Server4You Review
-
Let's take a real world example.... you have Redis running in a Docker container. It needs to talk to Redis instances around the world. As well as its Sentinel services need to do this. How do you expose them?
-
Most of the data that changes would be stored on a volume on the host also. Just for a quick example the /etc/nginx/conf.d folder would be stored in say /var/lib/nginx or whatever folder you create. That way you just add a conf file there and the container reads it. This keeps you from needing to access containers via SSH and keeping them light.
-
@johnhooks said:
Most of the data that changes would be stored on a volume on the host also. Just for a quick example the /etc/nginx/conf.d folder would be stored in say /var/lib/nginx or whatever folder you create. That way you just add a conf file there and the container reads it. This keeps you from needing to access containers via SSH and keeping them light.
Why are you talking about SSH? You don't consume any services normally over SSH other than SSH Itself. But in that case, say you are making SSH Proxies, how are you going to do this without exposing SSH?
-
Then you just attach the ports. That was the whole point of the original post. You dont have multiple ip addresses, either you use something like nginx to reverse proxy or use hardcoded ports.
-
@scottalanmiller You could expose those services via simple port forwarding.
ie: MyServerHost.My.Domain:5539 could point to my Redis instance running inside of a docker instance on port 5678(port numbers pulled from magic hat).
-
@johnhooks said:
Then you just attach the ports. That was the whole point of the original post. You dont have multiple ip addresses, either you use something like nginx to reverse proxy or use hardcoded ports.
Okay, so you are adding PAT in front of it and doing something really messy. Yes, that will work and something like HA-Proxy is probably best. But doing odd ports is sloppy and having to have PAT in front of your hosted Docker instances is messy and how do you handle graceful scaling? You can but it becomes much more complicated.
-
@johnhooks said:
Then you just attach the ports.
You can do that without NGinx too, but in the case of HTTP we can sense how messy that is.
-
Right. I'm only planning to use NGinX for the Web interfaces in each of my docklets (Is that even the right term? lol).
so my NGinX will reverse proxy for those.
In the event that I need to communicate between docklets, then I will simply attach the services to various ports and connect that way. IE: MySQL could be on port 3306 for my main instance, and 5306 for a Wordpress Docklet, and 9958 for a PGSQL Docklet... or whatever else I decide to set up.
-
@dafyre said:
Right. I'm only planning to use NGinX for the Web interfaces in each of my docklets (Is that even the right term? lol).
so my NGinX will reverse proxy for those.
In the event that I need to communicate between docklets, then I will simply attach the services to various ports and connect that way. IE: MySQL could be on port 3306 for my main instance, and 5306 for a Wordpress Docklet, and 9958 for a PGSQL Docklet... or whatever else I decide to set up.
Ya its cool. You don't even needto manage any of that. Just type
docker run --name some-app --link some-mysql:mysql -dand it links them together. -
I didn't mean to make this into a giant discussion. I agree that PAT is kind of clumsy, but it's how they have it documented. So if you need a specific service from a container to the outside world you do it with ports. Inter-container communication is either done by linking the containers together. Controlling a container is either done by connecting to the container from the host and giving commands directly, by creating a throwaway container which links to the original container, or by using a shared volume on the host. This is all done behind 1 ip address.
With CoreOS you can link multiple hosts together with etcd and then have certain containers on certain hosts, but that's a whole different conversation.
-
@johnhooks said:
I didn't mean to make this into a giant discussion. I agree that PAT is kind of clumsy, but it's how they have it documented. So if you need a specific service from a container to the outside world you do it with ports.
All communication is with ports
It's ports sharing a single IP that I've not seen done anywhere. Not that you can't, but it is very clumsy having to manage ports in that way for all systems. -
@scottalanmiller said:
@johnhooks said:
I didn't mean to make this into a giant discussion. I agree that PAT is kind of clumsy, but it's how they have it documented. So if you need a specific service from a container to the outside world you do it with ports.
All communication is with ports
It's ports sharing a single IP that I've not seen done anywhere. Not that you can't, but it is very clumsy having to manage ports in that way for all systems.I could be 100% wrong, but I think that's why it's been more of a dev tool and not exploded in the production area. However, with CoreOS and etcd that might be different.
-
@johnhooks said:
I could be 100% wrong, but I think that's why it's been more of a dev tool and not exploded in the production area. However, with CoreOS and etcd that might be different.
Docker is very much a production tool. I just left Change.org where it is being used for production. Lots of devs use it, of course, but Docker is not being produced for development, it is for production.
-
@scottalanmiller said:
@johnhooks said:
I could be 100% wrong, but I think that's why it's been more of a dev tool and not exploded in the production area. However, with CoreOS and etcd that might be different.
Docker is very much a production tool. I just left Change.org where it is being used for production. Lots of devs use it, of course, but Docker is not being produced for development, it is for production.
So how do they handle the port issue?
-
Containerization was developed by Sun (not Oracle) and has been the only way for deploying Solaris for a decade now. Linux has had product containers for almost as long.
-
@johnhooks said:
So how do they handle the port issue?
They don't run Docker on a single IP address VM
No different than how you host any VM, you get one IP per machine. -
Oh ok, makes sense.
-
I think that the average business using Docker is doing so on private IP ranges. And anything exposed to the outside world is going through load balancers or connecting to a proxy service like CloudFlare Enterprise.
-
@scottalanmiller said:
I think that the average business using Docker is doing so on private IP ranges. Any anything exposed to the outside world is going through load balancers or connecting to a proxy service like CloudFlare Enterprise.
Oh ok that makes sense.
-
@johnhooks said:
I didn't mean to make this into a giant discussion.
Sure, sure, I believe you!
But without giant discussions nobody learns anything.If I were to use docker, on my home network, then I would use my own DHCP server and let each docklet (please... somebody correct me if I'm wrong... or tell me if I just created a new buzzaord, lol) get it's own IP address.
However, in the sense that I am running this from a VPS provider, I will have only a single IP address (right now) to run things from.
This does bring me to another question though... are Docker containers migratable like VMs? ie: Can I move this docklet from Server A to Server B ?
