Advice for configuring mergerfs for home medis server with 4x14tb drives and 128gb ram?

[Betonhaus]

/mnt/dsk/*/ /mnt/pool fuse.mergerfs defaults,nonempty,allow_other,use_ino,cache.files=partial,moveonenospc=true,dropcacheonclose=true,minfreespace=10G,category.create=mspmfs,fsname=mergerfs 0 0
I've been having issues where if I'm importing like 300-900gb of files at once then reading from the pool at the same time becomes very, very slow. I have 4 14tb drives formatted as ext4 in the pool, and the server has like 128gb and 32 cpu cores so i can allocate much more ram and threads to mergerfs but I'm not entirely sure how to do that? or is my config fine but I should be formatting my drives as XFS? I'm running mergerfs 2.40.2 on Ubuntu 24.04, most individual files are 1-30gb in size

$ xattr -l /mnt/pool/.mergerfs
user.mergerfs.async_read: true
user.mergerfs.auto_cache: false
user.mergerfs.branches: /mnt/dsk/1=RW:/mnt/dsk/2=RW:/mnt/dsk/3=RW:/mnt/dsk/4=RW
user.mergerfs.cache.attr: 1
user.mergerfs.cache.entry: 1
user.mergerfs.cache.files: partial
user.mergerfs.cache.negative_entry: 0
user.mergerfs.cache.readdir: false
user.mergerfs.cache.statfs: 0
user.mergerfs.cache.symlinks: false
user.mergerfs.cache.writeback: false
user.mergerfs.category.action: epall
user.mergerfs.category.create: mspmfs
user.mergerfs.category.search: ff
user.mergerfs.direct_io: false
user.mergerfs.dropcacheonclose: true
user.mergerfs.follow-symlinks: never
user.mergerfs.fsname: mergerfs
user.mergerfs.func.access: ff
user.mergerfs.func.chmod: epall
user.mergerfs.func.chown: epall
user.mergerfs.func.create: mspmfs
user.mergerfs.func.getattr: ff
user.mergerfs.func.getxattr: ff
user.mergerfs.func.link: epall
user.mergerfs.func.listxattr: ff
user.mergerfs.func.mkdir: mspmfs
user.mergerfs.func.mknod: mspmfs
user.mergerfs.func.open: ff
user.mergerfs.func.readlink: ff
user.mergerfs.func.removexattr: epall
user.mergerfs.func.rename: epall
user.mergerfs.func.rmdir: epall
user.mergerfs.func.setxattr: epall
user.mergerfs.func.symlink: mspmfs
user.mergerfs.func.truncate: epall
user.mergerfs.func.unlink: epall
user.mergerfs.func.utimens: epall
user.mergerfs.fuse_msg_size: 256
user.mergerfs.ignorepponrename: false
user.mergerfs.inodecalc: hybrid-hash
user.mergerfs.kernel_cache: false
user.mergerfs.link-exdev: passthrough
user.mergerfs.link_cow: false
user.mergerfs.log.metrics: false
user.mergerfs.minfreespace: 10737418240
user.mergerfs.mount: /mnt/pool
user.mergerfs.moveonenospc: mfs
user.mergerfs.nfsopenhack: off
user.mergerfs.nullrw: false
user.mergerfs.pid: 833
user.mergerfs.posix_acl: false
user.mergerfs.readdirplus: false
user.mergerfs.rename-exdev: passthrough
user.mergerfs.security_capability: true
user.mergerfs.srcmounts: /mnt/dsk/1:/mnt/dsk/2:/mnt/dsk/3:/mnt/dsk/4
user.mergerfs.statfs: base
user.mergerfs.statfs_ignore: none
user.mergerfs.symlinkify: false
user.mergerfs.symlinkify_timeout: 3600
user.mergerfs.threads: 36
user.mergerfs.version: 2.40.4
user.mergerfs.xattr: passthrough

thanks.

[Betonhaus]

/mnt/dsk/*/ /mnt/pool fuse.mergerfs defaults,nonempty,allow_other,use_ino,func.readdir=cor,func.getattr=newest,lazy-umount-mountpoint=true,cache.files=off,moveonenospc=true,dropcacheonclose=true,minfreespace=10G,category.create=mspmfs,fsname=mergerfs 0 0
does this look right?

[trapexit]

I would suggest reading the docs. Much, if not everything you are looking for is there.

• https://trapexit.github.io/mergerfs/quickstart/
• https://trapexit.github.io/mergerfs/faq/configuration_and_policies/#how-can-i-ensure-files-are-collocated-on-the-same-branch
• https://trapexit.github.io/mergerfs/config/threads/
• https://trapexit.github.io/mergerfs/config/functions_categories_and_policies/
• https://trapexit.github.io/mergerfs/performance/

Without seeing the state of the system I can't say why things are slow but one guess would be that you are using mspmfs which is likely resulting in the same underlying branch being used for creates and you are bottlenecked by the drive.

I don't know where you got those settings but given they use old and deprecated values it isn't from the docs. You should get rid of nonempty, allow_other, use_ino, defaults. They do nothing or are unnecessary. cache.files should be off probably. And if you don't want to overwhelm your underlying devices you should change the create policy to rand or pfrd.

[Betonhaus]

/mnt/dsk/*/ /mnt/pool fuse.mergerfs func.readdir=cor,func.getattr=newest,lazy-umount-mountpoint=true,cache.files=off,moveonenospc=true,dropcacheonclose=true,minfreespace=10G,category.create=pfrd,fsname=mergerfs 0 0
so this should work fine then?

[trapexit]

It looks alright, yes.