I encountered a problem with an online cluster:
Peer recovery is very slow on translog stage. Ioutil of source node full intermittently.
We don't have much read/write tasks and from Jstack, we see source node use LuceneChange snapshot as history source which search document from segment.
So if we want to keep soft delete enable, is there any way to use Translog as a HistorySource during recovery, or is there any risk if translog is forced used.