>>>>> "Emmanuel" == Emmanuel Kasper <emman...@libera.cc> writes:
Emmanuel> Le 16/08/2016 à 17:56, Sam Hartman a écrit :
>>
>> Please note the follow-up to the cloud list.
Emmanuel> Hi Sam
Emmanuel> Thank your for this very thorough review of image creation
Emmanuel> tools. You made me google for "object oriented shell
Emmanuel> programming" ...
Emmanuel> But ... could you maybe also have a looker at packer ? (
Emmanuel> https://packages.debian.org/stretch/packer ) It can create
Emmanuel> image for various virtualization platforms and cloud
Emmanuel> providers.
Without prejudice to packer, it's not a tool I'd use. I do agree that
we should evaluate it here, but I'm not the best person to do that
evaluation. I'd recommend that someone who favors the approach do that
work.
For me, my concerns are:
1) I think involving d-i is the wrong approach for building images. The
main complexity of d-i over debootstrap seems to be able to deal with
per-system variation and configuration. Perhaps it would be nice to
share some of the partitioning infrastructure that d-i has, but for me,
when building an image, the rest of d-i adds complexity and possible
unwanted variation. As someone who cares about customization,
customization with d-i will be even more difficult than with
bootstrap-vz.
2) For me, the lack of shared libraries in Go is a deal-killer for
infrastructure tools I'm going to depend on.
I want to stress that these are reasons why I personally am not very
interested in packer, and should only be taken as my opinion in the
broader discussion. I think someone should do the analysis of packer.
As we're discussing how to approach what we do as official images, I'll
inject at least the first concern above into the discussion as my
opinion about why I don't think a d-i based solution is the way to go.
I would eagerly read a packer evaluation written by an advocate with an
open mind and consider that I may be wrong.
--Sam
