juju status shows ip address from public-api space rather than internal-api space

here is one example where juju machine show different subnet for public address. I have two network 10.200.2.x and 10.200.5.x and bot routable to network. PXE boot network is 10.200.5.x but nodes public address is on 10.200.2.x

Unit Workload Agent Machine Public address Ports Message
aodh/0 waiting allocating 0/lxd/0 waiting for machine
ceilometer/0 waiting allocating 2/lxd/0 waiting for machine
ceph-mon/0 waiting allocating 2/lxd/1 waiting for machine
ceph-mon/1 waiting allocating 1/lxd/0 waiting for machine
ceph-mon/2 waiting allocating 0/lxd/1 waiting for machine
ceph-osd/0 waiting allocating 0 10.200.2.15 waiting for machine
ceph-osd/1 waiting allocating 1 10.200.2.13 waiting for machine
ceph-osd/2 waiting allocating 2 10.200.2.14 waiting for machine
ceph-radosgw/0 waiting allocating 0/lxd/2 waiting for machine
cinder/0 waiting allocating 1/lxd/1 waiting for machine
glance/0 waiting allocating 0/lxd/3 waiting for machine
heat/0 waiting allocating 0/lxd/4 waiting for machine
keystone/0 waiting allocating 2/lxd/2 waiting for machine
mongodb/0 waiting allocating 0/lxd/5 waiting for machine
mysql/0 waiting allocating 0/lxd/6 waiting for machine
neutron-api/0 waiting allocating 1/lxd/2 waiting for machine
neutron-gateway/0 waiting allocating 0 10.200.2.15 waiting for machine
nodes/0 waiting allocating 0 10.200.2.15 waiting for machine
nodes/1 waiting allocating 1 10.200.2.13 waiting for machine
nodes/2 waiting allocating 2 10.200.2.14 waiting for machine
nova-cloud-controller/0 waiting allocating 1/lxd/3 waiting for machine
nova-compute/0 waiting allocating 1 10.200.2.13 waiting for machine
nova-compute/1 waiting allocating 2 10.200.2.14 waiting for machine
openstack-dashboard/0 waiting allocating 1/lxd/4 waiting for machine
opnfv-promise/0 waiting allocating 0/lxd/7 waiting for machine
rabbitmq-server/0 waiting allocating 2/lxd/3 waiting for machine

I would actually expect "juju status" to show a public address if possible,
rather than an internal address. That doesn't mean it is the address that a
given unit should be advertising to its peers (that would be "juju run
--unit ceph-mon/0 network-get --primary-address BINDING").
Did you actually deploy the services bound into spaces, or you just
deployed them on a system that has more than one network interface and
hoped it would always pick what is "right" for you?

On Thu, Jan 26, 2017 at 9:18 PM, Narinder Gupta <email address hidden>
wrote:

> here is one example where juju machine show different subnet for public
> address. I have two network 10.200.2.x and 10.200.5.x and bot routable
> to network. PXE boot network is 10.200.5.x but nodes public address is
> on 10.200.2.x
>
> Unit Workload Agent Machine Public address
> Ports Message
> aodh/0 waiting allocating 0/lxd/0
> waiting for machine
> ceilometer/0 waiting allocating 2/lxd/0
> waiting for machine
> ceph-mon/0 waiting allocating 2/lxd/1
> waiting for machine
> ceph-mon/1 waiting allocating 1/lxd/0
> waiting for machine
> ceph-mon/2 waiting allocating 0/lxd/1
> waiting for machine
> ceph-osd/0 waiting allocating 0 10.200.2.15
> waiting for machine
> ceph-osd/1 waiting allocating 1 10.200.2.13
> waiting for machine
> ceph-osd/2 waiting allocating 2 10.200.2.14
> waiting for machine
> ceph-radosgw/0 waiting allocating 0/lxd/2
> waiting for machine
> cinder/0 waiting allocating 1/lxd/1
> waiting for machine
> glance/0 waiting allocating 0/lxd/3
> waiting for machine
> heat/0 waiting allocating 0/lxd/4
> waiting for machine
> keystone/0 waiting allocating 2/lxd/2
> waiting for machine
> mongodb/0 waiting allocating 0/lxd/5
> waiting for machine
> mysql/0 waiting allocating 0/lxd/6
> waiting for machine
> neutron-api/0 waiting allocating 1/lxd/2
> waiting for machine
> neutron-gateway/0 waiting allocating 0 10.200.2.15
> waiting for machine
> nodes/0 waiting allocating 0 10.200.2.15
> waiting for machine
> nodes/1 waiting allocating 1 10.200.2.13
> waiting for machine
> nodes/2 waiting allocating 2 10.200.2.14
> waiting for machine
> nova-cloud-controller/0 waiting allocating 1/lxd/3
> waiting for machine
> nova-compute/0 waiting allocating 1 10.200.2.13
> waiting for machine
> nova-compute/1 waiting allocating 2 10.200.2.14
> waiting for machine
> openstack-dashboard/0 waiting allocating 1/lxd/4
> waiting for machine
> opnfv-promise/0 waiting allocating 0/lxd/7
> waiting for machine
> rabbitmq-server/0 waiting allocating 2/lxd/3
> waiting for machine
>
> --
> You receive...

We track Juju 2.x issues in "juju" project. Re-targeting and marking as Incomplete since we are awaiting Narinder Gupta's reply.

John,
Will you please define a public address here? As i have two networks here 10.200.2.x (eno1) and 10.200.5.x (eno2) and both are routable to outside world through a gateway 10.200.2.254 and 10.200.5.254. I can configure all my servers pxe boot on any of the network. Problem is with insistency as i do configure my script to configure based on juju status of an service. How can i predict juju status will show the ip address from the specific subnet all the time?

look like it picks the random subnet which are routable. In this example in MAAS i configure two spaces as internel-api and public-api. All services are showing the openstack API on correct network as expected but my scripts based on juju status fails. As i want juju status to show only the ip address from 10.200.5.x subnet rather than 10.200.2.x

Thanks and Regatds,
Narinder Gupta

Can you confirm that "juju show-machine 0", and "juju status --format=yaml" are showing both addresses as you would expect?

Generally "juju status" in tabular form is not intended to be scraped for scripting. I would recommend either using "juju status --format=json" or "--format=yaml" which are both more ameniable to automatic processing.

As for whether status reports 10.200.5.x or 10.200.2.x. Generally, there isn't a reason why one set of addresses would be preferred for external access than the other. (There may be sites where PXE network is the one with external routing, and others where it *isn't* externally routable.)

We should be able to give you all addresses (and do everywhere but in tabular output).

I'm curious why when you say "both are routable" but one of them is failing.

This bug has not been updated in 2 years, so we're marking it Low importance. If you believe this is incorrect, please update the importance.