Hi David,
I am aware of this already being achievable via Daemonsets, although last
time I checked a Daemonset had to run on all nodes and did not honor a node
selector (this may have been fixed).
Assuming that this has been fixed, it is often not desirable for a "Puppet"
script to run
Just to be clear, DaemonSet does allow all of these
> - Run or schedule a job on all machines
> - Run or schedule a job on a sub-set of machines using a label selector
> - Run or schedule a job on a single machine"
It just requires that the "job" be a "run forever" thing, not a "run to
I feel this feature is even more important with the deprecation of Fleet.
"For me this is a very important issue but more in the broader scope of
Jobs (not just scheduled).
CoreOS has deprecated Fleet (distributed systemd). With Fleet one could
broadcast a global systemd unit to run on and
I would recommend filing a Github issue about this. Discussing on the
mailing list is not good for tracking this long-term.
On Mon, Jan 23, 2017 at 10:30 AM, Ben Kochie wrote:
> This sounds like a job that should happen at provisioning time, or by
> config management
This sounds like a job that should happen at provisioning time, or by
config management software.
On Jan 23, 2017 10:00, "Mayank" wrote:
> Yeah, my use case is basically change the permissions of the hostPath so
> that my pods running as non root can access it. I dont want
On Fri, Jan 20, 2017 at 1:40 PM 'Tim Hockin' via Kubernetes user discussion
and Q wrote:
> Concretely the "tweak a sysctl" thing leaves machines that are
> "dirty". Once you allow any users to do this, the machines become
> less useful for anyone else who
If you want to tweak something about a machine, you probably want to
actually occasionally check to ensure that it stayed tweaked (no one put it
back). So I guess I'd expect the current behavior of daemonset is actually
correct for that.
On Fri, Jan 20, 2017 at 1:40 PM, 'Tim Hockin' via
Concretely the "tweak a sysctl" thing leaves machines that are
"dirty". Once you allow any users to do this, the machines become
less useful for anyone else who doesn't specifically tolerate that
tweak. Almost every sysctl represents a tradeoff. Optimize for
low-latency network? Pay higher CPU
Brandon, would you like to file an issue in kubernetes/kubernetes to start?
FWIW the privileged run-to-completion node configuration script is a use
case we have also seen at Google, but the semantics get a bit tricky. We
could start with just a run-to-completion DaemonSet which I think covers
the