Hacker News new | ask | show | jobs
by tcfunk 3199 days ago
Never heard of Scaleway...those prices seem to beat the pants of DigitalOcean. Might have to look into that, thank you for mentioning.
2 comments

(Disclaimer: anecdata follows)

The Mastodon instance I use is hosted at Scaleway. It's had multiple multi-day outages due to failures at Scaleway's end.

Cheap is great and all, but this experience has not been the kind that would lead me to recommend using Scaleway for any service you need to be reliable.

They (Scaleway) are ARM machines, no?
Depends which offer you are looking at. They also have x86 VPS options using Atom chips.
You can still run mail servers on them though. You might have to compile stuff up for yourself if your distro doesn't support it but I've found Arch to be pretty good in this respect.
That's all fine, but then you're one hardware failure away from not being able to receive email.

I'm working on a guide on how to setup a replicated fault tolerant email cluster (galera/dsync) here [1] -- feedback appreciated.

This costs far more than something like fastmail, however. Depending on your situation you might value cost over peace of mind.

1: https://medium.com/@cyberpunk_networks/nsa-proof-your-email-...

> galera

Unfortunately, with Galera, you're not even a hardware failure away from losing / corrupting email.

https://aphyr.com/posts/327-jepsen-mariadb-galera-cluster

> Unfortunately, even in totally healthy clusters, with no node failures or network failures, Galera Cluster does not satisfy its claims of Snapshot Isolation.

In this scenario the galera DB is only hosting virtual aliases/account info etc -- it's essentially read-only unless I'm adding accounts so the fail cases are less severe than if I was constantly writing data.

I'll add a section on recovery, but even with a few hundred thousand email inboxes my database size is likely to be under 100mb and backing that up will cause no issues in live since I'm only doing SELECT 99.99% of the time; each of my nodes dumps the db into /var somewhere every hour.