coreos: fleet: obsolete version(s) included by default, thus omitting fleet bug fixes

[megahall]

Description

A Fleet bug fix which affects use of fleetctl destroy in my deployment is available upstream in Fleet v0.12.0 and up, but the CoreOS Fleet is quite far behind the upstream Fleet.

Would it be possible either to backport the fix in coreos/fleet#1417 into Fleet v0.11.x or ship a standard train CoreOS which includes Fleet 0.12.x or up? Because no current version of CoreOS includes Fleet 0.12.x or up as of yet.

coreos/fleet#1383 (comment)
coreos/fleet#1417 (comment)

CoreOS Version

Issue applies to all published CoreOS releases up to and including the current latest, 1097.0.0. This can be confirmed by verifying the Fleet versions listed in all releases noted in the release notes at https://coreos.com/releases/ .

Environment

This is a base dependency issue which is independent of hardware, cloud, or hypervisor releases.

Expected Behavior

CoreOS should include more recent branches of upstream Fleet on a regular basis to prevent missing important bug fixes.

Actual Behavior

CoreOS does not include more recent branches of upstream Fleet on a regular basis to prevent missing important bug fixes.

Reproduction Steps

NOTE: See coreos/fleet#1383 for more details.

1. Run fleetctl destroy on a service that's not active for whatever reason.
2. Obsolete copies of Fleet print storms of Error destroying units: googleapi: Error 404: unit does not exist messages and return spurious POSIX process return codes.
3. Current copies of Fleet do not print the messages or return errors and just work.

Other Information

N/A.

[kayrus]

/cc @jonboulle @dongsupark

[crawford]

We aren't going to update the version of fleet in Container Linux anymore. We are going to sunset the project and remove it from the OS.

[codepunkt]

@crawford What does this mean exactly? We're setting up a CoreOS cluster at the moment. Should we continue using Fleet or rather switch to Kubernetes or some other solution? Is there a timeframe for this or an official announcement? What is the reasoning behind it?

[crawford]

Early on, we built fleet to help bridge the gap between single-instance scheduling (solved by systemd) and cluster-wide scheduling. Since then, kubernetes was released and it was clear to us that this was the superior technology and this is what we would focus our efforts on instead. I've been fighting to officially sunset the fleet project for a long time now and we finally have a roadmap. Unfortunately, it has taken far too long to relay that plan to our users and I feel forced to warn people now.

All new projects should not use fleet. Our official recommendation is kubernetes. It can do everything fleet can do (and much more) except it can do it much quicker and more efficiently.

We will be making an official announcement early next year and will provide details on the timeline, how to continue using fleet, and how to migrate to kubernetes.

[codepunkt]

@crawford thanks a lot for your thorough explanation!