Skip to content

server.md

Latest commit

 

History

History
86 lines (61 loc) · 2.82 KB

server.md

File metadata and controls

86 lines (61 loc) · 2.82 KB

Server

Some notes about my server configuration (as I'll be moving it to NixOS soon). After the migration, the information here can be removed once it is all in code.

Functions:

  • Main storage (/mnt/storage)
  • Run docker containers for my homelab
  • Serve all the Windows Games/Utilities folders to Zephyr over SAMBA

Drive configuration

Each volume/dataset may have:

  • Snapshots
  • Backups (6-monthly)
  • Scrubs (monthly)

ZFS

  • Dataset properties: atime=off, compression=zstd-3
Pool Drives Scrub
storage 4x HDD, RAIDZ2 ✅
Dataset Snaps Backup
storage/data ✅ ✅

Note: RAIDZ should not be used with databases.

Btrfs

Mount options: noatime,compress=zstd

Volume Drives Scrub
/ 2x SSD, mirror ✅
Subvolume Snaps Backup
root ✅ ✅
server ✅ ✅
var/log
var/lib/docker
nix

Direct writes and checksum errors

Btrfs has issues syncing the checksum with the written data when direct writes (O_DIRECT) are used. Some sources of direct writes are libvirt's cache=none and fio's direct=1 flags.

For this reason, in libvirt, avoid using cache=none when writing to Btrfs.

Further reading:

ZFS Backup + Scrub Command

The following can be run on a local computer with the archive drive imported prior, to backup the server's snapshots as well as scrub the archive drive:

sudo bash -c "
  syncoid \
    --delete-target-snapshots \
    --sshkey ~/.ssh/id_ed25519 \
    --sshport 39483 \
    server:storage/data archive/storage/data && \
  zpool scrub -w archive && \
  zpool status archive
"

The archive drive pool's root dataset is configured with compression=zstd-3 and atime=off.

Note: We don't preserve properties (--preserve-properties) because we don't want to preserve the mountpoint property.