This is actually really big news for anyone running or interested in running a Docker-based PaaS system such as Deis or Flynn. DigitalOcean's cheap instances are a great match for Docker containers.
As of Deis 0.8.0 it only runs on CoreOS, and I believe most other DIY PaaS systems are moving the same way.
IMO Docker + etcd is a far more sane configuration than endless Ruby Chef scripts, or worse, Amazon OpsWorks.
its not about its shortcomings, its a database. your saying it replaces chef and opsworks. you're missing a few pieces to make that picture.
ie. how do i update the load balancer as i add capacity to my web app, how do i setup a new db instance and open up the firewall rules to an app on a different host. etc.
docker, and etcd are both nice pieces, but they don't make a complete picture. nor does fleet imo.
Don't forget Ansible; out of the box, it can perform pretty much any of the glue operations necessary to manage a set of containers and servers.
It's still pretty early, so much of the tooling will need more improvement before container-based infrastructure gets more mainstream (e.g. enterprise, smaller teams) uptake.
While I agree that etcd and config management systems do not solve the same problems, and esp the same way, the use of etcd by all of your services does enable your examples.
For instance, there is a tutorial on using vulcand as your http load balancer, which is driven by configuration in etcd. A tutorial on the CoreOS site shows you how to use this to cleanly deploy a new version of an app as new versions of a container, and to rotate them into the LB and to rotate the old ones out.
You can also connect a script that controls firewall rules based on etcd, its' changes would be reflected almost immediately, rather than the splayed 30-60min period of seemingly-randomly-applying-changes you typically see with config runs of tools like chef and puppet.
I've been trying to figure out where tools like chef, fabric, and docker/coreos/etcd/fleet will play together, what the boundaries will be, etc.. In situations where I'm using Docker and CoreOS, I don't expect to use chef, but I'm not sure it will be used for things like database servers, which we typically dedicate and tune hosts for.
It would be nice to see something like chef running on CoreOS for, say, user management so that people who want to talk to, say, fleet aren't required to all ssh as 'core'.
etcd is better compared to zookeeper (SOA orchestration, service discovery, etc) and etcd (after researching for the last two weeks for a production environment) is nowhere near as stable (ie non-beta/alpha) as zookeeper or Netflix's Eureka.
Configuration management (puppet, chef, salt, ansible) is a completely different beats than service discovery, health management, etc.
agreed, etcd is like zookeeper, their both databases that you write applications against. No! neither does service discovery and orchestration out of the box. The applications you write on top of them do.
Right, and you can achieve service discovery, orchestration, and convergence driven by something like zookeeper or etcd, rather than by bulky config runs.
While I enjoy working with Chef, have had some pretty reasonable times with Puppet, these full-run tools do have issues sometimes where you introduce a narrow bug in your user management or some other code, and all of a sudden you can't update some random conf file that happens after it. I've also seen tools like capistrano and fabric bastardized to allow this sort of precise updating, but lose the cohesion of typical config-managed systems.
With something like etcd, certain problems are solved by not relying on static configuration. That is the paradigm shift, and it is similar to Hadoop. When I've built Hadoop systems with puppet, we simply wrote out the same configs and files on every machine, then chose which services to start, and via zookeeper things like primary / secondary failover take care of themselves.
Another is that your configuration is wide open to every node in the cluster (e.g. your Web app's etcd client can read the key that stores the master database user password). Hope you never get hacked!
IMO Docker + etcd is a far more sane configuration than endless Ruby Chef scripts, or worse, Amazon OpsWorks.
For what use case? Doesn't using Docker for everything take you into 'golden master image' dead-ends, as outlined by the Opscode guys in comments to a Docker blog post last year [1].
What Lamont and Joshua had to say in that thread resonated, but I haven't really looked into DevOps approaches with Docker.
I'm also not sure what you mean by 'endless Ruby Chef scripts'?
I'm really out of touch with Docker and CoreOS, so please forgive my ignorance if this is a ridiculous Q: could this combo be used for spinning up machines with specific apps for thin clients? Or is it more about scaling one app?
Essentially it lets you run 20 super-lightweight VMs on a single machine... Each one of those could be running different apps, and Docker makes it really easy to build a custom machine image using a single script (called a Dockerfile) using any standard linux OS as a base.
So yes, powering 20 thin clients running different apps from a single server is a perfect use case.
What about isolation between apps? I have app A connected to database A, and app B connected to database B. Is there a way to deny connections from app A to database B etc?
What I want is to make groups of containers that can talk only to each other(only to containers within one group). Does CoreOS provide something like that? Maybe kubernetes? What are possible options?
Docker has a sophisticated system for controlling what ports are open on each container and which other containers it can "see" when it uses these ports.
I still am very impressed with CoreOS but I find sad that they were forced to move away from their initially touted "only systemd+etcd". (No blame on them, that was expected as the Docker ecosystem is still young and very fast-moving; plenty of interleaved problems must be solved elegantly.) When you say that Docker+etcd is saner that other solutions, I wonder for how long this will remain the case as the field matures.
Well "Docker + etcd" still needs a surprising amount of tooling around it to be considered a viable production environment. CoreOS provides a lot of this and deis sits atop that.
Deis is the best thing I've seen come out of Docker's DevOps gold rush.
Indeed... Great news for everyone working with DIY PaaS and 12factor apps. Smart move from both, DO and CoreOS, but will also benefit projects like Deis and its users.
Comments
This is actually really big news for anyone running or interested in running a Docker-based PaaS system such as Deis or Flynn. DigitalOcean's cheap instances are a great match for Docker containers.
As of Deis 0.8.0 it only runs on CoreOS, and I believe most other DIY PaaS systems are moving the same way.
IMO Docker + etcd is a far more sane configuration than endless Ruby Chef scripts, or worse, Amazon OpsWorks.
etcd is not a magical replacement for a configuration system.
Just curious - what do you see as its shortcomings?
its not about its shortcomings, its a database. your saying it replaces chef and opsworks. you're missing a few pieces to make that picture.
ie. how do i update the load balancer as i add capacity to my web app, how do i setup a new db instance and open up the firewall rules to an app on a different host. etc.
docker, and etcd are both nice pieces, but they don't make a complete picture. nor does fleet imo.
Don't forget Ansible; out of the box, it can perform pretty much any of the glue operations necessary to manage a set of containers and servers.
It's still pretty early, so much of the tooling will need more improvement before container-based infrastructure gets more mainstream (e.g. enterprise, smaller teams) uptake.
While I agree that etcd and config management systems do not solve the same problems, and esp the same way, the use of etcd by all of your services does enable your examples.
For instance, there is a tutorial on using vulcand as your http load balancer, which is driven by configuration in etcd. A tutorial on the CoreOS site shows you how to use this to cleanly deploy a new version of an app as new versions of a container, and to rotate them into the LB and to rotate the old ones out.
You can also connect a script that controls firewall rules based on etcd, its' changes would be reflected almost immediately, rather than the splayed 30-60min period of seemingly-randomly-applying-changes you typically see with config runs of tools like chef and puppet.
I've been trying to figure out where tools like chef, fabric, and docker/coreos/etcd/fleet will play together, what the boundaries will be, etc.. In situations where I'm using Docker and CoreOS, I don't expect to use chef, but I'm not sure it will be used for things like database servers, which we typically dedicate and tune hosts for.
It would be nice to see something like chef running on CoreOS for, say, user management so that people who want to talk to, say, fleet aren't required to all ssh as 'core'.
etcd is better compared to zookeeper (SOA orchestration, service discovery, etc) and etcd (after researching for the last two weeks for a production environment) is nowhere near as stable (ie non-beta/alpha) as zookeeper or Netflix's Eureka.
Configuration management (puppet, chef, salt, ansible) is a completely different beats than service discovery, health management, etc.
agreed, etcd is like zookeeper, their both databases that you write applications against. No! neither does service discovery and orchestration out of the box. The applications you write on top of them do.
Right, and you can achieve service discovery, orchestration, and convergence driven by something like zookeeper or etcd, rather than by bulky config runs.
While I enjoy working with Chef, have had some pretty reasonable times with Puppet, these full-run tools do have issues sometimes where you introduce a narrow bug in your user management or some other code, and all of a sudden you can't update some random conf file that happens after it. I've also seen tools like capistrano and fabric bastardized to allow this sort of precise updating, but lose the cohesion of typical config-managed systems.
With something like etcd, certain problems are solved by not relying on static configuration. That is the paradigm shift, and it is similar to Hadoop. When I've built Hadoop systems with puppet, we simply wrote out the same configs and files on every machine, then chose which services to start, and via zookeeper things like primary / secondary failover take care of themselves.
+1
If you want to do orchestration of both the infra and the app, you need things like opsworks.
One simple solution would be to add etcd configuration items as a CloudFormation/Heat custom resource type.
For one thing, it's not production ready yet: https://github.com/coreos/etcd/blob/769c043537263dd5701f5254...
Another is that your configuration is wide open to every node in the cluster (e.g. your Web app's etcd client can read the key that stores the master database user password). Hope you never get hacked!
For what use case? Doesn't using Docker for everything take you into 'golden master image' dead-ends, as outlined by the Opscode guys in comments to a Docker blog post last year [1].
What Lamont and Joshua had to say in that thread resonated, but I haven't really looked into DevOps approaches with Docker.
I'm also not sure what you mean by 'endless Ruby Chef scripts'?
[1] http://blog.relateiq.com/why-docker-why-not-chef/#comment-43...
I'm really out of touch with Docker and CoreOS, so please forgive my ignorance if this is a ridiculous Q: could this combo be used for spinning up machines with specific apps for thin clients? Or is it more about scaling one app?
Essentially it lets you run 20 super-lightweight VMs on a single machine... Each one of those could be running different apps, and Docker makes it really easy to build a custom machine image using a single script (called a Dockerfile) using any standard linux OS as a base.
So yes, powering 20 thin clients running different apps from a single server is a perfect use case.
What about isolation between apps? I have app A connected to database A, and app B connected to database B. Is there a way to deny connections from app A to database B etc?
What I want is to make groups of containers that can talk only to each other(only to containers within one group). Does CoreOS provide something like that? Maybe kubernetes? What are possible options?
Docker has a sophisticated system for controlling what ports are open on each container and which other containers it can "see" when it uses these ports.
Great, thanks for that! Not a sysadmin by trade so didn't know whether I'd misinterpreted how it worked.
I have another noob question. Would this be a good way to get redundancy for serving a website?
I still am very impressed with CoreOS but I find sad that they were forced to move away from their initially touted "only systemd+etcd". (No blame on them, that was expected as the Docker ecosystem is still young and very fast-moving; plenty of interleaved problems must be solved elegantly.) When you say that Docker+etcd is saner that other solutions, I wonder for how long this will remain the case as the field matures.
Well "Docker + etcd" still needs a surprising amount of tooling around it to be considered a viable production environment. CoreOS provides a lot of this and deis sits atop that.
Deis is the best thing I've seen come out of Docker's DevOps gold rush.
"Deis is the best thing I've seen come out of Docker's DevOps gold rush."
I have to agree - Deis fills a lot of holes, although it still has some way to go.
Would you be willing to share more about what's in CoreOS that you feel like shouldn't be?
Indeed... Great news for everyone working with DIY PaaS and 12factor apps. Smart move from both, DO and CoreOS, but will also benefit projects like Deis and its users.