Merge pull request #10112 from jsmartin/ec2_asg_rolling_integration_tests

Adds integration tests for new ec2_asg features

test/integration/cleanup_ec2.py

@@ -12,6 +12,7 @@ import optparse
 import yaml
 import os.path
 import boto.ec2.elb
+import time
 
 def delete_aws_resources(get_func, attr, opts):
     for item in get_func():
@@ -19,6 +20,37 @@ def delete_aws_resources(get_func, attr, opts):
         if re.search(opts.match_re, val):
             prompt_and_delete(item, "Delete matching %s? [y/n]: " % (item,), opts.assumeyes)
 
+def delete_autoscaling_group(get_func, attr, opts):
+    assumeyes = opts.assumeyes
+    group_name = None
+    for item in get_func():
+        group_name = getattr(item, attr)
+        if re.search(opts.match_re, group_name):
+            if not opts.assumeyes:
+                assumeyes = raw_input("Delete matching %s? [y/n]: " % (item).lower()) == 'y'
+            break
+    if assumeyes and group_name:
+        groups = asg.get_all_groups(names=[group_name])
+        if groups:
+            group = groups[0]
+            group.max_size = 0
+            group.min_size = 0
+            group.desired_capacity = 0
+            group.update()
+            instances = True
+            while instances:
+                tmp_groups = asg.get_all_groups(names=[group_name])
+                if tmp_groups:
+                    tmp_group = tmp_groups[0]
+                    if not tmp_group.instances:
+                       instances = False
+                time.sleep(10)
+
+            group.delete()
+            while len(asg.get_all_groups(names=[group_name])):
+                time.sleep(5)
+            print ("Terminated ASG: %s" % group_name)
+
 def delete_aws_eips(get_func, attr, opts):
 
     # the file might not be there if the integration test wasn't run
@@ -128,7 +160,7 @@ if __name__ == '__main__':
         delete_aws_resources(aws.get_all_security_groups, 'name', opts)
 
         # Delete matching ASGs
-        delete_aws_resources(asg.get_all_groups, 'name', opts)
+        delete_autoscaling_group(asg.get_all_groups, 'name', opts)
 
         # Delete matching launch configs
         delete_aws_resources(asg.get_all_launch_configurations, 'name', opts)

test/integration/roles/test_ec2_asg/tasks/main.yml

@@ -1,31 +1,41 @@
 ---
 # tasks file for test_ec2_asg
 
+# we are using a custom built AMI that runs an apache server to verify
+# ELB health checks and perform rolling ASG updates
+# this will only work on us-east-1
+
 # ============================================================
 # create and kill an ASG
-- name: lookup ami id
-  ec2_ami_search: distro=ubuntu region={{ ec2_region }} release=trusty
-  register: ubuntu_image
 - name: ensure launch config exists
   ec2_lc:
     name: "{{ resource_prefix }}-lc"
     ec2_access_key: "{{ ec2_access_key }}"
     ec2_secret_key: "{{ ec2_secret_key }}"
     region: "{{ ec2_region }}"
-    image_id: "{{ ubuntu_image.ami }}"
+    image_id: ami-964a0efe
-    instance_type: t1.micro
+    instance_type: t2.micro
-- name: launch asg
+
+- name: launch asg and wait for instances to be deemed healthy (no ELB)
   ec2_asg:
     name: "{{ resource_prefix }}-asg"
     ec2_access_key: "{{ ec2_access_key }}"
     ec2_secret_key: "{{ ec2_secret_key }}"
     launch_config_name: "{{ resource_prefix }}-lc"
+    desired_capacity: 1
     min_size: 1
     max_size: 1
     region: "{{ ec2_region }}"
     state: present
-- name: pause for a bit to make sure that the group can't be trivially deleted
+    wait_for_instances: yes
-  pause: seconds=30
+  register: output
+
+- assert:
+    that:
+      - "output.viable_instances == 1"
+
+# - name: pause for a bit to make sure that the group can't be trivially deleted
+#   pause: seconds=30
 - name: kill asg
   ec2_asg:
     name: "{{ resource_prefix }}-asg"
@@ -34,3 +44,167 @@
     region: "{{ ec2_region }}"
     state: absent
   async: 300
+
+
+- name: launch asg and do not wait for instances to be deemed healthy (no ELB)
+  ec2_asg:
+    name: "{{ resource_prefix }}-asg"
+    ec2_access_key: "{{ ec2_access_key }}"
+    ec2_secret_key: "{{ ec2_secret_key }}"
+    launch_config_name: "{{ resource_prefix }}-lc"
+    desired_capacity: 1
+    min_size: 1
+    max_size: 1
+    region: "{{ ec2_region }}"
+    wait_for_instances: no
+    state: present
+  register: output
+
+- assert:
+    that:
+      - "output.viable_instances == 0"
+
+- name: kill asg
+  ec2_asg:
+    name: "{{ resource_prefix }}-asg"
+    ec2_access_key: "{{ ec2_access_key }}"
+    ec2_secret_key: "{{ ec2_secret_key }}"
+    region: "{{ ec2_region }}"
+    state: absent
+  async: 300
+
+- name: launch load balancer
+  ec2_elb_lb:
+    name: "{{ resource_prefix }}-lb"
+    region: "{{ ec2_region }}"
+    state: present
+    ec2_access_key: "{{ ec2_access_key }}"
+    ec2_secret_key: "{{ ec2_secret_key }}"
+    zones:
+      - "{{ ec2_region }}b"
+      - "{{ ec2_region }}c"
+    connection_draining_timeout: 60
+    listeners:
+      - protocol: http
+        load_balancer_port: 80
+        instance_port: 80
+    health_check:
+        ping_protocol: http 
+        ping_port: 80
+        ping_path: "/"
+        response_timeout: 5 
+        interval: 30
+        unhealthy_threshold: 3
+        healthy_threshold: 3
+  register: load_balancer
+
+
+- name: launch asg and wait for instances to be deemed healthy (ELB)
+  ec2_asg:
+    name: "{{ resource_prefix }}-asg"
+    availability_zones:
+      - "{{ ec2_region }}b"
+      - "{{ ec2_region }}c"
+    ec2_access_key: "{{ ec2_access_key }}"
+    ec2_secret_key: "{{ ec2_secret_key }}"
+    launch_config_name: "{{ resource_prefix }}-lc"
+    health_check_type: ELB
+    desired_capacity: 1
+    min_size: 1
+    max_size: 1
+    health_check_period: 120
+    load_balancers: "{{ resource_prefix }}-lb"
+    region: "{{ ec2_region }}"
+    wait_for_instances: yes
+    wait_timeout: 600
+    state: present
+  register: output
+
+- assert:
+    that:
+      - "output.viable_instances == 1"
+
+
+# grow scaling group to 3 
+
+- name: add 2 more instances wait for instances to be deemed healthy (ELB)
+  ec2_asg:
+    name: "{{ resource_prefix }}-asg"
+    availability_zones:
+      - "{{ ec2_region }}b"
+      - "{{ ec2_region }}c"
+    ec2_access_key: "{{ ec2_access_key }}"
+    ec2_secret_key: "{{ ec2_secret_key }}"
+    launch_config_name: "{{ resource_prefix }}-lc"
+    health_check_type: ELB
+    desired_capacity: 3
+    min_size: 3
+    max_size: 5
+    health_check_period: 120
+    load_balancers: ec2-asg-int-test
+    region: "{{ ec2_region }}"
+    wait_for_instances: yes
+    wait_timeout: 600
+    state: present
+  register: output
+
+- assert:
+    that:
+      - "output.viable_instances == 3"
+
+# # create new launch config with alternate AMI
+
+- name: ensure launch config exists
+  ec2_lc:
+    name: "{{ resource_prefix }}-lc-2"
+    ec2_access_key: "{{ ec2_access_key }}"
+    ec2_secret_key: "{{ ec2_secret_key }}"
+    region: "{{ ec2_region }}"
+    image_id: ami-2a4a0e42
+    instance_type: t2.micro
+
+
+# # perform rolling replace
+
+- name: perform rolling update to new AMI
+  ec2_asg:
+    name: "{{ resource_prefix }}-asg"
+    availability_zones:
+      - "{{ ec2_region }}b"
+      - "{{ ec2_region }}c"
+    ec2_access_key: "{{ ec2_access_key }}"
+    ec2_secret_key: "{{ ec2_secret_key }}"
+    launch_config_name: "{{ resource_prefix }}-lc-2"
+    health_check_type: ELB
+    desired_capacity: 3
+    min_size: 3
+    max_size: 5
+    health_check_period: 120
+    load_balancers: ec2-asg-int-test
+    region: "{{ ec2_region }}"
+    wait_for_instances: yes
+    replace_all_instances: yes
+    wait_timeout: 600
+    state: present
+  register: output
+
+# ensure that all instances have new launch config
+- assert:
+    that:
+      - "item.value.launch_config_name == '{{ resource_prefix }}-lc-2'"
+  with_dict: output.instance_facts
+
+# assert they are all healthy
+- assert:
+    that:
+      - "output.viable_instances >= 3"
+
+
+- name: kill asg
+  ec2_asg:
+    name: "{{ resource_prefix }}-asg"
+    ec2_access_key: "{{ ec2_access_key }}"
+    ec2_secret_key: "{{ ec2_secret_key }}"
+    region: "{{ ec2_region }}"
+    state: absent
+  async: 300