juju should add nodes IPs to no-proxy list

[Ashley Lai]

When setting proxy in juju environment the nodes cannot communicate with each other. Juju should allow this communication by adding all the nodes' IPs to no-proxy list.

[Ashley Lai (alai)]

Bug to add bootstrap node to no-proxy list:

https://bugs.launchpad.net/juju-core/+bug/1478660

[Tim Penhey (thumper)]

Can anyone think of a reason why we shouldn't do this by default?

[Marco Ceppi (marcoceppi)]

No, making it default behavior where it appends the nodes IP addresses to the existing user defined no-proxy would be preferable.

[Cheryl Jennings (cherylj)]

In bug #1556207, we have automatically added state servers / controllers to the no-proxy list.

[David Britton (dpb)]

If we want to claim http_proxy support for the autopilot or cloud deployments. This is one key step that needs to happen before getting there. the 'no_proxy' list needs to be managed and curated by Juju.

[Anastasia (anastasia-macmood)]

Removing 2.1 milestone as we will not be addressing this issue in 2.1.

[John A Meinel (jameinel)]

Note that most applications require you to list *every* IP address that you don't want to proxy separately *or* create a Domain and only reference things via the Domain Name.
See bugs like https://bugs.launchpad.net/juju/+bug/1421650

If we *could* we'd set no_proxy with nice short-form CIDR notation around what spaces (subnets) we are deployed in, which would leave you with a reasonably short list.

We could try to do "no_proxy everything in the model" and hope you never have 1000s of machines. (Note that no_proxy becomes an O(N^2) problem as everytime you deploy a new machine, you now have to tell all the other machines that they shouldn't proxy traffic to that machine.)

We could try "no_proxy everything related to units on the machine" which ignores some realities, but would at least keep N smaller. (Still N^2 in the number of machines with units related to each-other.)

Its a similar problem that they ran into with Hadoop, where everything *had* to be a Hostname, which meant they were populating /etc/hosts with the 100 machines that were being spun up for the hadoop charm. The problem is still that a given application can have 100 machines.

One option would be to re-open the idea of Juju becoming a DNS server, and giving a common suffix to every machine in the model, and only configuring traffic between machines to use the DNS names.

That is a rather major overhaul of the communication model to handle no_proxy.

[Canonical Juju QA Bot (juju-qa-bot)]

This bug has not been updated in 5 years, so we're marking it Expired. If you believe this is incorrect, please update the status.