The install file is over 75MB and incurs a significant transfer overhead
during the initialization of the ansible environments when used with a
provisioning framework like vagrant which stages ansible on the target
node.  This install file should likely be downloaded from an online source.

The default user aligns with the conventions of the image provisioning
framework.  For the centos cloud image that user name is "centos".

The internal network nic for the head node relies on static
IP address assignement. Explicitly assign the IP address
via so NetworkManager doesn't default to requesting an IP
address via DHCP and causing the eth1 interface to pend
configuration idefinitly.  Without eth1 all communication
with the head node on the cluster network will fail, e.g.
sacctmgr setup.

…-openstack

Corrects master internal IP set in /etc/hosts from a static value in config to the actual assigned value for the interface.

For packer to build compute image

Now support 10 compute nodes: c0 to c9

The two files added here come from compute_build_nodes
and nodes_vivify roles. These two new files will be
modified for suitable cloud operations.

The home dir would not mount on OOD if we did not disable the firewalld
in OHPC, it was working in our previous setup while deploying in vagrant
as the VBox VM was running and was never turned off during the setup,
but in the cloud deploy we have to preserve the state we have in the
build phase (i.e. after packer builds the image). We need to disable
it so it would not get turned back on after we boot the image during
deploy

We dont use DHCP broadcasts for auto node discovery so we set this flag
to false.

In cloud deploy PXE boot is no longer required so those tasks are
removed.

we don't use DHCP broadcast for node discovery, so we remove the
wwnodescan task and DHCP restart task.

This conditional will check if a node exists in WW DB, if yes then it
will just update the IP and MAC address during node registration, if it
doesn't then it will add it as a new node.

In the cloud deploy we no longer need this pause, it was required in
a warewulf based deploy to let the compute nodes boot so that pdsh
commands can be run, with terraform the compute nodes will boot from
the compute-image and be ready for OHPC to run the tasks because of
the dependency we defined in terraform.

install_ww_bin
compute_build_image

Running scontrol right after start up of slurmctld on
nodes causes the command to fail. A slight delay allows
the system state to settle and the ansible playbook to
exit cleanly.

This task was in compute_build_vnfs role but due to rearrangement
of tasks for cloud deploy, it is moved to ohpc_config_build

…file target

Based on ood.yml with addition of role for warewulf tools install
developed during compute build.  Also installs user reg app.
This unbundles build/install steps from deploy steps to support
cloud deployment.
These tasks were previously bundled in a single-phase build+deploy step.

Based on ood_user_reg role with the post-deploy
reg app account create steps removed for inclusion
in an ops phase.
This delays running the reg app account create
until the ops phase when the user and group ids
are known.

Add ohpc and copr repos to add cluster tools and slurm to image at
build time.
Add autofs to remove static nfs-mounts from boot sequence dependencies.
Change selinux to permissive.

ood-portal-generator needs to be run after the apache config
is changed for the user reg app in order to deploy the apache
config.

The CentOS 7.6 repos for SCLo stopped working during testing.
Manually setting baseurl works around the issue.

See:

https://discourse.osc.edu/t/errors-with-rhel-server-rhscl-7-rpms-repo-in-the-past-week/494/8

This improves #cd58c183 to use just the major os release rather
than build base path to ensure software collection dependencies
continue to be installed over subsequent releases.

Ensure the user registration app is active during the build so
it is also available in the deployed environment.

This role creates the reg app user account during the ops phase
when the ood node is instantiated.
It starts the Apache web server once the reg app user account is
created.

Compute build image onto feat-openstack

This captures all the requirements to build the compute nodes for a deploy with terraform

Devint ood compute build

There is no need for tasks to configure the boot parameters of
compute nodes in warewulf because it isn't responsible for node
booting.
Also remove harness for automatic node detection for the same
reason.

This task is not needed since warewulf does not manage node boot
and provisioning.

Devint ohpc build

Using the group name for headnode instead of the ohpc host name
from the inventory in the playbooks makes it easier to build
images with custom host names.
It's also slightly more correct to use the group name instead of
the host name and aligns with other uses of group name in playbooks.

Change the playbook target to the headnode group name

…d db

The build-time setting of the slurm controller and db hostname cause
slurm startup to fail during deploy (ops) phase.  The ohpc_install_ops
role updates the slurm.conf from the template to include the current
names. The conditional avoids multiple writers to the file, since there
is one call to the code for each client node in the cluster due to the
current null resource config.

Update build image with runtime instance name for slurm controller and db

Easybuild bootstrap script is failing with following error
"Failed to find right 'easy_install' command!" during easybuild
installation. To avoid this we install python3 and pip3, so that setup
tools for python3 can be installed using pip3.

It also avoids easy_install error about finding the right command

Leaving password blank, when runs playbook, pass in with arguments:
 --extra-vars rabbitmq_user_password="YoUrPasSwOrd"

patch to fix job state directory space in slurm.conf

firewall will block message passing to/from the node,
  so turn it off for now.

* Specify the full paht for git repo to be installed in
* Changing the pattern for what to find and change
* Using shell script to start ood_account_create in background

…-error

Patch fix easybuild error

It seems this was inherited from earlier use on ood.  Firewall should be off by default on ohpc as currently deployed in dev.  We should figure out the ports needed.

Installing rabbitmq-server on ohpc node

 Add rabbitmq agents as service

python pip version update for Flask User reg app