While fixing ISPN-15394 we now drop a value that is not in the index. Unfortunately, we only drop the value and the index can still remain. This is not an issue from removing a segment as we delete the entire segment tree lookup. There is some case that can cause an index to point to a value that shouldn't exist still.. This JIRA is not to fix that but to fix a side effect of it.

The issue is this code at https://github.com/infinispan/infinispan/blob/4b01f9a55f00519997b0bdcada1aa05f300e4b3c/core/src/main/java/org/infinispan/persistence/sifs/Compactor.java#L583 that uses a continue instead of continuing processing for the entry and actually dropping the index.

Unfortunately, I cannot add a test for this as I do not know the exact case that causes it, but it seems extremely likely that it is from a concurrent clear call.