[dredmorbius]

Another lesson is that "multi-cloud" should mean "multi cloud-provider", if you're going to go that way.

And that self-hosting as a fallback (if only to test backup and recovery procedures) is also highly underappreciated.

[adrr]

Multi region at least doubles your infra costs. Multi cloud will probably at least triple your costs since things like terraform don’t transfer cleanly and you can’t use even hosted services like RDS/CloudSQL, EKS/GKE because you won’t be able to replicate it across cloud providers due to all vendor specific differences.

[dredmorbius]

Note that the case here was of free-tier usage.

Multi-provider setup may incur additional effort, though the payoff is in liberating yourself from lock-in to any one vendor. Note that those costs are being incurred here regardless, with both the costs of having to develop those on the fly and with downtime. Effectively, multi-provider set-up was a deferred cost for yewtu.be, now being realised.

One trick for using hosted services to to avoid the service-specific tooling of the hosts in question. Yes, that decreases the value-add of such services, but again, the trade-off is reduced lock-in.

There are also multi-platform solutions which stand as middleware between your own application and/or services and that of the host platform.

[adrr]

Then you incur costs around people who have specific knowledge around the technology. I can push 500k reads per sec on a RDS on instance which be a very challenging to do on on vanilla Postgres on EC2 instances. And when stuff breaks like broken file systems, I can just rely on their support. Same goes with Kubernetes at scale. EKS allows me to not to hire an ops/systems role.

[nucleardog]

If you just need durability not reliability it doesn’t need to be _nearly_ this expensive.

I run a bunch of stuff single region, single cloud, but I’ve got a regular job that takes the dumps of my databases and other user content and pushes it over into a Backblaze B2 account tied to a separate credit card.

At $0.005/gb/mo, every 100GB costs you about $0.50/mo in storage. Depending on how much data is changing, if it’s say 10GB/mo you’re looking at about $0.90/mo in AWS’s punishing egress charges.

For less than the cost of a basic McDonald’s cheeseburger every month a lot of services can add _durability_ which is more important in many cases. If your service has a lot of value people will wait for it to come back. If it’s never coming back, then you’re probably done. (I worked with one client who was entirely offline for a MONTH. People called CS every day begging them to fix it faster, but not a single one cancelled.)

Yeah, if AWS nukes my entire AWS organization it’s going to be a bad week getting everything set back up on $AnotherCloud. But it will be back up in a week, not lost forever along with my users’ trust.

[hnlmorg]

I do love the ideal of multi-cloud in theory but the practicalities aren’t so simple. It’s hard enough for businesses to pull it off, let alone someone who’s providing a free service.