> From: "Axel Dörfler" <axeld@xxxxxxxxxxxxxxxx> > > > > heh... I think we also have a big compatibility-problem here... > > > net_server > > > being 10 times faster than the original, OBFS being faster and > > > more > > > stable - > > > heck, this has to wait for R2..! ;-)) > > > > That it's more stable is yet to be proven - at least you can > > unmount > > the original BFS when a thread dies while running a query; it's not > > affected by that kernel bug. > > Hmmmmm... Maybe it does have a workaround in place... I will try to > figure > it out... :) I just think it doesn't cache the index and B+tree inodes, so that it only leaks some memory in that case. While for our BFS, the cached nodes prevents the fs from being unmounted. The only work-around (and that would be the cleanest possibility for R5) is to have a timeout for queries. Adios... Axel.