Merge ~arif-ali/stsstack-bundles:add_script_tools into stsstack-bundles:master

[Arif Ali]

Add script to add ceph-osd to nova-compute

In some scenarios the ceph-osd is co-located on nova-compute
and its useful for reproducers to be able to have this scenario

The script is something that worked for me, but may require more
work for the wider audience

[Edward Hope-Morley (hopem)]

Hi Arif, I wonder if there is a way that we could do this without it having to be a two-phase deployment. Also, I think this could be simplified by taking the existing ceph overlay [1] and making the number of osd devices configurable, that way Juju will automatically create the volumes and attach them to the node that ceph-osd is running on. Thoughts?

[1] https://git.launchpad.net/stsstack-bundles/tree/overlays/ceph.yaml#n25

[Arif Ali (arif-ali)]

Yeah, that would be the ultimate solution, as the cleanup process would be a lot easier too

I will need to have a play, and test, and see how we can do this. I think we'll probably need a new option with generate_bundles.sh to enable a feature so that the overlays can change dynamically

Again, this was a quick solution to a problem I was trying to solve

[Edward Hope-Morley (hopem)]

@arif-ali fyi there is now a --num-cepn-osds option that allows you to modify the number of OSDs per ceph-osd unit at deploy time. Hopefully that can help you a bit.

review: Needs Information

[Edward Hope-Morley (hopem)]

The repository has moved to https://github.com/canonical/stsstack-bundles/. Please resubmit there if you are still working on this patch.

Unmerged commits

3ff7758... by Arif Ali on 2020-09-01

Add script to add ceph-osd to nova-compute

In some scenarios the ceph-osd is co-located on nova-compute
and its useful for reproducers to be able to have this scenario

The script is something that worked for me, but may require more
work for the wider audience

diff --git a/openstack/tools/add_volumes_to_nova_compute.sh b/openstack/tools/add_volumes_to_nova_compute.sh
new file mode 100755
index 0000000..dce0ff5
--- /dev/null
+++ b/openstack/tools/add_volumes_to_nova_compute.sh
@@ -0,0 +1,53 @@
+#!/bin/bash
+
+# This script wil create 3 10G volumes for the purpose of adding ceph-osd to nova-compute units
+#
+# This is more in-line with how some customers would do it. This should automate everything
+# based on this scenario. This will also use less instances and hence less RAM
+
+# TODO: need to paramterise this
+series=bionic
+ceph_mon_deploy="cs:ceph-mon"
+ceph_osd_deploy="cs:ceph-osd"
+cinder_ceph_deploy="cs:cinder-ceph"
+
+. ~/novarc
+
+# let's deploy ceph-mon, while the volumes are being created
+juju deploy ${ceph_mon_deploy} ceph-mon --constraints mem=1G -n 3 --series ${series}
+
+juju deploy ${cinder_ceph_deploy} cinder-ceph --series ${series}
+
+# The machines that are allocated to nova-compute
+machines=$(juju status nova-compute --format=json | jq '.["applications"]["nova-compute"]["units"][]["machine"]' | sed s/\"//g | xargs)
+
+# Go through all the machines and create the 3 volumes, and attach it to the instance
+for machine in $machines
+do
+  instance=$(juju status nova-compute --format=json | jq '.["machines"]["'${machine}'"]["instance-id"]' | sed s/\"//g)
+  for i in `seq 1 3`
+  do
+    vol=vol${machine}${i}
+    openstack volume create --size 10 ${vol}
+    openstack server add volume ${instance} ${vol}
+  done
+done
+
+# Now deploy ceph-osd to all of the machines
+juju deploy ${ceph_osd_deploy} ceph-osd  --config osd-devices="/dev/vdb /dev/vdc /dev/vdd" -n $(echo ${machines} | wc -w) --to $(echo ${machines} | tr ' ' ',') --series ${series}
+
+# Add all relations
+juju add-relation ceph-osd:mon                ceph-mon:osd
+juju add-relation ceph-mon:client             nova-compute:ceph-access
+juju add-relation ceph-mon:client             glance:ceph
+
+juju add-relation cinder-ceph:ceph            ceph-mon:client
+juju add-relation cinder-ceph:storage-backend cinder:storage-backend
+juju add-relation cinder-ceph:ceph-access     nova-compute:ceph-access
+
+# TODO: echo back the ids of the volumes that may need to be deletes.
+# Probably better, list the commands that would do it, and maybe create
+# clean_volumes script for it to be available
+#
+# Note that the volumes won't delete if they are attached, so should be
+# good to do in most circumstances