ec2_asg: fix #28087 and #35993 (#36679)

Fixes #35993 - Changes to update_size in commit eb4cc31 made it so
the group dict passed into update_size was not modified. As a result,
the 'replace' call does not see an updated min_size like it previously
did and doesn't pause to wait for any new instances to spin up. Instead,
it moves straight into terminating old instances. Fix is to add batch_size
to min_size when calling wait_for_new_inst.

Fixes #28087 - Make replace_all_instances and replace_instances behave
exactly the same by setting replace_instances = current list of instances
when replace_all_instances used. Root cause of issue was that without lc_check
terminate_batch will terminate all instances passed to it and after updating
the asg size we were querying the asg again for the list of instances - so terminate batch
saw the list including new ones just spun up.

When creating new asg with replace_all_instances: yes and lc_check: false
the instances that are initially created are then subsequently replaced.
This change makes it so replace only occurs if the asg already existed.

Add integration tests for #28087 and #35993.

lib/ansible/modules/cloud/amazon/ec2_asg.py

@@ -1191,6 +1191,7 @@ def replace(connection):
     desired_capacity = module.params.get('desired_capacity')
     lc_check = module.params.get('lc_check')
     replace_instances = module.params.get('replace_instances')
+    replace_all_instances = module.params.get('replace_all_instances')
 
     as_group = describe_autoscaling_groups(connection, group_name)[0]
     if desired_capacity is None:
@@ -1199,6 +1200,10 @@ def replace(connection):
     wait_for_new_inst(connection, group_name, wait_timeout, as_group['MinSize'], 'viable_instances')
     props = get_properties(as_group)
     instances = props['instances']
+    if replace_all_instances:
+        # If replacing all instances, then set replace_instances to current set
+        # This allows replace_instances and replace_all_instances to behave same
+        replace_instances = instances
     if replace_instances:
         instances = replace_instances
     # check to see if instances are replaceable if checking launch configs
@@ -1235,7 +1240,7 @@ def replace(connection):
 
     as_group = describe_autoscaling_groups(connection, group_name)[0]
     update_size(connection, as_group, max_size + batch_size, min_size + batch_size, desired_capacity + batch_size)
-    wait_for_new_inst(connection, group_name, wait_timeout, as_group['MinSize'], 'viable_instances')
+    wait_for_new_inst(connection, group_name, wait_timeout, as_group['MinSize'] + batch_size, 'viable_instances')
     wait_for_elb(connection, group_name)
     wait_for_target_group(connection, group_name)
     as_group = describe_autoscaling_groups(connection, group_name)[0]
@@ -1414,6 +1419,12 @@ def wait_for_new_inst(connection, group_name, wait_timeout, desired_size, prop):
     return props
 
 
+def asg_exists(connection):
+    group_name = module.params.get('name')
+    as_group = describe_autoscaling_groups(connection, group_name)
+    return bool(len(as_group))
+
+
 def main():
     argument_spec = ec2_argument_spec()
     argument_spec.update(
@@ -1483,13 +1494,16 @@ def main():
                             endpoint=ec2_url,
                             **aws_connect_params)
     changed = create_changed = replace_changed = False
+    exists = asg_exists(connection)
 
     if state == 'present':
         create_changed, asg_properties = create_autoscaling_group(connection)
     elif state == 'absent':
         changed = delete_autoscaling_group(connection)
         module.exit_json(changed=changed)
-    if replace_all_instances or replace_instances:
+
+    # Only replace instances if asg existed at start of call
+    if exists and (replace_all_instances or replace_instances):
         replace_changed, asg_properties = replace(connection)
     if create_changed or replace_changed:
         changed = True

test/integration/targets/ec2_asg/tasks/main.yml

@@ -385,6 +385,122 @@
         that:
           - "output.viable_instances == 3"
 
+    # ============================================================
+
+    # perform rolling replace with new launch configuration and lc_check:false
+
+    # Note - this is done async so we can query asg_facts during
+    # the execution. Issues #28087 and #35993 result in correct
+    # end result, but spin up extraneous instances during execution.
+    - name: "perform rolling update to new AMI with lc_check: false"
+      ec2_asg:
+        name: "{{ resource_prefix }}-asg"
+        launch_config_name: "{{ resource_prefix }}-lc-2"
+        health_check_type: EC2
+        desired_capacity: 3
+        min_size: 1
+        max_size: 5
+        health_check_period: 900
+        load_balancers: []
+        vpc_zone_identifier: "{{ testing_subnet.subnet.id }}"
+        wait_for_instances: yes
+        replace_all_instances: yes
+        replace_batch_size: 3
+        lc_check: false
+        wait_timeout: 1800
+        state: present
+        <<: *aws_connection_info
+      async: 1800
+      poll: 0
+      register: asg_job
+
+    - name: get ec2_asg facts for 3 minutes
+      ec2_asg_facts:
+        name: "{{ resource_prefix }}-asg"
+        <<: *aws_connection_info
+      register: output
+      loop_control:
+          pause: 15
+      with_sequence: count=12
+
+    - set_fact:
+        inst_id_json_query: 'results[*].results[*].instances[*].instance_id'
+
+    # Since we started with 3 servers and replace all of them.
+    # We should see 6 servers total.
+    - assert:
+        that:
+          - "lookup('flattened',output|json_query(inst_id_json_query)).split(',')|unique|length == 6"
+
+    - name: Ensure ec2_asg task completes
+      async_status: jid="{{ asg_job.ansible_job_id }}"
+      register: status
+      until: status.finished
+      retries: 200
+      delay: 15
+
+    # ============================================================
+
+    - name: kill asg
+      ec2_asg:
+        name: "{{ resource_prefix }}-asg"
+        state: absent
+        <<: *aws_connection_info
+      async: 300
+
+    # Create new asg with replace_all_instances and lc_check:false
+
+    # Note - this is done async so we can query asg_facts during
+    # the execution. Issues #28087 results in correct
+    # end result, but spin up extraneous instances during execution.
+    - name: "new asg with lc_check: false"
+      ec2_asg:
+        name: "{{ resource_prefix }}-asg"
+        launch_config_name: "{{ resource_prefix }}-lc"
+        health_check_type: EC2
+        desired_capacity: 3
+        min_size: 1
+        max_size: 5
+        health_check_period: 900
+        load_balancers: []
+        vpc_zone_identifier: "{{ testing_subnet.subnet.id }}"
+        wait_for_instances: yes
+        replace_all_instances: yes
+        replace_batch_size: 3
+        lc_check: false
+        wait_timeout: 1800
+        state: present
+        <<: *aws_connection_info
+      async: 1800
+      poll: 0
+      register: asg_job
+
+    # Collect ec2_asg_facts for 3 minutes
+    - name: get ec2_asg facts
+      ec2_asg_facts:
+        name: "{{ resource_prefix }}-asg"
+        <<: *aws_connection_info
+      register: output
+      loop_control:
+          pause: 15
+      with_sequence: count=12
+
+    - set_fact:
+        inst_id_json_query: 'results[*].results[*].instances[*].instance_id'
+
+    # Get all instance_ids we saw and assert we saw number expected
+    # Should only see 3 (don't replace instances we just created)
+    - assert:
+        that:
+          - "lookup('flattened',output|json_query(inst_id_json_query)).split(',')|unique|length == 3"
+
+    - name: Ensure ec2_asg task completes
+      async_status: jid="{{ asg_job.ansible_job_id }}"
+      register: status
+      until: status.finished
+      retries: 200
+      delay: 15
+
 # ============================================================
 
   always: