1 files changed, 27 insertions, 16 deletions

diff --git a/mcp/config/states/maas b/mcp/config/states/maas
index 8f7a86611..10cfd01bf 100755
--- a/mcp/config/states/maas
+++ b/mcp/config/states/maas
@@ -14,6 +14,9 @@ ERASE_ENV=${ERASE_ENV:-0}
 source "$(dirname "${BASH_SOURCE[0]}")/../../scripts/globals.sh"
 source "$(dirname "${BASH_SOURCE[0]}")/../../scripts/lib.sh"
 
+bm_nodes=$(salt --out yaml 'mas01*' pillar.get maas:region:machines | \
+           awk '/^\s+\w+[[:digit:]]+:$/ {gsub(/:$/, "*"); printf "%s ", $1}')
+
 # Wait for MaaS commissioning/deploy to finish, retry on failure
 function maas_fixup() {
   local statuscmd="salt 'mas01*' --out yaml state.apply maas.machines.status"
@@ -27,12 +30,19 @@ function maas_fixup() {
 
   local fcnodes=$(echo "${statusout}" | \
     grep -Pzo 'status: Failed commissioning\n\s+system_id: \K.+\n')
+  local ftnodes=$(echo "${statusout}" | \
+    grep -Pzo 'status: Failed testing\n\s+system_id: \K.+\n')
   for node_system_id in ${fcnodes}; do
     salt -C 'mas01*' state.apply maas.machines.delete \
       pillar="{'system_id': '${node_system_id}'}"
     sleep 10
   done
-  if [ -n "${fcnodes}" ]; then
+  for node_system_id in ${ftnodes}; do
+    salt -C 'mas01*' state.apply maas.machines.override_failed_testing \
+      pillar="{'system_id': '${node_system_id}'}"
+    sleep 10
+  done
+  if [ -n "${fcnodes}" ] || [ -n "${ftnodes}" ]; then
     salt -C 'mas01*' state.apply maas.machines
     return 1
   fi
@@ -47,6 +57,16 @@ function maas_fixup() {
     sleep 10
   done
   if [ -n "${fdnodes}" ] || [ -n "${rnodes}" ]; then
+    for node_system_id in ${fdnodes} ${rnodes}; do
+      # For now, we allocate 30GB (fixed) for / on cmp nodes
+      local node_hostname=$(echo "${statusout}" | \
+        grep -Pzo 'hostname: \K.+(?=\n.+\n\s+system_id: '"${node_system_id}"')')
+      if [[ "${node_hostname}" =~ ^cmp ]]; then
+        salt -C 'mas01*' state.apply maas.machines.set_storage_layout \
+          pillar="{'system_id': '${node_system_id}', 'lv_size': '32212254720'}"
+        sleep 10
+      fi
+    done
     salt -C 'mas01*' state.apply maas.machines.deploy
     return 1
   fi
@@ -68,16 +88,19 @@ fi
 
 # MaaS rack/region controller, node commissioning
 salt -C 'mas01*' state.apply linux,salt,openssh,ntp
+salt -C 'mas01*' state.apply maas.pxe_nat
 salt -C 'mas01*' state.apply maas.cluster
 
 wait_for 10 "salt -C 'mas01*' state.apply maas.region"
 
-salt -C 'mas01*' state.apply maas.machines
+if [ -n "${bm_nodes}" ]; then
+  salt -C 'mas01*' state.apply maas.machines
+fi
 
 # cleanup outdated salt keys
 sleep 30
 salt-key --out yaml | awk '!/^(minions|- cfg01|- mas01)/ {print $2}' | \
-  xargs -I{} salt-key -yd {}
+  xargs --no-run-if-empty -I{} salt-key -yd {}
 
 # MaaS node deployment
 wait_for 10 maas_fixup
@@ -87,18 +110,6 @@ salt -C 'mas01*' pillar.item\
   maas:region:admin:password
 
 # Check all baremetal nodes are available
-rc=1
-attempt=0
-total_attempts=10
-while [ $rc -ne 0 ] && [ ${attempt} -lt ${total_attempts} ]; do
-  bm_nodes=$(salt --out yaml 'mas01*' pillar.get maas:region:machines | \
-             awk '/^\s+\w+[[:digit:]]+:$/ {gsub(/:$/, "*"); print $1}')
-  rc=0
-  for node in $bm_nodes; do
-    salt "$node" test.ping 2>/dev/null || { rc=$?; break; };
-  done
-  sleep 5
-  ((attempt+=1))
-done
+wait_for 5.0 "(for n in ${bm_nodes}; do salt \${n} test.ping 2>/dev/null || exit; done)"
 
 wait_for 10.0 "salt -C '* and not cfg01* and not mas01*' saltutil.sync_all"