Major Roadblocks - real life use cases

James Beedy jamesbeedy at gmail.com
Tue May 29 05:23:21 UTC 2018


I want to shed some light on a few blockers for me right now.

1) I am having a hard time actually using Hadoop charms for Hadoop deploys on metal due to 2 reasons:
        * Maas support for 3rd party drivers a pain
        * Hadoop charms don’t have storage support

2) I am having difficulty deploying Ceph and Neutron due to:
        * maas support for 3rd party drivers a pain

Here are my two asks, hopefully you will find them as pertinent as I do.

1) Hadoop charms need storage support
        * hdfs
        * cephfs

Currently without support for storage, Hadoop slaves just use a local directory on / for HDFS.

This is entirely illegitimate for any real batch processing use case using HDFS (something I’m trying to do right now but can’t use juju Hadoop to do because of the lack of support for storage).

2) Maas needs better support for 3rd party drivers.
      * Getting my Mellanox drivers hooked up at commissioning so maas recognizes the 40Gb interfaces is taking me a few weeks now. Supporting user defined 3rd party driver should be a primary supported capability of MAAS.
      * how are people doing Hadoop,and Ceph and Openstack without this, possibly someone knows something I don’t know here?

Any feedback would be greatly appreciated.

Thanks





More information about the Juju mailing list