Configuring auto scaling for Compute instances
Learn how to automatically scale out your Compute instances in response to heavy system usage. By using pre-defined rules that consider factors such as CPU or memory usage, you can configure Orchestration (heat) to add and remove additional instances automatically, when they are needed.
1.1. Architectural overview of auto scaling
The core component providing automatic scaling is Orchestration (heat). You can use Orchestration to define rules using human-readable YAML templates. These rules are applied to evaluate system load based on Telemetry data to find out whether there is need to add more instances into the stack. When the load drops, Orchestration can automatically remove the unused instances again.
Use Telemetry to monitor the performance of your Red Hat OpenStack Platform (RHOSP) environment, collecting data on CPU, storage, and memory utilization for instances and physical hosts. Orchestration templates examine Telemetry data to assess whether any pre-defined action should start.
1.1.1. Key auto scaling terms
Stack - A stack stands for all the resources necessary to operate an application. It can be as simple as a single instance and its resources, or as complex as multiple instances with all the resource dependencies that comprise a multi-tier application.
Templates - YAML scripts that define a series of tasks for Heat to execute. For example, it is preferable to use separate templates for certain functions:
Template File - This is where you define thresholds that Telemetry responds to and define the auto scaling group.
Environment File - Defines the build information for your environment: which flavor and image to use, how to configure the virtual network, and what software to install.
1.2. Example: Auto scaling based on CPU use
In this example, Orchestration examines Telemetry data, and automatically increases the number of instances in response to high CPU use. Create a stack template and environment template to define the rules and subsequent configuration. This example uses existing resources, such as networks, and uses names that might be different to those in your own environment.
Note
The cpu_util metric was deprecated and removed from Red Hat OpenStack Platform.
Procedure
- Create the environment template, describing the instance flavor, networking configuration, and image type. Save the template in the
/home/<user>/stacks/example1/cirros.yamlfile. Replace the<user>variable with a real user name.
heat_template_version: 2016-10-14
description: Template to spawn an cirros instance.
parameters:
metadata:
type: json
image:
type: string
description: image used to create instance
default: cirros
flavor:
type: string
description: instance flavor to be used
default: m1.tiny
key_name:
type: string
description: keypair to be used
default: mykeypair
network:
type: string
description: project network to attach instance to
default: internal1
external_network:
type: string
description: network used for floating IPs
default: external_network
resources:
server:
type: OS::Nova::Server
properties:
block_device_mapping:
- device_name: vda
delete_on_termination: true
volume_id: { get_resource: volume }
flavor: {get_param: flavor}
key_name: {get_param: key_name}
metadata: {get_param: metadata}
networks:
- port: { get_resource: port }
port:
type: OS::Neutron::Port
properties:
network: {get_param: network}
security_groups:
- default
floating_ip:
type: OS::Neutron::FloatingIP
properties:
floating_network: {get_param: external_network}
floating_ip_assoc:
type: OS::Neutron::FloatingIPAssociation
properties:
floatingip_id: { get_resource: floating_ip }
port_id: { get_resource: port }
volume:
type: OS::Cinder::Volume
properties:
image: {get_param: image}
size: 1
- Register the Orchestration resource in
~/stacks/example1/environment.yaml:
- Create the stack template. Describe the CPU thresholds to watch for and how many instances to add. An instance group is also created that defines the minimum and maximum number of instances that can participate in this template.
Note
The cpu_util metric was deprecated and removed from Red Hat OpenStack Platform. To obtain the equivalent functionality, use the cumulative cpu metric and an archive policy that includes the rate:mean aggregation method. For example, ceilometer-high-rate and ceilometer-low-rate. You must convert the threshold value from % to ns to use the cpu metric for the CPU utilisation alarm. The formula is: time_ns = 1,000,000,000 x {granularity} x {percentage_in_decimal}. For example, for a threshold of 80% with a granularity of 1s, the threshold is 1,000,000,000 x 1 x 0.8 = 800,000,000.0
- Save the following values in
~/stacks/example1/template.yaml:
heat_template_version: 2016-10-14
description: Example auto scale group, policy and alarm
resources:
scaleup_group:
type: OS::Heat::AutoScalingGroup
properties:
cooldown: 300
desired_capacity: 1
max_size: 3
min_size: 1
resource:
type: OS::Nova::Server::Cirros
properties:
metadata: {"metering.server_group": {get_param: "OS::stack_id"}}
scaleup_policy:
type: OS::Heat::ScalingPolicy
properties:
adjustment_type: change_in_capacity
auto_scaling_group_id: { get_resource: scaleup_group }
cooldown: 300
scaling_adjustment: 1
scaledown_policy:
type: OS::Heat::ScalingPolicy
properties:
adjustment_type: change_in_capacity
auto_scaling_group_id: { get_resource: scaleup_group }
cooldown: 300
scaling_adjustment: -1
cpu_alarm_high:
type: OS::Aodh::GnocchiAggregationByResourcesAlarm
properties:
description: Scale up if CPU > 80%
metric: cpu
aggregation_method: rate:mean
granularity: 1
evaluation_periods: 3
threshold: 800000000.0
resource_type: instance
comparison_operator: gt
alarm_actions:
- str_replace:
template: trust+url
params:
url: {get_attr: [scaleup_policy, signal_url]}
query:
str_replace:
template: '{"=": {"server_group": "stack_id"}}'
params:
stack_id: {get_param: "OS::stack_id"}
cpu_alarm_low:
type: OS::Aodh::GnocchiAggregationByResourcesAlarm
properties:
metric: cpu
aggregation_method: rate:mean
granularity: 1
evaluation_periods: 3
threshold: 200000000.0
resource_type: instance
comparison_operator: lt
alarm_actions:
- str_replace:
template: trust+url
params:
url: {get_attr: [scaledown_policy, signal_url]}
query:
str_replace:
template: '{"=": {"server_group": "stack_id"}}'
params:
stack_id: {get_param: "OS::stack_id"}
outputs:
scaleup_policy_signal_url:
value: {get_attr: [scaleup_policy, signal_url]}
scaledown_policy_signal_url:
value: {get_attr: [scaledown_policy, signal_url]}
- Build the environment and deploy the instance:
$ openstack stack create -t template.yaml -e environment.yaml example
+---------------------+--------------------------------------------+
| Field | Value |
+---------------------+--------------------------------------------+
| id | 248a98bb-f56e-4934-a281-fffde62d78d8 |
| stack_name | example |
| description | Example auto scale group, policy and alarm |
| creation_time | 2017-03-06T15:00:29Z |
| updated_time | None |
| stack_status | CREATE_IN_PROGRESS |
| stack_status_reason | Stack CREATE started |
+---------------------+--------------------------------------------+
- Orchestration creates the stack and launches a defined minimum number of cirros instances, as defined in the
min_sizeparameter of thescaleup_groupdefinition. Verify that the instances were created successfully:
$ openstack server list
+--------------------------------------+-------------------------------------------------------+--------+------------+-------------+-------------------------------------+
| ID | Name | Status | Task State | Power State | Networks |
+--------------------------------------+-------------------------------------------------------+--------+------------+-------------+-------------------------------------+
| e1524f65-5be6-49e4-8501-e5e5d812c612 | ex-3gax-5f3a4og5cwn2-png47w3u2vjd-server-vaajhuv4mj3j | ACTIVE | - | Running | internal1=10.10.10.9, 192.168.122.8 |
+--------------------------------------+-------------------------------------------------------+--------+------------+-------------+-------------------------------------+
- Orchestration also creates two cpu alarms which trigger scale-up or scale-down events, as defined in
cpu_alarm_highandcpu_alarm_low. Verify that the triggers exist:
$ openstack alarm list
+--------------------------------------+--------------------------------------------+-------------------------------------+-------------------+----------+---------+
| alarm_id | type | name | state | severity | enabled |
+--------------------------------------+--------------------------------------------+-------------------------------------+-------------------+----------+---------+
| 022f707d-46cc-4d39-a0b2-afd2fc7ab86a | gnocchi_aggregation_by_resources_threshold | example-cpu_alarm_high-odj77qpbld7j | insufficient data | low | True |
| 46ed2c50-e05a-44d8-b6f6-f1ebd83af913 | gnocchi_aggregation_by_resources_threshold | example-cpu_alarm_low-m37jvnm56x2t | insufficient data | low | True |
+--------------------------------------+--------------------------------------------+-------------------------------------+-------------------+----------+---------+
1.3. Testing automatic scaling up of instances
Orchestration can scale instances automatically based on the cpu_alarm_high threshold definition. When the CPU utilization reaches a value defined in the threshold parameter, another instance starts up to balance the load. The threshold value in the template.yaml file is set to 80%.
Procedure
- Log on to the instance and run several
ddcommands to generate the load:
$ ssh -i ~/mykey.pem cirros@192.168.122.8
$ sudo dd if=/dev/zero of=/dev/null &
$ sudo dd if=/dev/zero of=/dev/null &
$ sudo dd if=/dev/zero of=/dev/null &
- After you run the
ddcommands, you can expect to have 100% CPU use in the cirros instance. Verify that the alarm has been triggered:
$ openstack alarm list
+--------------------------------------+--------------------------------------------+-------------------------------------+-------+----------+---------+
| alarm_id | type | name | state | severity | enabled |
+--------------------------------------+--------------------------------------------+-------------------------------------+-------+----------+---------+
| 022f707d-46cc-4d39-a0b2-afd2fc7ab86a | gnocchi_aggregation_by_resources_threshold | example-cpu_alarm_high-odj77qpbld7j | alarm | low | True |
| 46ed2c50-e05a-44d8-b6f6-f1ebd83af913 | gnocchi_aggregation_by_resources_threshold | example-cpu_alarm_low-m37jvnm56x2t | ok | low | True |
+--------------------------------------+--------------------------------------------+-------------------------------------+-------+----------+---------+
- After approximately 60 seconds, Orchestration starts another instance and adds it into the group. To verify this, enter the following command:
$ openstack server list
+--------------------------------------+-------------------------------------------------------+--------+------------+-------------+---------------------------------------+
| ID | Name | Status | Task State | Power State | Networks |
+--------------------------------------+-------------------------------------------------------+--------+------------+-------------+---------------------------------------+
| 477ee1af-096c-477c-9a3f-b95b0e2d4ab5 | ex-3gax-4urpikl5koff-yrxk3zxzfmpf-server-2hde4tp4trnk | ACTIVE | - | Running | internal1=10.10.10.13, 192.168.122.17 |
| e1524f65-5be6-49e4-8501-e5e5d812c612 | ex-3gax-5f3a4og5cwn2-png47w3u2vjd-server-vaajhuv4mj3j | ACTIVE | - | Running | internal1=10.10.10.9, 192.168.122.8 |
+--------------------------------------+-------------------------------------------------------+--------+------------+-------------+---------------------------------------+
- After another short period, observe that Orchestration has auto scaled again to three instances. The configuration is set to three instances maximally, so it cannot scale any higher (the
scaleup_groupdefinition:max_size). To verify that there are three instances, enter the following command:
$ openstack server list
+--------------------------------------+-------------------------------------------------------+--------+------------+-------------+---------------------------------------+
| ID | Name | Status | Task State | Power State | Networks |
+--------------------------------------+-------------------------------------------------------+--------+------------+-------------+---------------------------------------+
| 477ee1af-096c-477c-9a3f-b95b0e2d4ab5 | ex-3gax-4urpikl5koff-yrxk3zxzfmpf-server-2hde4tp4trnk | ACTIVE | - | Running | internal1=10.10.10.13, 192.168.122.17 |
| e1524f65-5be6-49e4-8501-e5e5d812c612 | ex-3gax-5f3a4og5cwn2-png47w3u2vjd-server-vaajhuv4mj3j | ACTIVE | - | Running | internal1=10.10.10.9, 192.168.122.8 |
| 6c88179e-c368-453d-a01a-555eae8cd77a | ex-3gax-fvxz3tr63j4o-36fhftuja3bw-server-rhl4sqkjuy5p | ACTIVE | - | Running | internal1=10.10.10.5, 192.168.122.5 |
+--------------------------------------+-------------------------------------------------------+--------+------------+-------------+---------------------------------------+
1.4. Scaling down instances automatically
You can use Orchestration to automatically scale down instances based on the cpu_alarm_low threshold. In this example, the instances are scaled down when CPU use is below 5%.
Procedure
- Terminate the running
ddprocesses and observe Orchestration begin to scale the instances back down.
- When you stop the
ddprocesses, this triggers thecpu_alarm_low eventalarm. As a result, Orchestration begins to automatically scale down and remove the instances. Verify that the corresponding alarm has triggered:
$ openstack alarm list
+--------------------------------------+--------------------------------------------+-------------------------------------+-------+----------+---------+
| alarm_id | type | name | state | severity | enabled |
+--------------------------------------+--------------------------------------------+-------------------------------------+-------+----------+---------+
| 022f707d-46cc-4d39-a0b2-afd2fc7ab86a | gnocchi_aggregation_by_resources_threshold | example-cpu_alarm_high-odj77qpbld7j | ok | low | True |
| 46ed2c50-e05a-44d8-b6f6-f1ebd83af913 | gnocchi_aggregation_by_resources_threshold | example-cpu_alarm_low-m37jvnm56x2t | alarm | low | True |
+--------------------------------------+--------------------------------------------+-------------------------------------+-------+----------+---------+
After a few minutes, Orchestration continually reduce the number of instances to the minimum value defined in the min_size parameter of the scaleup_group definition. In this scenario, the min_size parameter is set to 1.
1.5. Troubleshooting for auto scaling
If your environment is not working properly, you can look for errors in the log files and history records.
Procedure
- To retrieve information on state transitions, list the stack event records:
$ openstack stack event list example
2017-03-06 11:12:43Z [example]: CREATE_IN_PROGRESS Stack CREATE started
2017-03-06 11:12:43Z [example.scaleup_group]: CREATE_IN_PROGRESS state changed
2017-03-06 11:13:04Z [example.scaleup_group]: CREATE_COMPLETE state changed
2017-03-06 11:13:04Z [example.scaledown_policy]: CREATE_IN_PROGRESS state changed
2017-03-06 11:13:05Z [example.scaleup_policy]: CREATE_IN_PROGRESS state changed
2017-03-06 11:13:05Z [example.scaledown_policy]: CREATE_COMPLETE state changed
2017-03-06 11:13:05Z [example.scaleup_policy]: CREATE_COMPLETE state changed
2017-03-06 11:13:05Z [example.cpu_alarm_low]: CREATE_IN_PROGRESS state changed
2017-03-06 11:13:05Z [example.cpu_alarm_high]: CREATE_IN_PROGRESS state changed
2017-03-06 11:13:06Z [example.cpu_alarm_low]: CREATE_COMPLETE state changed
2017-03-06 11:13:07Z [example.cpu_alarm_high]: CREATE_COMPLETE state changed
2017-03-06 11:13:07Z [example]: CREATE_COMPLETE Stack CREATE completed successfully
2017-03-06 11:19:34Z [example.scaleup_policy]: SIGNAL_COMPLETE alarm state changed from alarm to alarm (Remaining as alarm due to 1 samples outside threshold, most recent: 95.4080102993)
2017-03-06 11:25:43Z [example.scaleup_policy]: SIGNAL_COMPLETE alarm state changed from alarm to alarm (Remaining as alarm due to 1 samples outside threshold, most recent: 95.8869217299)
2017-03-06 11:33:25Z [example.scaledown_policy]: SIGNAL_COMPLETE alarm state changed from ok to alarm (Transition to alarm due to 1 samples outside threshold, most recent: 2.73931707966)
2017-03-06 11:39:15Z [example.scaledown_policy]: SIGNAL_COMPLETE alarm state changed from alarm to alarm (Remaining as alarm due to 1 samples outside threshold, most recent: 2.78110858552)
- To read the alarm history log, enter the following command:
$ openstack alarm-history show 022f707d-46cc-4d39-a0b2-afd2fc7ab86a
+----------------------------+------------------+-----------------------------------------------------------------------------------------------------+--------------------------------------+
| timestamp | type | detail | event_id |
+----------------------------+------------------+-----------------------------------------------------------------------------------------------------+--------------------------------------+
| 2017-03-06T11:32:35.510000 | state transition | {"transition_reason": "Transition to ok due to 1 samples inside threshold, most recent: | 25e0e70b-3eda-466e-abac-42d9cf67e704 |
| | | 2.73931707966", "state": "ok"} | |
| 2017-03-06T11:17:35.403000 | state transition | {"transition_reason": "Transition to alarm due to 1 samples outside threshold, most recent: | 8322f62c-0d0a-4dc0-9279-435510f81039 |
| | | 95.0964497325", "state": "alarm"} | |
| 2017-03-06T11:15:35.723000 | state transition | {"transition_reason": "Transition to ok due to 1 samples inside threshold, most recent: | 1503bd81-7eba-474e-b74e-ded8a7b630a1 |
| | | 3.59330523447", "state": "ok"} | |
| 2017-03-06T11:13:06.413000 | creation | {"alarm_actions": ["trust+http://fca6e27e3d524ed68abdc0fd576aa848:delete@192.168.122.126:8004/v1/fd | 224f15c0-b6f1-4690-9a22-0c1d236e65f6 |
| | | 1c345135be4ee587fef424c241719d/stacks/example/d9ef59ed-b8f8-4e90-bd9b- | |
| | | ae87e73ef6e2/resources/scaleup_policy/signal"], "user_id": "a85f83b7f7784025b6acdc06ef0a8fd8", | |
| | | "name": "example-cpu_alarm_high-odj77qpbld7j", "state": "insufficient data", "timestamp": | |
| | | "2017-03-06T11:13:06.413455", "description": "Scale up if CPU > 80%", "enabled": true, | |
| | | "state_timestamp": "2017-03-06T11:13:06.413455", "rule": {"evaluation_periods": 1, "metric": | |
| | | "cpu_util", "aggregation_method": "mean", "granularity": 300, "threshold": 80.0, "query": "{\"=\": | |
| | | {\"server_group\": \"d9ef59ed-b8f8-4e90-bd9b-ae87e73ef6e2\"}}", "comparison_operator": "gt", | |
| | | "resource_type": "instance"}, "alarm_id": "022f707d-46cc-4d39-a0b2-afd2fc7ab86a", | |
| | | "time_constraints": [], "insufficient_data_actions": null, "repeat_actions": true, "ok_actions": | |
| | | null, "project_id": "fd1c345135be4ee587fef424c241719d", "type": | |
| | | "gnocchi_aggregation_by_resources_threshold", "severity": "low"} | |
+----------------------------+------------------+-----------------------------------------------------------------------------------------------------+-------------------------------------
- To view the records of scale-out or scale-down operations that heat collects for the existing stack, you can use the
awkcommand to parse theheat-engine.log:
$ awk '/Stack UPDATE started/,/Stack CREATE completed successfully/ {print $0}' /var/log/containers/heat/heat-engine.log
- To view
aodh-related information, examine theevaluator.log: