Hi All, I’ve drafted a summary of the Mesos townhall that took place at TwitterHQ on 11/19. Approximately 70 people attended the event, and earlier that same day OpenTable hosted a Mesos hackathon with 20 attendees.
The event format featured lightning talks (4-minute presentations), and unconference sessions that were proposed by attendees and ran concurrently. Lightning talks included presentations about the new Slave Recovery feature, Elastic Mesos, the Mesos CLI, launching frameworks on Mesos, and Apache Aurora. Video of the lighting talks will be posted online at a later date (I’ll be sure to share links on the user list). Session Summaries At the conclusion of the townhall, the group reconvened to brief us on their discussions and any next steps. Below I’ve shared the notes I was able to write down; others in attendance should feel free to chime-in and help fill in the gaps. Resource Isolation and Containers The group discussed Docker specifically, including its packaging capabilities and how Mesos could leverage that and running it in Mesos. The group also discussed using a pluggable isolator, possibly KVM or Docker? How could you make it easier? Next steps were a combination of online conversations on the mailing list ( d...@mesos.apache.org) and possibly setting up an in-person hackathon to collaborate moving forward. Apache Aurora The Apache Aurora <http://aurora.incubator.apache.org> team released their codebase earlier in the day, and gave a lightning talk that demonstrated the scheduler. Much of the unconference discussion was centered around the capabilities of the project, and questions related to how Twitter uses it in production. There was interest in using it for long-running services as well as cron. Next steps include making progress on the open source project (now part of the Apache Incubator) and improving the project documentation. Future of the Mesos API This session was proposed to discuss the scheduler and executer apis, but the conversation quickly moved to discuss capabilities and feature requests. For example, there was interest in changing the Mesos UI to be centered around tasks instead of frameworks (which is something already being worked on). There was also discussion from the operator side of Mesos, such as host drain, interacting with the cluster when doing kernel upgrades, etc. Requested features included making making service discovery be more of a generic thing so frameworks could leverage it, even if it’s slightly different between them. Also, there was interest in exposing more information from slaves back to masters. As next steps, community members were encouraged to submit new JIRA issues to track these feature requests and future work. Mesos on Amazon Web Services and Autoscaling Capabilities Discussion centered around the production needs of running Mesos on AWS, particularly as it relates to autoscaling. The goal of this potential feature would be to grow and shrink the size of a cluster on AWS. There were questions about the best way to implement this, and whether feature could that become integrated into the Mesos core. Next steps: Netflix has a lot of experience in auto scaling AWS, and plans to work on this feature for Mesos in the next few quarters. Community members were encouraged to talk to them as they begin. Packaging Discussion was about packaging and distributing Mesos, and how to push this forward in the future. There was a shared interest in seeing official Mesos packages distributed in the future. Unfortunately, some key members of the community who have worked on packaging were unable attend to attend the townhall. Next steps may include regrouping on the mailing list, or possibly setting up regular meetings to push this forward.