deadlock in LMDB if stressing multiple forks
It seems reproducible reliably with various versions of kresd (1.2.6, 1.3.3) and lmdb (0.9.18, 0.9.21). It's enough to stress a pair of kresd processes with resperf, and they tend to lockup within a minute, both during start of a read-only transaction in lmdb (pthread lock apparently taken by "someone third" ATM).
I've been unable to reproduce the problem with a single fork. I think we use the LMDB API correctly, but on the whole more investigation is needed.