Hi Christopher, On Mon, Sep 18, 2017 at 10:59:44AM +0200, Christopher Faulet wrote: > Could you check if it works ? In the mean time, I'll see with Willy if > he see a better (and probably deeper) way to fix the bug.
We've discussed off-list with Bin and I think that the problem is not in the cache itself but in the way our splicing works. It's not normal that we can ping-pong between FDs like this. At some point, one should face either a pipe full or lack of data on the socket and should stop bothering its peer. I'm currently trying to figure how this can happen and why, to see if we can address it there. Willy