2014-02-18 12:10 GMT+00:00 Tom Arnfeld <[email protected]>: > I’ve been working the past few days on building a Pluggable > Containerizer for Docker (I am aware the guys at Mesosphere > are already doing this, but i’ve not seen anything shared and > wanted to get started on testing mesos as I intend to use it). > I’d very much appreciate if anyone involved in the new > external/pluggable containerizer feature could take a few > minutes to have a look and/or test it out! > > It’s on github with a getting-started readme: > http://tfld.me/3m2m1v322w0y > > I’ve implemented it pretty blindly and the example > containerizers don’t help too much as they’re pretty basic. > Some documentation on what all of the different commands mean > would go a long way with others users that want to implement > their own containerizers, i’m sure. > > There are a few outstanding bits worth noting; > > - The `usage` stats don’t reflect the usage of the container > - I’m not sure how this works with custom executors hosted on hdfs:// or > s3n:// (needs testing and documenting) > - The `recover` method of the containerizer isn’t implemented > I’m not sure what this is for exactly… since the comments/docs are pretty > sparse > - The resource limits are not enforced on the container
We're opening up the code that we have today. https://github.com/mesosphere/medea There are some things we do already -- like resource limitations -- but we haven't finished standard error/out integration or filesystem browsing (probably to be handled via a read-only mount in to the sandbox). Medea is different in that it is packaged with the mesos.proto file and intended to be installable with setuptools/distutils. As the interface between the containerizer and Mesos is all protos over standard in/out, it is feasible to ship it as a true Python package. Please look forward to an implementation of `usage` later today. -- Jason Dusek pgp // solidsnack // C1EBC57DC55144F35460C8DF1FD4C6C1FED18A2B
