I’ve just arrived from Dockercon EU 2015 and it has been a really interesting thematic conference. This is the first post of a series of two posts, in which I will share some picks and links of the talks I have attended to.
The General session was performed by Ben Golub (@golubbe), Solomon Hykes (@solomonstre) and Aanand Prasad (@aanandprasad), among others. It gave a high-level view about news in the Docker ecosystem. It also set the general theme of the conference, based upon three concepts:
- Empower the makers
- Docker in production
- End-to-end solutions
They described the docker stack, which is comprised of solutions, dev tools, infrastructure and standards, with fantastic cartoon-style designs (Docker designers really rock!).
Docker Compose supports all new Swarm / Engine features. Thus, this tool is the one officially supported to define and manage your infrastructure, from dev environments, to production clusters
Kitematic is part of the Docker tools as a simple GUI for packaged applications
The Docker Content Trust and Docker Notary (both available in the Docker Trusted Registry) allow secure and usable content distribution for developers, allowing survivable key compromise, proof of origin and protection against untrusted transports. And now, the Yubikey can be used to sign images before pushing them to the registry. Cool!
There is also a vulnerability scanner for containers, called Project Nautilus, that has been running in the official repos since 2 months ago, and that eventually will be released for the public as a self-service. It uses deep content analysis of the images and matches against a database of known code vulnerabilities.
Docker Swarm 1.0 is out. It has multi-host newtworking and persistent storage out-of-the-box, based upon the new plugin architecture for both features.
To close this session, Andrea Luzzardi (@aluzzardi) perfomed a live demo with a cluster of 1000 dockerhosts, spinning up 50000 containers, managing swarm with
docker-compose commands. Impressive!
The talk The latest in Docker Engine, by Jessie Frazelle(@frazelledazzell) and Arnaud Porterie (@icecrime) was loaded with meaty information about version 1.9:
docker networkcommand, explained with an example of an Elasticsearch container linked against a Kibana container, showing some stats of the Docker project. Multihost networking is out of experimental, and extensible via plugins
Minor improvements and build-time arguments (
ARGDockerfile instruction, custom STOP signal
docker volumecommand, also extensible via plugins
GID/UID remap (root is not the host root anymore!) and storage scoped by GID/UID
The also talked about the next features in the pipeline:
Distribution rework: multiarchitecture images (powered by the so called fat manifests)
Official ARM support
Windows Server 2016 support
IBM PowerSystems, IBMz Systems, Solaris support
Support to seccomp for syscall filtering. Jessie Frazzelle made a demo based on a proof of concept she wrote.
Default to Docker Content Trust
Splitting the tools among runtime (RunC) and builder, allowing client-side builds
Maybe some convergence of Swarm and Engine (they have some duplicated functionality)
The talk Docker orchestration at production scale, by Andrea Luzzardi and Victor Vieux (@vieux), explained the details about the Swarm scaling demo of the General Session and discussed the new plugin-based networking mode that ships with two networking drivers out-of-the-box: bridge (for single host) and overlay (for multi-host) and the new plugin-based volume management that ships with local bind mount. Networking plugins (Calico, Weave, …) and volume plugins (glusterfs, sshfs, keywhiz) are available (or almost available).
Nathan McCauley (@nathanmccauley) talked about Understanding Docker Security:
He explained that the current model of isolation is based on namespacing (PID, mount, IPC and network), cgroups, limiting capabilities and third-party systems as SELinux and AppArmor. The work in progress in this fields is: allowing user namespaces and enabling seccomp, a granular syscal whitelist/blacklists.
He detailed the Docker Content Trust concept, that consists on a Notary, that translates image names to a content address, and a content addressable registry, that allows self-verifiable pulls because the name of the content is the hash of the content (something like a Merkle tree)
About Project Nautilus, he told that it consists on a filesystem walker, with some heuristics to detect code, and a match agains a database of known vulnerable code
Regarding, authentication and authorizations, there are pull requests for third-party systems, beyond the Unix Domain Socket and the Client Certificate, such as Kerberos, SASL, PAM, LDAP/AD, unix users…
Dave Tucker (@dave_tucker) and Jana Radhakrishnam gave a fantastic talk about Docker networking deep dive. They explained both the bridge and overlay model of networking. The latter is a very interesting one, based on VxLAN, with a centralized key-value storage (by default token-based, but swappable with Consul or other) and SERF protocol in order to exchange information among the cluster.
They also explained the roadmap of networking:
Better IPv6 support: working by default out-of-the-box, address allocation and support for AAAA records
DNS-based service discovery
Encryption of the overlay network
Official proxy container for tying networks together
And that’s all for the first day!