SuspendProcesses
During a rolling update, you must suspend Auto Scaling processes to avoid making unexpected changes to the group. If an unexpected scaling action changes the state of the group during a rolling update, this can cause CloudFormation to have an inconsistent view of the group and can cause the rolling update to fail. Processes to suspend include: HealthCheck, ReplaceUnhealthy, AZRebalance, AlarmNotification, and ScheduledActions.
Note

MinSuccessfulInstancesPercent
If you are replacing a large number of instances during a rolling update and you are also waiting for a success signal from each instance, you might need to specify a MinSuccessfulInstancePercent value. This is useful to prevent CloudFormation from rolling back the entire stack if a single instance fails to launch. For example, if you perform a rolling update to change your AMI to replace 10 instances, and the first 9 are successful but the last one fails due to a transient issue, CloudFormation would not be able to update the entire stack and would attempt to roll back. If you set MinSuccessfulInstancePercent to 50%, CloudFormation would wait for 5 instances to signal success and then mark the overall AutoScaling Group resource as UPDATE_COMPLETE, because the new AMI is working correctly and there are enough instances in service to maintain traffic. If one of the subsequent instances fails to launch successfully, it would eventually be replaced by your Auto Scaling group health checks.

WaitOnResourceSignals + PauseTime
Enabling WaitOnResourceSignals allows CloudFormation to wait until you have received a success signal before performing the next scaling action.
Note: It is important to note that when WaitOnResourceSignals is set to true, PauseTime becomes a timeout value. CloudFormation waits until the maximum time specified in the PauseTime value for a success signal. If a signal is not received, it cancels the update and attempts to roll back the stack. In order to roll back, CloudFormation performs a rolling update with the same settings, so the same PauseTime value is used upon rollback. It is recommended that you specify a long enough PauseTime for your instances to bootstrap and signal a success.