See attached log in the description. When i fought with this test last night, i tried to make it easier to debug (print inner CFS content sizes etc).
But its still hard, maybe someone else should have a look at the test. We could try to summarize bytes by extension, that might help.
In this case its not the ending size after forceMerge that is the problem, but the peak disk usage as tracked by MDW.
Maybe it only happens in NIGHTLY because both of the two fields involved randomly got term vector options set (which is rare):
[junit4] 1> NOTE: LuceneTestCase: upgrade name=content type=indexed,tokenized,termVector,termVectorPosition,termVectorPayloads
[junit4] 1> NOTE: LuceneTestCase: upgrade name=id type=stored,indexed,tokenized,termVector,termVectorOffsets