Excessive Redistribution Events During Cluster Size Changes

[VinozzZ]

Description

Currently, Refinery’s trace redistribution logic can trigger multiple redistribution events when cluster size changes occur back-to-back in a short time frame. Each cluster size change immediately triggers a redistribution, leading to increased traffic on nodes, which can worsen their stability, especially in environments with frequent scaling events.

Proposed Solution:

Introduce an initial delay using a timer before starting the redistribution process. The behavior will be as follows:

• A timer will start after a cluster size change is detected.
• Redistribution will only begin once the timer expires.
• If another cluster size change occurs before the timer finishes, the timer will reset.
• This ensures that redistribution only happens once the cluster has stabilized, reducing unnecessary redistribution events.

[akvanhar]

Fixed by #1403