Member
Some kernel have bug with RAID10 reshape, my friend. I see on forum many problem with mdadm in 6.1. Check dmesg for "md: checkpoint" error. Best price for you to upgrade 6.6 LTS, very stable.
Restic has built-in check. My cron runs `restic check --read-data-subset=10%` weekly, full read-data monthly. But since you use rclone alone, you need something else.
Simplest: after rclone, run `rclone check source: dest:` and grep for ERROR. Pipe to gotify if you want push alerts. I did this before Restic, works fine.
Member
Obrigado mano! I will try this. Very simple, I like.
My RackNerd is in Los Angeles, very far from São Paulo. Ping 180ms, kkkkk. But for backup ok no problem.
I've been burned by "it backed up fine" too many times. My setup:
The Hetzner box is 10 EUR/TB, Wasabi is $7/TB but no egress fees. I restore-tested from Wasabi last month, 400GB in 45 minutes. Acceptable for the price.
Key point: two different backup programs, two different providers, two different countries. Paranoid? The one time Falkenstein had a routing issue, Amsterdam saved me.
Overkill for most. I run borg to rsync.net and call it done. $0.024/GB, pay in USD. Been five years, one restore needed when I rm -rf'd a Nextcloud data dir. Took 20 minutes to pull 80GB.
rsync.net runs on ZFS, you get snapshots for free. No S3 API nonsense, just ssh and borg.
S3 API is not nonsense when you have 12TB and need multipart parallel upload. Borg is single-threaded, my 10Gbps link sits idle.
I run Duplicati to MinIO cluster, 4 nodes, erasure coding. Can lose two nodes, keep writing. Overkill? I also thought so until a PSU took out a motherboard.
Real tip: if you self-host MinIO, use XFS not ext4. ext4 dir indexing chokes at 1M objects.
Appreciate it. Checked dmesg, clean. My reshape was months ago anyway, array stable.
Noted, but I'm not saturating 10Gbps. 340GB in 22 minutes is ~2.6Gbps average, fine for my wireguard tunnel over 1Gbps symmetric. Bottleneck is the crypto and the distance.
Speaking of which: anyone benchmarked wireguard vs plain ssh for rsync? Thinking of dropping the tunnel for a direct ssh with forced ciphers.
WireGuard kernel module adds maybe 3-5% overhead on modern CPUs. Your E5-2680v4 has AES-NI, negligible. SSH with [email protected] is comparable but you lose the always-on nature.
I ran iperf3 through both last year. WireGuard 940 Mbps, SSH 892 Mbps, same hardware. Not worth the config hassle for 48 Mbps.
If you're CPU bound at 2.6 Gbps effective, check your wireguard MTU. 1420 default plus PPPoE or whatever your ISP does can fragment. I run 1380 on one link.
Member
I use ssh with -c [email protected], much more fast than default chacha. My server in Mexico, target in Hetzner Ashburn. Without tunnel I get 700Mbps, with wireguard only 520Mbps. No tunnel for me.
But I use fail2ban and key only, no password.
I tried the multi-provider thing. Nightmare.
Vultr object storage for hot tier, Wasabi for cold, Backblaze B2 for "archive." Three dashboards, three billing cycles, three ways to fail. Vultr's S3 compatibility is 95%, that last 5% broke rclone sync twice.
Consolidated to AWS S3 + Glacier Deep Archive. Yes, Jeff gets my money. But one API, one SDK, one bill in SGD I can predict. Restore from Glacier is 12 hours, but I have S3 Standard for last 30 days if I need fast.
Singapore to Tokyo Vultr was fast. Singapore to anywhere else, meh.
