[Bug 293957] Kernel Panic in fusefs: page fault (0x78) in fuse_vnop_write during vnode recycling (csync2 synchro)

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=293957

--- Comment #2 from zjk <zjk7@wp.pl> ---
(In reply to Alan Somers from comment #1)
Yes, you're right. At first glance, it seems like FUSE (from the moosefs
mount), but after extensive testing, I concluded that the issue is related to
memory management – ​​this needs to be verified:

1. Situation: I started synchronizing a website (40GB, several hundred files)
from one server to multiple servers.

2. The website content contains mounts to the network moosefs – but I enabled
synchronization exclusions for the entire mounted resource (this disk is
mounted the same way on other servers).

3. csync2 causes panics not only during uploads, but also only during local
file scanning. I've now confirmed that rsync also causes panics. (Ultimately,
after many restarts, I have servers syncing, but it's not suitable for
production – the restarts are regular/repeatable during sync). The HTTP server
is running constantly, and sometimes the core dump trace leads to it.

4. I changed
kern.maxvnodes=500000-2000000 (better at 2000000)
vfs.fusefs.iov_credit=4-128 (better at 64)
vfs.fusefs.data_cache_mode=0-1 (better at 0)
where vfs.numvnodes changes from 60000 to 250000 (idle), when csync2 is running
- close to 800000.

5. for moosefs
mfsentrycacheto,mfsattrcacheto,mfsdirentrycacheto - ranges from 5 to 600
(better at 60). Moosefs mount options:
rw,late,failok,noatime,mfsdelayedinit,mfscachemode=DIRECT,mfsport=9421,mountprog=/usr/local/bin/mfsmount,mfsentrycacheto=60,mfsattrcacheto=60,mfsdirentrycacheto=60,mfstimeout=30,mfsioretries=60

6. I still have free memory (1GB) and swap is at 0 usage.

I'm sending a core dump, 2 examples:
https://www.filemail.com/d/sqswmdirehrziwk

Best regards
zjk

-- 
You are receiving this mail because:
You are the assignee for the bug.