What do you think about twitter's Aurora framework? Do you think Twitter will release some complementary tools like Mesosphere made?
I think Aurora is a good tool, but somewhat specific to Twitter's use case. I mention it a bit in chapter nine, but I see Aurora as being a bit more like Kubernetes to be honest. If I recall correctly, Aurora wants to be the only Mesos framework running on the cluster, and it provides primitives to developers and operators for running the cluster and deploying applications. I don't know how well it will play with other schedulers like Spark and Marathon, but in my experience Aurora tends to hold on to all resource offers and expects to have a world view of the cluster, so that things like task preemption and maintenance mode can function.
Having said that, it's entirely possible that Aurora fits your use case over Marathon.
You should try out both and see which one makes the most sense for your team (both developers and operators).
Because Aurora was developed at Twitter, it already has support for working with things like Finagle, which is open source. You might want to check out linkerd (
https://linkerd.io) which was built by some former Twitter engineers on top of Finagle. It's actually a really great tool!
And did you saw one of the latest Netflix presentation which shows their work to produce their internal new framework which works with containers and also deal with Amazon Security Groups and so on? Any thoughts about it?
Are you referring to Fenzo? (
https://github.com/Netflix/Fenzo)
Also , do you think the best framework and tools to go today would be DCOS as it is open source now?
Disclaimer: I'm a Mesosphere employee.
Since Mesos is the kernel, tooling is required in order to use it in production. Just like you can't run the Linux kernel by itself (you need an operating system like Ubuntu or RHEL), you need tools (and essentially, an operating system) to run Mesos. DC/OS aims to be that Mesos-based operating system, but across multiple machines instead of just one.
I spoke with someone from a rather large company at MesosCon this year who has taken individual components of DC/OS and used them with other components not included in DC/OS to build a Mesos-based cluster for his company. There are pros and cons to that approach though that need careful consideration:
Pro: you can pick all the best components for your stack, and you have intimate knowledge of how they all work togetherCon: you now have to maintain a stack, including testing compatiblity changes when you make changes to one or more components. Upgrades might be trickier depending on the number of engineers on staff.
If you're looking for something to get up and running quickly, and want all of your software tested together and know that releases will work, DC/OS is probably the way to go. If you want control over each component in the stack, including the option of when to upgrade it, at the expense of needing to maintain this stack, then Mesos + DC/OS components (mesos-dns, minuteman, etc) might be the way to go. Honestly, it all depends on your comfort level, team/staff, and timeframe. I'm a huge fan of experimentation though, so if you have the time, try out each in a development environment!
https://github.com/mesosphere/playa-mesoshttps://github.com/dcos/dcos-vagrant