Hammer FS dies during pruning after massive write load
I have one or two apps that use leveldb badly, and read and write a lot of data to disk.
I have a 256GB SSD, and to avoid running out of space within an hour or two of starting these apps, I have them use the DB on a no history PFS and clean the FS intraday. This kernel freezes around 40% of the time during pruning. The other operations are never a problem.
I don't really know what I can give as useful information. Normal usage without these apps is fine; hammer never gives a problem. Just after 100s of GBs of writes pruning sometimes kills the kernel. The FS is always fine and not corrupt after rebooting, and redoing the clean usually then succeeds (though once it died again, after making progress freeing space).
Incidentally, during these pruning phases that do succeed, X completely freezes for 20-30s several times; the mouse doesn't move and the desktop clock I have stops ticking. Not losing userspace interactivity would be nice.
$ uname -a
DragonFly zotac.akihabara.co.uk 4.5-DEVELOPMENT DragonFly v22.214.171.1241.g2e03c8-DEVELOPMENT #2: Sun Mar 20 21:54:05 JST 2016 email@example.com:/usr/obj/usr/src/sys/X86_64_GENERIC x86_64
Updated by tkusumi about 6 years ago
Hmm.. let me think what I can do.
It'll probably be the same with 4.7, since no one has really touched the real core part of hammer1 in the past 1 year or so.
BTW, have you ever used any other databases like postgres or mysql on your DragonFly ?