All fractal trees share the same space within the cachetable and other common resources down in PerconaFT. It is a known issue that a drop/delete of tables/indices can take a long time and possibly hold some common locks that might cause other accessors to slow or stall https://tokutek.atlassian.net/browse/FT-715
I can't say for certain if that is what you are seeing but it sounds like a possibility. It would be helpful if we could get a PMP or similar trace when you experience these stalls. It might help point us in some direction.
All fractal trees share the same space within the cachetable and other common resources down in PerconaFT. It is a known issue that a drop/delete of tables/indices can take a long time and possibly hold some common locks that might cause other accessors to slow or stall https:/ /tokutek. atlassian. net/browse/ FT-715
I can't say for certain if that is what you are seeing but it sounds like a possibility. It would be helpful if we could get a PMP or similar trace when you experience these stalls. It might help point us in some direction.