We have decided to reverse the direction of 3rd party integrations so this epic is mostly de-scoped.

Rejecting tasks , , , , and .

For future reference, the work done so far can be found in:

• https://github.com/mendersoftware/mender/pull/888

• https://github.com/mendersoftware/mender/pull/878 (review not completed)

• Complementary PRs to the above listed here (not reviewed)

We want it to be a drop-in solution which doesn't require any changes to the clients. If we switched to UNIX sockets, all of mender-connect, mender-configure, and mender-monitor would have to change. Besides, it's still speaking HTTP, and we plan to extend it to HTTP proxying later on, so using a network port seems like a natural choice.

There are some advantages to UNIX sockets, such as better control over who can access it. But this isn't needed, because we already use access control on the DBus endpoint to get the JWT token, and without that you cannot do anything with the localhost interface, even if you technically can access the port.

I was not into this a lot, but I have this comment: isn't UNIX socket a better interface than localhost?

Planning done!

I have gone for extremely limited documentation. This is very advanced functionality, which requires implementing whole daemons from scratch. Therefore the only documentation I have included is that you can disable the built-in daemon in . Beyond that, I think we should get comfortable with writing IoT integrations ourselves before we try to steer anyone else in that direction.

Thank you, indeed it seems like a good idea to start with this next sprint, so let's aim for that!