readiness-override-options.md

Readiness Override Options (Supported with Caveats)

The autoscaler supports the following readiness override options for cluster scaling flows:

• node_filter_ignore_drain
• node_filter_ignore_init
• ignore_asg_events

These options are supported, but they trade strict readiness behavior for faster progress during node transitions.

Risks to consider

• potential scaling oscillation
• temporary capacity overshoot/undershoot while transitions are still converging
• reduced accuracy of “stable cluster” assumptions during evaluation windows

When to use

• spot-heavy clusters with frequent replacement events
• environments with long drain windows (for example, long-running jobs)
• workloads where meeting SLA and restoring headroom quickly is prioritized

When not to use

• clusters prioritizing strict stability and conservative scaling behavior
• environments where capacity accounting accuracy is more important than scaling responsiveness
• workloads sensitive to short-term over/under scaling during transitions