Architecture for a WordPress Docker Project

General
1

Hello,

I have a project where we want to offer a WordPress install (on subdomains) per registered user. Will it be correct if in a swarm

  • for each user i will create a service (each with 2 containers - wordpress and mysql )

  • have a nginx as reverse proxy to route each sub domain to the right container .

  • maybe use AWS EFS for volumes (if is possible ) in order to have persistent data for images and mysql db

I have tried a similar approach with Aws Ecs but is very expensive . Any suggestions for this project architecture ?
Thank you

2

it is possible to create user specific docker stacks, which each describe the two (swarm) services (wordpress and mysql). In the Swarm world, Each service will be scheduled as a task from the scheduler. Then a task executes the container on a node.

Instead of nginx, I would strongly suggest to go with traefik. You can simply add reverse proxy rules using service labels. Whenever such a labeled service is starting, the rules are instantly applied in traefik. Another approach would be to pusblish each wordpress instance on a different port and leverage ELB to forward traffic for a specific sub domains to a specific target group. A target group can consist of one or more ec2 instances and ports.

Also I am not sure if EFS is realy the right kind of storage. From my experience EFS is blody slow and preferably used to store data at rest. When used as a target for docker volumes, it can quickly become a bottleneck.

Don’t forget to put additional machines into account for log management (ELK?) and system monitoring (e.g. Swarmprom).

3

I agree with @meyay, i have setup something similar to what you’re doing, just with traefic and multiple swarm stacks, 1 stack pr customer.

But personally, i think aws is overkill for this (because of price), thought about something like digitalocean ?

4

Thank you for your input.

I will definitely use traefik as the reverse proxy.

Regarding EFS - Is there any other solution to prevent data loss ? I’m thinking that one EBS/Hdd i will have many clients and a hardware failure could turn into a nightmare,

Regarding Digital Ocean - i will definitely take a look and see if is more affordable.

Thank you again.

5

if you plan to have more than a single node, you will need to solve the “how to make the volume available on all nodes?” problem. The default volume driver creates volumes that are always local to a node, which will lead to problems if a servce task is scheduled on a different node. Also, pinning a service to a specific node (using labels) breaks the idea of self healing and spreading the containers across the nodes.

You either need a remote share like nfs or a volume plugin that handles “the magic” for you. We ended up with a dedicated EC2 self managed NFSv4 server per cluster(!) and putting the volume declaration for our nfsv4 shares inside the compose.yml of each stack. as a result the local volume handle will be created on a node when the service is executed on a node the first time.

6

I used glusterfs between nodes