Elastic Beanstalk Rolling Environment Configuration Updates
When a configuration change requires instances to be replaced, Elastic Beanstalk can perform the update in batches to avoid downtime while the change is propagated. During a rolling update, capacity is only reduced by the size of a single batch, which you can configure. Elastic Beanstalk takes one batch of instances out of service, terminates them, and then launches a batch with the new configuration. After the new batch starts serving requests, Elastic Beanstalk moves on to the next batch.
Rolling configuration update batches can be processed periodically (time-based), with a delay between each batch, or based on health. For time-based rolling updates, you can configure the amount of time that Elastic Beanstalk waits after completing the launch of a batch of instances before moving on to the next batch. This pause time allows your application to bootsrap and start serving requests.
With health-based rolling updates, Elastic Beanstalk waits until instances in a batch pass health checks before moving on to the next batch. The health of an instance is determined by the health reporting system, which can be basic or enhanced. With basic health, a batch is considered healthy as soon as all instances in it pass ELB health checks.
With enhanced health reporting, all of the instances in a batch must pass multiple consecutive health checks before Elastic Beanstalk will move on to the next batch. In addition to ELB health checks, which only check your instances, enhanced health monitors application logs and the state of your environment's other resources. In a web server environment with enhanced health, all instances must pass 12 health checks over the course of two minutes (18 checks over three minutes for worker environments). If any instance fails one health check, the count resets.
If a batch does not become healthy within the rolling update timeout (default is 30
minutes), the update is cancelled. Rolling update timeout is a configuration option that is available in the
namespace. If your application does not pass health checks with
Ok status but is
stable at a different level, you can set the
HealthCheckSuccessThreshold option in
aws:elasticbeanstalk:command namespace to change the level at which
Elastic Beanstalk considers an instance to be healthy.
If the rolling update process fails, Elastic Beanstalk starts another rolling update to rollback to the previous configuration. If the rollback fails, then you must terminate the environment and then launch a new environment. A rolling update can fail due to failed health checks or if launching new instances causes you to exceed the limits on your account. An alternative method is to deploy the new version of your application to a different environment and then perform a CNAME swap to redirect traffic with zero downtime. See Blue/Green Deployments with AWS Elastic Beanstalk for more information.
Rolling Updates vs Rolling Deployments
Rolling updates occur when you change settings that require new EC2 instances to be provisioned for your environment. This includes changes to the Auto Scaling group configuration such as instance type and key pair settings, and changes to VPC settings. In a rolling update, each batch of instances is terminated prior to a new batch being provisioned to replace it.
Rolling deployments occur whenever you deploy your application and can typically be performed without replacing instances in your environment. Elastic Beanstalk takes each batch out of service, deploys the new application version, and then places it back in service.
The exception to this is if you change settings that require instance replacement at the same time you deploy a new application version. For example, if you change the key name settings in a configuration file in your source bundle and deploy it to your environment, you trigger a rolling update. Instead of deploying your new application version to each batch of existing instances, a new batch of instances is provisioned with the new configuration. A separate deployment does not occur in this case because the new instances are brought up with the new application version.
Any time that new instances are provisioned as part of an environment update, there is a deployment phase where your application's source code is deployed to the new instances and any configuration settings that modify the operating system or software on the instances are applied. Deployment health check settings (Ignore health check, Healthy threshold and Command timeout) also apply to health based rolling udates and immutable updates during the deployment phase.
Configuring Rolling Updates
You can enable and configure rolling updates in the Elastic Beanstalk Management Console.
To enable rolling updates
The Configuration Updates section of the Updates and Deployments page has the following options for rolling updates:
Update type – Choose from the following options: Elastic Beanstalk waits after it finishes updating a batch of instances before moving on to the next batch, to allow those instances to finish bootstrapping and start serving traffic.
Rolling based on Health – Wait until instances in the current batch are healthy before placing instances in service and starting the next batch.
Rolling based on Time – Specify an amount of time to wait between launching new instances and placing them in service before starting the next batch.
Immutable – Apply the configuration change to a fresh group of instances by performing an immutable update.
Maximum batch size – The number of instances to replace in each batch, between
10000. By default, this value is one-third of the minimum size of the autoscaling group, rounded up.
Minimum instances in service – The minimum number of instances to keep running while other instances are being updated, between
9999. The default value is either the minimum size of the autoscaling group or one less than the maximum size of the autoscaling group, whichever number is lower.
Pause time (time-based only) – The amount of time to wait after a batch is updated before moving on to the next batch, to allow your application to start receiving traffic. Between 0 seconds to 1 hour.
The aws:autoscaling:updatepolicy:rollingupdate namespace
RollingUpdateEnabled option to enable rolling updates, and
RollingUpdateType to choose the update type. The following values are supported
Health.– Wait until instances in the current batch are healthy before placing instances in service and starting the next batch.
Time– Specify an amount of time to wait between launching new instances and placing them in service before starting the next batch.
Immutable– Apply the configuration change to a fresh group of instances by performing an immutable update.
When you enable rolling updates, set the
MinInstancesInService options to configure the size of each batch. For
time-based and health-based rolling updates, you can also configure a
For example, to launch up to five instances at a time, while maintaining at least two instances in service, and wait five minutes and 30 seconds between batches, specify the following options and values:
option_settings: aws:autoscaling:updatepolicy:rollingupdate: RollingUpdateEnabled: true MaxBatchSize: 5 MinInstancesInService: 2 RollingUpdateType: Time PauseTime: PT5M30S
To enable health based rolling updates, with a 45 minute timeout for each batch, specify the following options and values:
option_settings: aws:autoscaling:updatepolicy:rollingupdate: RollingUpdateEnabled: true MaxBatchSize: 5 MinInstancesInService: 2 RollingUpdateType: Health Timeout: PT45M
Timeout and PauseTime values must be specified in ISO8601 duration:
where each # is the number of hours, minutes, and/or seconds, respectively.
The EB CLI and Elastic Beanstalk console apply recommended values for the preceding options. These settings must be removed if you want to use configuration files to configure the same. See Recommended Values for details.