This post is part of a bigger topic Autoscaling Publishers in AWS.
Now that we have a Launch Configuration, based on it, we can create an Auto Scaling Group. This will be in charge of creating/terminating publisher instances based on some rules (Scaling Policies).
We base our Auto Scaling Group on the Launch Configuration created earlier (sdl_publisher_lc). The very important properties are the Desired, Min and Max number of instances in the group. The Desired is set usually by the Scaling Policies, but more about those later. The Min and Max are the limits of this group. In my case I use Min=1 and Max=3, meaning that I want to have 1 publisher running at all times and when needed, based on load, 2 additional publishers can be added to the group by an 'Increasing size' Scaling Policy.
Once load passes, a 'Decreasing size' Scaling Policy reduces the number of instances in the group.
Scaling Policies
These policies represent the rules for adding/removing instances to the group. They can monitor metrics on the instance itself (e.g. CPU), CloudWatch metrics, or even CloudWatch alarms (e.g. Publish Alarm defined earlier) in order to increase and decrease the number of instances.
We define an 'increase_group_size' as a Scaling Policy with Steps in order to add more publisher instances as the size of the Publish Queue increases.
We also define a 'decrease_group_size' as Simple Scaling Policy that reduces the size of the group. But more details about these policies in a followup post.
Lifecycle Hooks
We are going to use a lifecycle hook when scaling-in (decreasing) the size of out group, when publishing load has passed.
More details about the termination hook in a later post. For now, we create one hook that is going to be raised when the group attempts to terminate an instance. This hook can be intercepted in a CloudWatch event that can then trigger a Lambda Function that will instruct the publisher to shutdown gracefully. Once that happens, the termination hook is released and termination occurs normally.
The Lifecycle Hook Name is important, because in our Lambda Function we will instruct this particular hook to continue termination.
Heartbeat Timeout specifies the time needed to this hook to expire. This means that in the case the termination Lambda did not release the hook in the meantime, the hook will be automatically released once this timeout expires.
Now that we have a Launch Configuration, based on it, we can create an Auto Scaling Group. This will be in charge of creating/terminating publisher instances based on some rules (Scaling Policies).
We base our Auto Scaling Group on the Launch Configuration created earlier (sdl_publisher_lc). The very important properties are the Desired, Min and Max number of instances in the group. The Desired is set usually by the Scaling Policies, but more about those later. The Min and Max are the limits of this group. In my case I use Min=1 and Max=3, meaning that I want to have 1 publisher running at all times and when needed, based on load, 2 additional publishers can be added to the group by an 'Increasing size' Scaling Policy.
Once load passes, a 'Decreasing size' Scaling Policy reduces the number of instances in the group.
Scaling Policies
These policies represent the rules for adding/removing instances to the group. They can monitor metrics on the instance itself (e.g. CPU), CloudWatch metrics, or even CloudWatch alarms (e.g. Publish Alarm defined earlier) in order to increase and decrease the number of instances.
We define an 'increase_group_size' as a Scaling Policy with Steps in order to add more publisher instances as the size of the Publish Queue increases.
We also define a 'decrease_group_size' as Simple Scaling Policy that reduces the size of the group. But more details about these policies in a followup post.
Lifecycle Hooks
We are going to use a lifecycle hook when scaling-in (decreasing) the size of out group, when publishing load has passed.
More details about the termination hook in a later post. For now, we create one hook that is going to be raised when the group attempts to terminate an instance. This hook can be intercepted in a CloudWatch event that can then trigger a Lambda Function that will instruct the publisher to shutdown gracefully. Once that happens, the termination hook is released and termination occurs normally.
The Lifecycle Hook Name is important, because in our Lambda Function we will instruct this particular hook to continue termination.
Heartbeat Timeout specifies the time needed to this hook to expire. This means that in the case the termination Lambda did not release the hook in the meantime, the hook will be automatically released once this timeout expires.
Comments