On Fri, Jul 8, 2016 at 8:17 AM, Hardy Ferentschik <[email protected]> wrote:
> Hi, > > > Not running the Red Hat Docker is a serious problem for OpenShift / > > Kube, simply given the instability and gaps in upstream Docker. While > > we're not running production workloads, it's really difficult to > > certify and fix issues. > > Here is a bit I don't understand. We keep saying we should run the Red Hat > Docker > and somehow this seems to imply that you also need the CentOS/RHEL base > image. > > Can one not replace the default Docker binary of any distribution with the > Red Hat version? Or does the Red Hat version depend on things which are > only > available on RHEL. > The Red Hat version does not truly depend on RHEL, but there are a fair number of things between device mapper and docker to get exactly right, so it's not just Docker but also libdevmapper and a bunch of other things to be aligned. > > My understanding was that Red Hat even offered a patch for Docker, but they > are just not willing to accept this. So there should be not extras > required, right? > There are two classes of things in the Red Hat version of Docker: 1. Backports from newer Docker versions - Docker does *not* patch older versions, so you can't get fix A without upgrading to Docker next version without also getting bug B. In practice we were finding that it was impossible to stabilize a cluster with Docker without back porting fixes. Along with that comes specific security and other fixes that are critical to having a long term production version of docker. 2. Specific fixes that Docker refuses to carry - things like allowing registries to be reparented, or the ability to disable Docker schema2 push to registries (allowing Docker to continue to push in a way that preserves pull-by-digest for older Docker versions). The former is critical. The latter is just very useful. > > --Hardy > >
_______________________________________________ Devtools mailing list [email protected] https://www.redhat.com/mailman/listinfo/devtools
