On Wed, Aug 19, 2015 at 3:04 AM, Bill Nottingham <notting(a)splat.cc> wrote:
Peter Robinson (pbrobinson(a)gmail.com) said:
> I think atomic is an excellent use case for that style of updates and
> with decent testing would even provide decent rolling style of update
> between releases with the ability to do rollbacks too with one boot
> type of functionality (update, set watchdog, reboot, test connectivity
> and core functionality, unset one-boot flag for rollback or if tests
> fail/watchdog triggers).
> The first two items are covered to some degree by atomic, the later
> would need some form of push management platform. I've not looked
> closely at feedhenry bits as I don't believe they've been opensourced
> yet, or I missed the announcement but there could be building blocks
Definitely fits way way better for the use case than our traditional model.
Although given that each IoT thingamajig is likely going to want a level of
customization, that means either concentrating more on the container build
aspect for the customized bits, or if they need heavier customization,
concentrating more on the atomic-producing tools rather than the One True
Atomic IoT tree.
Completely agree, I was thinking something akin to docker layered
images on top of the base. But you also have IoT gateways, and the
various other components of IoT that aren't thingamajig endpoints.