I don't think the time behind the scenes will ever be zero so I'd rather 100% be guaranteed not to lose my IP address. Spin up a new VM and assign prod-new the IP of prod-old similar to how you might do it with AWS Elastic IP
I do agree with this. Adding storage to an existing droplet would be nice. Even if it required adding a separate "disk", this would be useful. But obviously, adding, for example, 20GB disk after disk after disk could lead to a lot of annoyance.
What if you start out small with one server, but then need to expand to multiple droplets? You won't be able to without powering off and making an image.
That's bad planning, but unfortunately also a very common situation. If you really can't afford a couple late hours downtime when you're small then this is what I think I'd do:
- Get a new pair of droplets in front of your current droplet to load balance the traffic (HAProxy, Nginx, you name it). Point the NS records to those :)
+ Why two? Well, you can't afford downtime so there we go.
- Get a new droplet and manually clone the initial one, i.e. copy over config and assets. Add this second droplet to the LB pool for your site/app.
+ Manual work? Well, again you can't afford downtime so you'll need to put some man-hours on this.
- Now you can shut down the initial droplet and image it while the second one gets the traffic coming from the LB.
Eventually you'd want to separate functionality on different servers, so try to also plan that in advance. Also bear in mind that each 512MB costs like $5/mo, that makes this solution work for $20/mo, how cool is that?
I agree there's room for improvement on DO's service, but there's nothing that a good architect can't work around.
You're absolutely right. How about your MySQL write server? You typically have one master that you update to and one or more slaves where you read from. If you get a sudden burst of traffic and your master can't handle all the writes, what do you do?
I understand how HTTP works so I know how to scale it, but unfortunately I don't have the knowledge to scale DB infrastructure.
Anything I say would be a shot in the dark, but I'm interested on your point of view about this one. Caching? Write on two servers? In the second case do you setup a cluster or you just load balance it with a third device/service?
I'm still learning this myself, but from the research I've done I've figured this out:
1. Most people don't go with two writes to a relational database (like MySQL). Syncing is hard, and when it is done, it loses its relational features.
2. Common set up is one mysql write server, scaling it up to fit your traffic, and automatically syncing it with additional servers where SELECT statements only will be made. A random read server is selected if there are multiple to choose from.
3. This also means that in the application, doing a write and immediate read on what you just wrote won't work, so you have to code your application around that.
4. In you app you separate your connections for the reads and writes.
Now, in terms of making changes to the write server is what I'm trying to figure out. Perhaps there's a way to make upgrades on one of the slaves (the reads), then automatically make that slave the master and the master a slave? Not sure, I'm reading a MySQL book (The MySQL Bible) right now that I hope will point me in the right direction.
and
https://digitalocean.uservoice.com/forums/136585-digitalocea...
I don't think the time behind the scenes will ever be zero so I'd rather 100% be guaranteed not to lose my IP address. Spin up a new VM and assign prod-new the IP of prod-old similar to how you might do it with AWS Elastic IP