Christoph Badura <b...@bsd.de> writes: > I've meant to reply early. One question would be if the files need to > contain any data or if rm'ing a bunch of zero-length files triggers the > bug too. I would then guess that it may be connected to freeing file > metadata.
The files I have been creating have just the index of the file, enough to make them not hit any deduplication that might happen. I suspected it was about metadata because of previous experiences with freezing. However, I have now found that just running the system out of memory provokes freezeups, even without zfs activity. Next on my list to check is comparing a system w/o zfs load and with, otherwise the same. > P.S. Seems to me you have some printf debugging ahead of you. Yes, and I've done some of it, and am continuing. I believe there is a deadlock/livelock/missing-wakeup situation in zfs code provoked by lack of system memory, even if there is not a lot of zfs memory in use.
