversion',
type=str,
help=
'Version of the database. Schemas differ, pick the appropriate one.',
choices=[query.LIBERTY, query.ROCKY],
default=query.ROCKY)
parser.add_argument('--kvm', help='Run at KVM site', action='store_true')
osapi.add_arguments(parser)
args = parser.parse_args(argv[1:])
mysqlargs.extract(args)
kvm = args.kvm
slack = Slackbot(args.slack,
script_name='orphan-detector') if args.slack else None
try:
db = mysqlargs.connect()
db.version = args.dbversion
if kvm:
# at kvm site
os_vars = {
k: os.environ[k]
for k in os.environ if k.startswith('OS_')
}
if args.osrc:
os_vars.update(osapi.load_osrc(args.osrc))
auth = v2.Password(username=os_vars['OS_USERNAME'],
password=os_vars['OS_PASSWORD'],
tenant_name=os_vars['OS_TENANT_NAME'],
auth_url=os_vars['OS_AUTH_URL'])
sess = session.Session(auth=auth)
keystone = client.Client(session=sess)
orphan_instances = get_orphan_instances_kvm(db, keystone)
else:
orphan_instances = get_orphan_instances(db)
orphan_instances_report = generate_report(
orphan_instances, "-" * 45 + "ORPHAN INSTANCES" + "-" * 45)
if orphan_instances_report:
print(orphan_instances_report)
if slack:
slack.error(orphan_instances_report)
else:
print('No orphan instances detected')
# Additionally perform lease report for CHI
if not kvm:
orphan_leases_report = generate_report(
get_orphan_leases(db), "-" * 45 + "ORPHAN LEASES" + "-" * 45)
if orphan_leases_report:
print(orphan_leases_report)
if slack:
slack.error(orphan_leases_report)
else:
print('No orphan leases detected')
except:
if slack:
slack.exception()
raise
def main(argv=None):
if argv is None:
argv = sys.argv
parser = base_parser('floating IP reaper')
mysqlargs = MySqlArgs({
'user': '******',
'password': '',
'host': 'localhost',
'port': 3306,
})
mysqlargs.inject(parser)
parser.add_argument(
'-w',
'--whitelist',
type=str,
help='File of project/tenant IDs to ignore, one per line.')
parser.add_argument(
'--grace-days',
type=int,
required=True,
help='Number of days since last used to consider to be idle')
parser.add_argument('--dryrun', help='dryrun mode', action='store_true')
args = parser.parse_args(argv[1:])
mysqlargs.extract(args)
auth = osapi.Auth.from_env_or_args(args=args)
slack = Slackbot(args.slack,
script_name='floatingip-reaper') if args.slack else None
whitelist = set()
if args.whitelist:
with open(args.whitelist) as f:
whitelist = {line.rstrip('\n') for line in f}
db = mysqlargs.connect()
db.version = query.ROCKY
try:
result = reaper(db=db,
auth=auth,
grace_days=args.grace_days,
whitelist=whitelist,
dryrun=args.dryrun)
if result and not args.dryrun:
message_lines = []
for proj, ips in result.items():
message_lines.append(
'Reclaimed *{} floating ips* from project {} ({:.0f} day grace-period)'
.format(str(len(ips)), proj, args.grace_days))
message = '\n'.join(message_lines)
print(message)
if slack:
slack.message(message)
except:
if slack:
slack.exception()
raise
def main(argv=None):
if argv is None:
argv = sys.argv
parser = base_parser('Floating IP and port reclaimer.')
mysqlargs = MySqlArgs({
'user': '******',
'password': '',
'host': 'localhost',
'port': 3306,
})
mysqlargs.inject(parser)
parser.add_argument(
'-q',
'--quiet',
action='store_true',
help='Quiet mode. No output to Slack if there was nothing to do.')
parser.add_argument(
'--multiport',
action='store_true',
help='Enable if Ironic nodes may have multiple ports associated.')
parser.add_argument('action',
choices=['info', 'clean'],
help='Just display info or actually fix them?')
args = parser.parse_args(argv[1:])
mysqlargs.extract(args)
auth = osapi.Auth.from_env_or_args(args=args)
slack = Slackbot(args.slack,
script_name='dirty-ports') if args.slack else None
assert_single = False if args.multiport else True
take_action = args.action == 'clean'
db = mysqlargs.connect()
try:
bad_ports = identify_dirty_ports(auth, assert_single)
if bad_ports:
str_ports = '\n'.join(
' • port `{uuid}` on node `{node_uuid}`'.format(**p)
for p in bad_ports)
if take_action:
clean_ports(db, bad_ports)
message = "Cleaned {} ports with `internal_info` data on `available` nodes:\n{}".format(
len(bad_ports), str_ports)
print(message)
if slack:
slack.success(message)
else:
print("(read-only mode, not cleaning ports):\n{}".format(
str_ports))
except:
if slack:
slack.exception()
raise
def main(argv=None):
if argv is None:
argv = sys.argv
parser = argparse.ArgumentParser(
description='Floating IP and port reclaimer.')
mysqlargs = MySqlArgs({
'user': '******',
'password': '',
'host': 'localhost',
'port': 3306,
})
mysqlargs.inject(parser)
osapi.add_arguments(parser)
parser.add_argument(
'-w',
'--whitelist',
type=str,
help='File of project/tenant IDs/names to ignore, one per line. '
'Ignores case and dashes.')
parser.add_argument(
'-q',
'--quiet',
action='store_true',
help='Quiet mode. No output if there was nothing to do.')
parser.add_argument(
'--slack',
type=str,
help=
'JSON file with Slack webhook information to send a notification to')
parser.add_argument(
'-d',
'--dbversion',
type=str,
help=
'Version of the database. Schemas differ, pick the appropriate one.',
choices=[query.LIBERTY, query.OCATA],
default=query.LIBERTY)
parser.add_argument('--kvm', help='Run at KVM site', action='store_true')
parser.add_argument('action',
choices=['info', 'delete'],
help='Just display info or actually delete them?')
parser.add_argument('type',
choices=list(RESOURCE_QUERY),
help='Grab floating IPs or ports?')
parser.add_argument(
'idle_days',
type=float,
help='Number of days since last active instance in project was '
'deleted to consider it idle.')
args = parser.parse_args(argv[1:])
mysqlargs.extract(args)
auth = osapi.Auth.from_env_or_args(args=args)
if args.action == 'delete' and args.type == 'port' and args.dbversion == 'ocata':
print(
'Checking ports on Ocata isn\'t validated, refusing to '
'automatically delete.',
file=sys.stderr)
sys.exit(1)
if args.slack:
slack = Slackbot(args.slack, script_name='neutron-reaper')
else:
slack = None
whitelist = set()
if args.whitelist:
with open(args.whitelist) as f:
whitelist = {normalize_project_name(line, args.kvm) for line in f}
db = mysqlargs.connect()
db.version = args.dbversion
kwargs = {
'db': db,
'auth': auth,
'type_': args.type,
'idle_days': args.idle_days,
'whitelist': whitelist,
'kvm': args.kvm,
'describe': args.action == 'info',
'quiet': args.quiet,
}
if slack:
with slack: # log exceptions
remove_count = reaper(**kwargs)
else:
remove_count = reaper(**kwargs)
if slack and (args.action == 'delete') and ((not args.quiet)
or remove_count):
thing = '{}{}'.format(
{
'ip': 'floating IP',
'port': 'port'
}[args.type],
('' if remove_count == 1 else 's'),
)
if remove_count > 0:
message = (
'Commanded deletion of *{} {}* ({:.0f} day grace-period)'.
format(remove_count, thing, args.idle_days))
color = '#000000'
else:
message = ('No {} to delete ({:.0f} day grace-period)'.format(
thing, args.idle_days))
color = '#cccccc'
slack.post('neutron-reaper', message, color=color)
def main(argv=None):
if argv is None:
argv = sys.argv
parser = base_parser('Unutilized Lease Reaper')
parser.add_argument('-w',
'--warn-hours',
type=int,
help='Number of hours after which to warn user.',
default=DEFAULT_WARN_HOURS)
parser.add_argument('-r',
'--grace-hours',
type=int,
help='Number of hours after which to remove lease.',
default=DEFAULT_GRACE_HOURS)
parser.add_argument('action',
choices=['info', 'delete'],
help='Just display info or actually delete them?')
parser.add_argument('--sender',
type=str,
help='Email address of sender.',
default='*****@*****.**')
args = parser.parse_args(argv[1:])
auth = osapi.Auth.from_env_or_args(args=args)
assert args.grace_hours > args.warn_hours, (
"Grace hours must be greater than warning period.")
if args.slack:
slack = Slackbot(args.slack, script_name='unutilized-leases-reaper')
else:
slack = None
try:
sender = args.sender
warn_period = args.warn_hours
grace_period = args.grace_hours
warn, terminate = find_leases_in_violation(auth, warn_period,
grace_period)
if (len(warn) + len(terminate) > 0):
if args.action == 'delete':
for lease in warn:
if lease not in terminate:
send_notification(
auth, lease, sender, warn_period, grace_period,
"Your lease {} is idle and may be terminated.".
format(lease['name']),
_email.IDLE_LEASE_WARNING_EMAIL_BODY)
for lease in terminate:
blazar.lease_delete(auth, lease['id'])
send_notification(
auth, lease, sender, warn_period, grace_period,
"Your lease {} has been terminated.".format(
lease['name']),
_email.IDLE_LEASE_TERMINATION_EMAIL_BODY)
message = ('Warned deletion of *{} idle leases* '
'Commanded deletion of *{} idle leases* '
'(Unutilized lease violation)'.format(
len(warn), len(terminate)))
print(message)
if slack:
slack.message(message)
else:
pprint(
dict(warn=[
dict(lease_id=l['id'],
nodes=[n['uuid'] for n in l['nodes']])
for l in warn
],
terminate=[
dict(lease_id=l['id'],
nodes=[n['uuid'] for n in l['nodes']])
for l in terminate
]))
else:
print('No leases to warn or delete.')
except Exception:
if slack:
slack.exception()
raise
def main(argv=None):
if argv is None:
argv = sys.argv
parser = base_parser(
'Kick Ironic nodes that refer to a deleted/nonexistant Nova instance')
parser.add_argument(
'mode',
choices=['info', 'delete'],
help='Just display data on the bound nodes or delete them')
parser.add_argument(
'--slack',
type=str,
help=
'JSON file with Slack webhook information to send a notification to')
parser.add_argument(
'--osrc',
type=str,
help='Connection parameter file. Should include password. envars used '
'if not provided by this file.')
parser.add_argument('-v', '--verbose', action='store_true')
parser.add_argument(
'--force-sane',
action='store_true',
help='Disable sanity checking (i.e. things really are that bad)')
parser.add_argument('--force-insane',
action='store_true',
help=argparse.SUPPRESS) # for testing
args = parser.parse_args(argv[1:])
slack = Slackbot(args.slack,
script_name='undead-instances') if args.slack else None
os_vars = {
k: os.environ[k]
for k in os.environ if k.startswith(OS_ENV_PREFIX)
}
if args.osrc:
os_vars.update(load_osrc(args.osrc))
missing_os_vars = set(Auth.required_os_vars) - set(os_vars)
if missing_os_vars:
print('Missing required OS values in env/rcfile: {}'.format(
', '.join(missing_os_vars)),
file=sys.stderr)
return -1
auth = Auth(os_vars)
nodes = osrest.ironic_nodes(auth)
instances = osrest.nova_instances(auth)
node_instance_map, unbound_instances = find_unbound_instances(
auth, nodes, instances)
if args.mode == 'info':
# no-op
if unbound_instances:
print('ZOMBIE INSTANCES ON NODES')
else:
print('No zombies currently.')
for inst_id in unbound_instances:
node = node_instance_map[inst_id]
assert inst_id not in instances, 'contradiction, this should be impossible'
print('-----')
print('Ironic Node\n' ' ID: {}'.format(node['uuid']))
print(' Instance: {}'.format(node['instance_uuid']))
print(' State: {}'.format(node['provision_state']))
elif args.mode == 'delete':
if not args.force_sane or args.force_insane:
# sanity check(s) to avoid doing something stupid
if len(instance_ids) == 0 and len(unbound_instances) != 0:
_thats_crazy('(in)sanity check: 0 running instances(?!)',
slack)
ubi_limit = 20 if not args.force_insane else -1
if len(unbound_instances) > ubi_limit:
_thats_crazy(
'(in)sanity check: it thinks there are {} unbound instances'
.format(len(unbound_instances)),
slack,
)
try:
for inst_id in unbound_instances:
node = node_instance_map[inst_id]
node_id = node['uuid']
if node['provision_state'] == 'available':
clear_node_instance_data(auth, node_id)
else:
osrest.ironic_node_set_state(auth, node_id, 'deleted')
message = 'Fixed Ironic nodes with nonexistant instances:\n{}'.format(
'\n'.join(' • node `{}` → instance `{}`'.format(
node_instance_map[i]['uuid'], node_instance_map[i]
['instance_uuid']) for i in unbound_instances))
print(message)
if slack:
slack.success(message)
except:
if slack:
slack.exception()
raise
