aws-eks-asg-rolling-update-handler

• Added execution timeout

While upgrading EKS from 1.14 to 1.15, I ran into an issue where an in-flight request would hang, and thus the execution would never end. Restarting the pod fixed the issue, and while I'm not exactly sure why it happened, what I do know is that since there's no health check that the liveness probe can perform, adding a timeout to each execution and panicking on failure to respect that execution time limit is a safe way to either allow the pod to recover on its own, or if the issue keeps happening, then at least it will increase visibility.

• Fixed issue with occasional deadlock during drain #6
• Increased drain timeout from 3 minutes to 5 minutes

The application now panics after 10 failed executions in a row.

• Instances are now classified as outdated if their instance type is no longer part of MixedInstancePolicy's instance types overrides

The MixedInstancePolicy available through the usage of a Launch Template is at the ASG level, meaning that if you modify the instance type overrides to allow the usage of multiple instance types on a single ASG, no new launch template version would be created. This update verifies if the instance's instance type is part of the overrides, and if it isn't, then it means that the ASG has been updated, which means that it's outdated.

• Fixed issue with node not getting deleted when it only has pods owned by a DaemonSet

• Do not decrease desired capacity on termination if the AutoScalingGroup's desired capacity is equal to its minimum size #3

• Pods from DaemonSets are now filtered out when calculating resources #1
• Make sure that the ASG doesn't get increased if the node has already been drained or scheduled for termination. Technically, if the node has been drained, this should never cause any problems, but it doesn't hurt to be careful.

• Removed duplicate call to AutoScaling API by breaking for loop on first scale up on ASG

• Rolling update now waits for the number of instances on the ASG to be equal or above the desired capacity to continue the process. This prevents increasing the desired capacity multiple times before an instance has enough time to be registered in the ASG.
• Added logs for the draining operation
• Improved test coverage
• Refactored code

• Fixed issue with AWS region not being properly passed down
• Added debugging logs
• AWS Session is now created only once rather than at every run

First release