There is some indication, that this issue doesn't occur with the kernel from oldstable (i.e. 3.16.51-3+deb8u1, with the rest of our userland still at stretch)... at least we have on node downgraded to that kernel, and so far that one wasn't hit in nearly two weeks, while many other were. If there's anything you'd need in terms of data/debug output, etc. do not hesitate to ask. Thanks, Chris.