Skip to content
Register

Backup strategy for 10 cheap VPS spread across 5 continents?

VPS Hosting by nerdbann 24 replies 2.7K views
nerdbann

Member OP

DNS Propagator

10 #1

Got ten boxes running random stuff on Contabo, OVHcloud, Vultr, Hetzner, and some no-name in Sao Paulo. Tired of logging into each one to run backups. Want something centralized I can trust.

Currently doing manual rsync to another cheap VPS but that feels like a house of cards. What do you all use for pull-based backups across providers? Or should I push from each node?

Also if anyone wants to check propagation, my main domain is on a 300s TTL and I can already see it hasn't hit all resolvers yet. dig @8.8.8.8 vs dig @1.1.1.1 gives different A records right now. Check your resolver before you blame me.

pieter_rtm

Member

4 #2

Pull needs SSH keys on every host and a central target with open inbound.

Bullet list for your setup:

  • 10 nodes, 5 providers, unknown bandwidth quotas
  • Pull model: one backup server, 10 outbound connections, single point of failure
  • Push model: 10 cron jobs, 10 credentials to rotate, harder to monitor

As said, I run pull with rclone over SFTP. 12 nodes, 3TB monthly, OVHcloud to Hetzner. Cost is 4 EUR for the backup target. Works.

Your TTL issue is not propagation. It is your authoritative server. Check there first.

Containers before it was cool
armstrongvds

Member

ARM Enjoyer

#3

Push from the nodes! I run this on Ampere Altras at three providers and it is insane value! The ARM chips run circles around x86 on compression tasks!

I use restic to a Vultr storage VPS, 2TB for like 6 bucks. Deduplication is wild, cut my backup size by 70%!

What CPUs are your targets running? If they are not ARM you are leaving efficiency on the table! The power draw alone!

one small ping for man...
FlowSana

Member

SRE

#4

1. Evaluate your threat model: provider failure, region outage, or configuration loss?
2. For 10 nodes, I recommend a hybrid approach: local snapshots (restic, borg 1.2.8) plus offsite replication.
3. Pull-based backups require: SSH key distribution, hardened backup host, network ACLs per provider.
4. Push-based backups require: credential management, per-node scheduling, monitoring for silent failures.
5. Bandwidth costs vary: Contabo charges €0.01/GB overage in APAC; Hetzner includes 5TB outbound on storage plans.

Practical recommendation: deploy a HostHatch storage instance in EU, use restic REST backend, push from all 10 nodes with systemd timers and healthchecks.io alerting.

It's always DNS. Always.
MarseilleMee

Member

Latency Poet

#5

nerdbann said:
Manual rsync to another cheap VPS
Actually I do the same than you @nerdbann ! I have 8 vps and I rsync to another cheap one ! How to say ... it is not perfect but it works ?

The problem is when the backup server goes down ... it doesn't go ! You understand ? You lose everything in same time !

I tried to use rclone but actually the config is very difficult for me !

torontofields

Member

TOS Reader

4 #6

Before you centralize, check your TOS. Contabo section 7.3 prohibits "automated data collection or aggregation services" on their standard VPS tier. I think OVHcloud's acceptable use policy has something about backup storage nodes needing a storage VPS with different terms.

Not legal advice, but "backup server" and "storage VPS" are distinct definitions. One is compute with incidental storage; the other is storage with incidental compute. If your provider audits, this distinction matters.

Also: rsync to another cheap VPS is not a backup. It is replication. A backup requires versioning and point-in-time recovery capability. Check your definitions.

DA

Member

#7

Which restic version are you running on those arm nodes

ZH

Member

#8

70% seems high

cpanelliw780

Member

shared hosting survivor

#9

What bandwidth quota does the no-name sao paulo box have

armstrongvds

Member

ARM Enjoyer

#10

daniel34 said:
Which restic version are you running on those arm nodes
restic 0.16.4, built with go1.21.6. Compiled it myself on the Altra, actually faster than the amd64 binary on x86 for my dataset. The SHA256 acceleration on ARMv8.2 is no joke. I built from https://restic.net.

zhang39 said:
70% seems high
It is what it is. Your mileage depends on how much duplicate data you have. I run similar configs across three nodes, so dedup hits hard.

one small ping for man...

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft