Both force merge and best_compression are resource intensive. If you set the ILM hot->warm rollover for the following actions:
- force merge
- change compression to best_compression
- change replica from 0 to 1
On which nodes is the heavy resource usage? Hot or warm?