CNI for Docker Containers with Weave and Calico

CNI for Docker Containers with Weave and Calico(weave.works)

57 points by polvi 10 years ago | 19 comments

I've got a strong interest on overlay networking solutions like Weave, but I'm not an expert on Docker and other container solutions.

What's the new thing here? If I understood correctly, it seems that you can connect your Docker host to an overlay network, so your containers can access other containers and resources through it. Am I correct to think this facilitates orchestration of the containers' network?

Disclosure: I am behind https://wormhole.network which could be seen as some sort of Weave competitor, but it's not. It covers other use cases, even though there's some overlapping e.g. overlay multi-host networking for containers https://github.com/pjperez/docker-wormhole - it doesn't require changes on the host itself, but can't be orchestrated.

vishvananda 10 years ago | |

I haven't come across your project before. Any relation to my (similarly named) project https://github.com/vishvananda/wormhole ? I always thought that one of the most interesting piece of my project was easy ipsec tunnel setup. It turns out that setting ip ipsec tunnels is pretty tricky.

NetStrikeForce 10 years ago | | |

Holy cr*p! haha! first time I see your project, too.

I think both are a bit different. My project is based on SoftEther and using an external server as a pivotal point, so all the members of the network only need outbound 443/TCP access. It's not point to point unfortunately. The idea is to make sure it works on as many scenarios as possible.

I'm just adding all the server management and simplification layer, but both server and clients are 100% SoftEther.

justincormack 10 years ago | |

No, Docker already supports overlay networks, including Weave a default one that ships with it, and Calico.

shykes 10 years ago |

I haven't looked at this in detail, but does this work with the standard networking features introduced in Docker 1.9 [1] and 1.10 [2]? Can I still use 'docker network create/connect' and the DNS service discovery features of Docker? Can containers interoperate regardless of the choice of Docker plugin, or will they only work on a plugin based on the weave proxy? The wording in the post leaves that ambiguous.

[1] https://blog.docker.com/2015/11/docker-multi-host-networking...

[2] https://blog.docker.com/2016/02/docker-1-10/

vidarh 10 years ago | |

Well, sort of. Before the Docker networking features, Weave relied on a proxy that intercepts Docker API calls to set up their network before passing requests on to Docker. It was/is a workaround for lacking plugin support in Docker, and even after the network plugin support was added it still allows additional functionality that is hard to implement via the plugin mechanism but there is now a Docker network plugin for Weave too.

What they've done now if I've understood it correctly, is that they've effectively leveraged that to allow them to intercept Docker API calls, and if that call requests a network provided by a CNI plugin, they call CNI "on behalf of Docker" and then pass on a modified API call to Docker, so you can have Docker/Kubernetes/Rocket on the same overlay network.

> Can containers interoperate regardless of the choice of Docker plugin, or will they only work on a plugin based on the weave proxy?

Containers don't care what network you configure. Basically Docker will just use a bridge interface, assign IP addresses for a container on that bridge, and optionally expose ports on the host. The Docker networking support lets Docker query an external plugin API to obtain the details to use for a container. Kubernetes and Rocket implements a different plugin API for the same purpose. But in both all of this happens before a container is started.

Once it's started, the container just sees a an interface bound to a suitable IP, so your containers should not need to care.

bboreham 10 years ago | |

The networking features introduced in Docker are one "standard"; CNI is another "standard" [1].

Weave Net as a product ships with support for the former; the blog post describes work towards support for the latter.

No they do not interoperate; there is a detailed argument at http://blog.kubernetes.io/2016/01/why-Kubernetes-doesnt-use-...

[1] Scare-quotes because all of these things are only a few months old.

politician 10 years ago |

Does Weave still have atrocious network performance?

DanielDent 10 years ago | |

For now, yes. https://github.com/weaveworks/weave/issues/1925 https://github.com/weaveworks/weave/issues/1931

The developers are friendly, helpful and responsive. But in my testing, Weave simply wasn't adequate. I hope they'll get there.

bboreham 10 years ago | | |

I think you were running with encryption turned on. This is a unique feature of Weave Net, but has some costs.

Without encryption, Weave Net uses the same VXLAN technology as Docker Overlay; goes at close to native speed.

(note I am a friendly Weave Net developer)

__monadic 10 years ago | |

No. http://www.weave.works/weave-fast-datapath/

politician 10 years ago | | |

Nice work. VXLAN works well for cloud providers that can support it (all of the them afaik, except for Azure).

danfitch 10 years ago |

Video that goes into detail about what Calico. https://www.youtube.com/watch?list=PLlh6TqkU8kg_Eydfk1Nyt6iK...

chrissnell 10 years ago |

Other than IPv6 support, is there a reason to use Calico/Weave over Flannel? We've been very happy with Flannel, especially using the clean CoreOS-Flannel integration.

fons 10 years ago | |

In the case of Weave: encryption, multicast, fault tolerance (fully distributed through a CRDT mesh), integrates beautifully with containers (can be used as a Docker network plugin and provides out-of-the-box service discovery). More over, it's stupidly simple to use and setup (zero-conf, no kv-store required).

Full disclamer: I work at Weaveworks.

vetrom 10 years ago |

I'm looking forward to setting up some trial deployments of calico at my workplace.

While there is an ease of understanding of bridged and/or overlayed networks, native end-to-end routing between containers with regular IP datagrams and container-level addressability has been on my wishlist for a long time.

emeraldd 10 years ago |

Am I the only one wondering who Calico is? (I'm not finding an obvious candidate with a quick Google search ...)

rconti 10 years ago | |

Some of these headlines feel like mad libs for technology.

Occasionally it makes me feel left behind.

Then I remember most of these "new hotness" technologies get abandoned as quickly as they get adopted, so I'm not missing much.

emeraldd 10 years ago | |

Ahh ... had to search for "calico tom denham" to find http://www.projectcalico.org/ ok ... that makes more sense than a company dealing with biology ...