Streamlined rolling udpate algorithm. Still need to account for partials, and not waiting for instances if we're mass terminating them.

This commit is contained in:
James Martin 2015-03-27 21:41:53 -04:00
parent f3e4704d92
commit b67e1a1ce9

View file

@ -191,9 +191,13 @@ to "replace_instances":
import sys
import time
import logging as log
from ansible.module_utils.basic import *
from ansible.module_utils.ec2 import *
log.getLogger('boto').setLevel(log.CRITICAL)
#log.basicConfig(filename='/tmp/ansible_ec2_asg.log',level=log.DEBUG, format='%(asctime)s: %(message)s', datefmt='%m/%d/%Y %I:%M:%S %p')
try:
import boto.ec2.autoscale
@ -266,8 +270,71 @@ def get_properties(autoscaling_group):
if getattr(autoscaling_group, "tags", None):
properties['tags'] = dict((t.key, t.value) for t in autoscaling_group.tags)
return properties
def elb_dreg(asg_connection, module, group_name, instance_id):
region, ec2_url, aws_connect_params = get_aws_connection_info(module)
as_group = asg_connection.get_all_groups(names=[group_name])[0]
wait_timeout = module.params.get('wait_timeout')
props = get_properties(as_group)
count = 1
if as_group.load_balancers and as_group.health_check_type == 'ELB':
try:
elb_connection = connect_to_aws(boto.ec2.elb, region, **aws_connect_params)
except boto.exception.NoAuthHandlerFound, e:
module.fail_json(msg=str(e))
else:
return
exists = True
for lb in as_group.load_balancers:
elb_connection.deregister_instances(lb, instance_id)
log.debug("De-registering {0} from ELB {1}".format(instance_id, lb))
wait_timeout = time.time() + wait_timeout
while wait_timeout > time.time() and count > 0:
count = 0
for lb in as_group.load_balancers:
lb_instances = elb_connection.describe_instance_health(lb)
for i in lb_instances:
if i.instance_id == instance_id and i.state == "InService":
count += 1
log.debug("{0}: {1}, {2}".format(i.instance_id, i.state, i.description))
time.sleep(10)
if wait_timeout <= time.time():
# waiting took too long
module.fail_json(msg = "Waited too long for instance to deregister. {0}".format(time.asctime()))
def elb_healthy(asg_connection, elb_connection, module, group_name):
healthy_instances = []
as_group = asg_connection.get_all_groups(names=[group_name])[0]
props = get_properties(as_group)
# get healthy, inservice instances from ASG
instances = []
for instance, settings in props['instance_facts'].items():
if settings['lifecycle_state'] == 'InService' and settings['health_status'] == 'Healthy':
instances.append(instance)
log.debug("ASG considers the following instances InService and Healthy: {0}".format(instances))
log.debug("ELB instance status:")
for lb in as_group.load_balancers:
# we catch a race condition that sometimes happens if the instance exists in the ASG
# but has not yet show up in the ELB
try:
lb_instances = elb_connection.describe_instance_health(lb, instances=instances)
except boto.exception.InvalidInstance, e:
pass
for i in lb_instances:
if i.state == "InService":
healthy_instances.append(i.instance_id)
log.debug("{0}: {1}".format(i.instance_id, i.state))
return len(healthy_instances)
def wait_for_elb(asg_connection, module, group_name):
region, ec2_url, aws_connect_params = get_aws_connection_info(module)
@ -278,36 +345,23 @@ def wait_for_elb(asg_connection, module, group_name):
as_group = asg_connection.get_all_groups(names=[group_name])[0]
if as_group.load_balancers and as_group.health_check_type == 'ELB':
log.debug("Waiting for ELB to consider intances healthy.")
try:
elb_connection = connect_to_aws(boto.ec2.elb, region, **aws_connect_params)
except boto.exception.NoAuthHandlerFound, e:
module.fail_json(msg=str(e))
wait_timeout = time.time() + wait_timeout
healthy_instances = {}
healthy_instances = elb_healthy(asg_connection, elb_connection, module, group_name)
while len(healthy_instances.keys()) < as_group.min_size and wait_timeout > time.time():
as_group = asg_connection.get_all_groups(names=[group_name])[0]
props = get_properties(as_group)
# get healthy, inservice instances from ASG
instances = []
for instance, settings in props['instance_facts'].items():
if settings['lifecycle_state'] == 'InService' and settings['health_status'] == 'Healthy':
instances.append(instance)
for lb in as_group.load_balancers:
# we catch a race condition that sometimes happens if the instance exists in the ASG
# but has not yet show up in the ELB
try:
lb_instances = elb_connection.describe_instance_health(lb, instances=instances)
except boto.exception.InvalidInstance, e:
pass
for i in lb_instances:
if i.state == "InService":
healthy_instances[i.instance_id] = i.state
while healthy_instances < as_group.min_size and wait_timeout > time.time():
healthy_instances = elb_healthy(asg_connection, elb_connection, module, group_name)
log.debug("ELB thinks {0} instances are healthy.".format(healthy_instances))
time.sleep(10)
if wait_timeout <= time.time():
# waiting took too long
module.fail_json(msg = "Waited too long for ELB instances to be healthy. %s" % time.asctime())
log.debug("Waiting complete. ELB thinks {0} instances are healthy.".format(healthy_instances))
def create_autoscaling_group(connection, module):
group_name = module.params.get('name')
@ -365,7 +419,7 @@ def create_autoscaling_group(connection, module):
try:
connection.create_auto_scaling_group(ag)
if wait_for_instances == True:
wait_for_new_instances(module, connection, group_name, wait_timeout, desired_capacity, 'viable_instances')
wait_for_new_inst(module, connection, group_name, wait_timeout, desired_capacity, 'viable_instances')
wait_for_elb(connection, module, group_name)
as_group = connection.get_all_groups(names=[group_name])[0]
asg_properties = get_properties(as_group)
@ -431,7 +485,7 @@ def create_autoscaling_group(connection, module):
module.fail_json(msg=str(e))
if wait_for_instances == True:
wait_for_new_instances(module, connection, group_name, wait_timeout, desired_capacity, 'viable_instances')
wait_for_new_inst(module, connection, group_name, wait_timeout, desired_capacity, 'viable_instances')
wait_for_elb(connection, module, group_name)
try:
as_group = connection.get_all_groups(names=[group_name])[0]
@ -472,6 +526,15 @@ def get_chunks(l, n):
for i in xrange(0, len(l), n):
yield l[i:i+n]
def update_size(group, max_size, min_size, dc):
log.debug("setting ASG sizes")
log.debug("minimum size: {0}, desired_capacity: {1}, max size: {2}".format(min_size, dc, max_size ))
group.max_size = max_size
group.min_size = min_size
group.desired_capacity = dc
group.update()
def replace(connection, module):
batch_size = module.params.get('replace_batch_size')
wait_timeout = module.params.get('wait_timeout')
@ -479,91 +542,191 @@ def replace(connection, module):
max_size = module.params.get('max_size')
min_size = module.params.get('min_size')
desired_capacity = module.params.get('desired_capacity')
# FIXME: we need some more docs about this feature
lc_check = module.params.get('lc_check')
replace_instances = module.params.get('replace_instances')
as_group = connection.get_all_groups(names=[group_name])[0]
wait_for_new_instances(module, connection, group_name, wait_timeout, as_group.min_size, 'viable_instances')
wait_for_new_inst(module, connection, group_name, wait_timeout, as_group.min_size, 'viable_instances')
props = get_properties(as_group)
instances = props['instances']
replaceable = 0
if replace_instances:
instances = replace_instances
for k in props['instance_facts'].keys():
if k in instances:
if props['instance_facts'][k]['launch_config_name'] != props['launch_config_name']:
replaceable += 1
if replaceable == 0:
# check to see if instances are replaceable if checking launch configs
new_instances, old_instances = get_instances_by_lc(props, lc_check, instances)
num_new_inst_needed = desired_capacity - len(new_instances)
if lc_check:
if num_new_inst_needed == 0 and old_instances:
log.debug("No new instances needed, but old instances are present. Removing old instances")
terminate_batch(connection, module, old_instances, instances, True)
as_group = connection.get_all_groups(names=[group_name])[0]
props = get_properties(as_group)
changed = True
return(changed, props)
# we don't want to spin up extra instances if not necessary
if num_new_inst_needed < batch_size:
log.debug("Overriding batch size to {0}".format(num_new_inst_needed))
batch_size = num_new_inst_needed
if not old_instances:
changed = False
return(changed, props)
# set temporary settings and wait for them to be reached
# This should get overriden if the number of instances left is less than the batch size.
as_group = connection.get_all_groups(names=[group_name])[0]
as_group.max_size = max_size + batch_size
as_group.min_size = min_size + batch_size
as_group.desired_capacity = desired_capacity + batch_size
as_group.update()
wait_for_new_instances(module, connection, group_name, wait_timeout, as_group.min_size, 'viable_instances')
update_size(as_group, max_size + batch_size, min_size + batch_size, desired_capacity + batch_size)
wait_for_new_inst(module, connection, group_name, wait_timeout, as_group.min_size, 'viable_instances')
wait_for_elb(connection, module, group_name)
as_group = connection.get_all_groups(names=[group_name])[0]
props = get_properties(as_group)
instances = props['instances']
if replace_instances:
instances = replace_instances
log.debug("beginning main loop")
for i in get_chunks(instances, batch_size):
terminate_batch(connection, module, i)
wait_for_new_instances(module, connection, group_name, wait_timeout, as_group.min_size, 'viable_instances')
# break out of this loop if we have enough new instances
break_early, desired_size, term_instances = terminate_batch(connection, module, i, instances, False)
wait_for_term_inst(connection, module, term_instances)
wait_for_new_inst(module, connection, group_name, wait_timeout, desired_size, 'viable_instances')
wait_for_elb(connection, module, group_name)
as_group = connection.get_all_groups(names=[group_name])[0]
# return settings to normal
as_group.max_size = max_size
as_group.min_size = min_size
as_group.desired_capacity = desired_capacity
as_group.update()
if break_early:
log.debug("breaking loop")
break
update_size(as_group, max_size, min_size, desired_capacity)
as_group = connection.get_all_groups(names=[group_name])[0]
asg_properties = get_properties(as_group)
log.debug("Rolling update complete.")
changed=True
return(changed, asg_properties)
def terminate_batch(connection, module, replace_instances):
group_name = module.params.get('name')
wait_timeout = int(module.params.get('wait_timeout'))
lc_check = module.params.get('lc_check')
def get_instances_by_lc(props, lc_check, initial_instances):
as_group = connection.get_all_groups(names=[group_name])[0]
props = get_properties(as_group)
new_instances = []
old_instances = []
# old instances are those that have the old launch config
if lc_check:
for i in props['instances']:
if props['instance_facts'][i]['launch_config_name'] == props['launch_config_name']:
new_instances.append(i)
else:
old_instances.append(i)
else:
log.debug("Comparing initial instances with current: {0}".format(initial_instances))
for i in props['instances']:
if i not in initial_instances:
new_instances.append(i)
else:
old_instances.append(i)
log.debug("New instances: {0}, {1}".format(len(new_instances), new_instances))
log.debug("Old instances: {0}, {1}".format(len(old_instances), old_instances))
return new_instances, old_instances
def list_purgeable_instances(props, lc_check, replace_instances, initial_instances):
instances_to_terminate = []
instances = ( inst_id for inst_id in replace_instances if inst_id in props['instances'])
# check to make sure instances given are actually in the given ASG
# and they have a non-current launch config
old_instances = []
instances = ( inst_id for inst_id in replace_instances if inst_id in props['instances'])
if lc_check:
for i in instances:
if props['instance_facts'][i]['launch_config_name'] != props['launch_config_name']:
old_instances.append(i)
instances_to_terminate.append(i)
else:
old_instances = instances
for i in instances:
if i in initial_instances:
instances_to_terminate.append(i)
return instances_to_terminate
# set all instances given to unhealthy
for instance_id in old_instances:
connection.set_instance_health(instance_id,'Unhealthy')
def terminate_batch(connection, module, replace_instances, initial_instances, leftovers=False):
batch_size = module.params.get('replace_batch_size')
min_size = module.params.get('min_size')
desired_capacity = module.params.get('desired_capacity')
group_name = module.params.get('name')
wait_timeout = int(module.params.get('wait_timeout'))
lc_check = module.params.get('lc_check')
decrement_capacity = False
break_loop = False
as_group = connection.get_all_groups(names=[group_name])[0]
props = get_properties(as_group)
desired_size = as_group.min_size
new_instances, old_instances = get_instances_by_lc(props, lc_check, initial_instances)
num_new_inst_needed = desired_capacity - len(new_instances)
# check to make sure instances given are actually in the given ASG
# and they have a non-current launch config
instances_to_terminate = list_purgeable_instances(props, lc_check, replace_instances, initial_instances)
log.debug("new instances needed: {0}".format(num_new_inst_needed))
log.debug("new instances: {0}".format(new_instances))
log.debug("old instances: {0}".format(old_instances))
log.debug("batch instances: {0}".format(",".join(instances_to_terminate)))
if num_new_inst_needed == 0:
decrement_capacity = True
if as_group.min_size != min_size:
as_group.min_size = min_size
as_group.update()
log.debug("Updating minimum size back to original of {0}".format(min_size))
#if are some leftover old instances, but we are already at capacity with new ones
# we don't want to decrement capacity
if leftovers:
decrement_capacity = False
break_loop = True
instances_to_terminate = old_instances
desired_size = min_size
log.debug("No new instances needed")
if num_new_inst_needed < batch_size and num_new_inst_needed !=0 :
instances_to_terminate = instances_to_terminate[:num_new_inst_needed]
decrement_capacity = False
break_loop = False
log.debug("{0} new instances needed".format(num_new_inst_needed))
log.debug("decrementing capacity: {0}".format(decrement_capacity))
for instance_id in instances_to_terminate:
elb_dreg(connection, module, group_name, instance_id)
log.debug("terminating instance: {0}".format(instance_id))
connection.terminate_instance(instance_id, decrement_capacity=decrement_capacity)
# we wait to make sure the machines we marked as Unhealthy are
# no longer in the list
return break_loop, desired_size, instances_to_terminate
def wait_for_term_inst(connection, module, term_instances):
batch_size = module.params.get('replace_batch_size')
wait_timeout = module.params.get('wait_timeout')
group_name = module.params.get('name')
lc_check = module.params.get('lc_check')
as_group = connection.get_all_groups(names=[group_name])[0]
props = get_properties(as_group)
count = 1
wait_timeout = time.time() + wait_timeout
while wait_timeout > time.time() and count > 0:
log.debug("waiting for instances to terminate")
count = 0
as_group = connection.get_all_groups(names=[group_name])[0]
props = get_properties(as_group)
instance_facts = props['instance_facts']
instances = ( i for i in instance_facts if i in old_instances)
instances = ( i for i in instance_facts if i in term_instances)
for i in instances:
if ( instance_facts[i]['lifecycle_state'] == 'Terminating'
or instance_facts[i]['health_status'] == 'Unhealthy' ):
lifecycle = instance_facts[i]['lifecycle_state']
health = instance_facts[i]['health_status']
log.debug("Instance {0} has state of {1},{2}".format(i,lifecycle,health ))
if lifecycle == 'Terminating' or healthy == 'Unhealthy':
count += 1
time.sleep(10)
@ -571,21 +734,24 @@ def terminate_batch(connection, module, replace_instances):
# waiting took too long
module.fail_json(msg = "Waited too long for old instances to terminate. %s" % time.asctime())
def wait_for_new_instances(module, connection, group_name, wait_timeout, desired_size, prop):
def wait_for_new_inst(module, connection, group_name, wait_timeout, desired_size, prop):
# make sure we have the latest stats after that last loop.
as_group = connection.get_all_groups(names=[group_name])[0]
props = get_properties(as_group)
log.debug("Waiting for {0} = {1}, currently {2}".format(prop, desired_size, props[prop]))
# now we make sure that we have enough instances in a viable state
wait_timeout = time.time() + wait_timeout
while wait_timeout > time.time() and desired_size > props[prop]:
log.debug("Waiting for {0} = {1}, currently {2}".format(prop, desired_size, props[prop]))
time.sleep(10)
as_group = connection.get_all_groups(names=[group_name])[0]
props = get_properties(as_group)
if wait_timeout <= time.time():
# waiting took too long
module.fail_json(msg = "Waited too long for new instances to become viable. %s" % time.asctime())
log.debug("Reached {0}: {1}".format(prop, desired_size))
return props
def main():