End to End Tour

A guided tour of everything involved in creating, realizing and materializing an experiment on Merge.

Core Concepts

To get a sense for how Merge works, and in particular how experiments are defined, realized and materialized, we need to go over a few core architectural features of Merge that are pervasive throughout the platform.

Model Driven Design

At the end of the day Merge is all about models. When a resource provider commissions a new testbed facility, they do that by providing a model of the facility to a Merge portal. When a user creates a new revision of an experiment, they do so by pushing a commit of their experiment model to a Git repository that is tracked by a Merge portal. When a user goes to realize an experiment, the Merge portal takes their experiment model, takes all the models it has of registered testbed facilities and creates an aggregate interconnected testbed model called the resource internet and then computes an embedding of the users experiment model onto the resource internet model.

Reconcilers

Both the Merge portal and the Mars testbed kernel are built around a reconciler architecture. Both have well defined API and an API server. When requests are made to the API, the API server translates a request into desired state. For both the Portal and Mars there are a collection of reconcilers running that monitor desired state, and drive actual state toward the target state. As a concrete example every Merge experiment comes with a Git repository. When a user creates a new experiment through the CLI, the mrg CLI tool calls the Merge Portal API which is implemented by the portal API server. The API server translates that new experiment request into a set of desired states. One piece of state in that set is having a Git repository associated with the new experiment. The API server places the parameters of that Git repository in the Portal’s etcd data store. The Git reconciler running in the portal is monitoring etcd for state updates pertaining to Git repos. When it sees this new state it ensures that the Git repo is present and ready to go for that users experiment. The same process happens when an experiment is deleted, except the Git reconciler is now observing the removal of a target state and thus removes the associated Git repository.

Setting Up A Testing Environment

There are two major systems to set up for running an end to end test.

  1. A Merge Portal appliance.
  2. A virtual Merge testbed running Mars. In this tutorial we’ll be using the Phobos virtual testbed environment.

Portal Appliance

The Portal appliance is a libvirt packaged virtual machine with a pre-installed Merge portal system. It can be stood up as follows

curl -L https://gitlab.com/mergetb/portal/appliance/-/raw/master/launch-appliance.sh | bash

Facility commissioning

The first thing we need to do on the portal is set up a facility.

Grab the latest version of the mrg CLI tool.

Point mrg at your appliance

mrg config set server grpc.mergetb.example.net

Login using the credentials the launch-appliance.sh script downloaded into your working directory.

pw=`cat portal-genconf* | grep opspw | awk '{print $2}'`
mrg login ops $pw --nokeys

Create a user and initialize a user that represents a testbed operator.

mrg logout
mrg register olive olive@mergetb.org muffins1701

mrg login ops $pw --nokeys
mrg init olive
mrg activate olive
mrg logout

Login as the testbed operator and commission the Phobos testbed facility.

mrg login olive muffins1701
curl -OL https://gitlab.com/mergetb/devops/vte/phobos/-/raw/master/model/cmd/phobos.xir
mrg new fac phobos phobos.example.com phobos.xir

Verify you can see the new facility

[ry@ryzen2 v0.7]$ mrg list fac
NAME      ADDRESS               DESCRIPTION    MODE
phobos    phobos.example.com                   Public

Now download the comissioned version of this model.

rm phobos.xir
mrg show fac phobos -mx

This will download version of the phobos facility that has routing and addressing information assigned to it by the portal that can be used to initialize the Phobos virtual testbed facility.

Logout

mrg logout

More detailed instructions on the use and operation of the Merge portal appliance are available here.

Phobos

rvn build
rvn deploy
rvn pingwait ops mgmt
rvn status
rvn configure
ansible-playbook -i .rvn/ansible-hosts -i ansible-interpreters bootstrap.yml
rvn reboot ifr

# !!!
# !!! Wait for ifr to provision (watch VM console)
# !!!
# !!! help-wanted: detect when ifr is ready
# !!!
eval $(rvn ssh ops)
sudo su
mrs init minio
mrs init images

update your testbed model with one that has been comissioned with a your portal appliance and was downloaded in the mrg show fac step above.

mrs update model <phobos.xir>

In a seprate terminal as root

rvn apiserver

Then continue with the setup

# initialize the harbor
mrs init harbor


# check network status,
mrs list net

# You should be able to rest one of the nodes (say x0) and successfully enter sled.
# Bring x0 up in virt-manager, set the view mode to serial console and force reset
# the node. You should get all the way into sled on reboot.

More detailed instructions for setting up and running Phobos are here.

Interconnection

Once the Appliance and Phobos are up, we need to make sure they can talk to each other from the host machine.

Setup your local firewall rules to allow the Appliance and Phobos to talk

  1. Lookup what interface phobos is on
virsh net-list

Locate the entry with the form phobos_<n>_raven-infra where <n> is an integer. Display that network

virsh net-dumpxml phobos_<d>_raven-infra

this will contain an entry that looks like this

<bridge name='virbr<m>' stp='on' delay='0'/>

where <m> is an integer.

  1. Add iptables forwarding rules.
sudo iptables -I FORWARD -o tt0 -i virbr<m> -j ACCEPT
sudo iptables -I FORWARD -i tt0 -o virbr<m> -j ACCEPT
  1. Add a /etc/hosts entry on your local workstation

Look up the 172.22.x.x address on the ops host and add it to your workstation /etc/hosts as follows.

172.22.x.x phobos.example.com

Then flush the dnsmasq cache so the portal can find phobos through libvirt dns

sudo killall -HUP dnsmasq

Testing

Now that we have a Merge portal and Mars-based facility running virtually on our workstation, we can begin doing things through the Merge API using the mrg command line tool.

First setup your mrg client to point at your Portal appliance

mrg config set server grpc.mergetb.example.net

Login using the credentials from the configuration that came with the appliance.

pw=`cat portal-genconf* | grep opspw | awk '{print $2}'`
mrg login ops $pw --nokeys

Experimentation

Create a user and initialize a user that represents a testbed user.

mrg logout
mrg register murphy murphy@mergetb.org muffins1701

mrg login ops $pw --nokeys
mrg init murphy
mrg activate murphy
mrg logout

Login as the testbed user and follow the hello world example.

Modify the experiment source to use bare-metal resources.

from mergexp import *

# Create a netwok topology object.
net = Network('hello-world')

# Create two nodes.
a,b = [net.node(name, metal == True) for name in ['a', 'b']]

# Create a link connecting the two nodes.
link = net.connect([a,b])

# Give IP addresses to the nodes on the link just created.
link[a].socket.addrs = ip4('10.0.0.1/24')
link[b].socket.addrs = ip4('10.0.0.2/24')

# Make this file a runnable experiment based on our two node topology.
experiment(net)

Push an update to the experiment git repo, relinquish (mrg relinquish from the CLI) the current realization and create a new one.

Now it’s time to materialize

mrg materialize world.hello.murphy

This will kick off a materialization on the Phobos site. If you open up a console to the Raven virtual machines that have been allocated by the realization engine (x0 and x1 usually chosen in my experience) using the virt-manager tool available on fedora, you will see that these nodes have been rebooted and they will go through the imaging process.

Once the nodes come up in the default Debian based operating system, you can login with the credentials user=test password=test. You can then setup addresses on the experiment interfaces of the nodes and the should be able to ping each other. At the time of writing the system that configures the nodes called Foundry is not yet fully integrated, so things like user accounts and automatic experiment network setup for things like addresses and routes is not there yet.


Last modified June 30, 2021: remove cumulus warning (97b7e45)