X-Git-Url: https://gerrit.opnfv.org/gerrit/gitweb?a=blobdiff_plain;f=mcp%2Fconfig%2Fstates%2Fmaas;h=f321b716062cfdd689476c7b814bed3bf25fbb37;hb=ab18375a629010525ac15bc11ce2d4e4cf393fe9;hp=5e33f3a0858fba93acd49c2875a840247f9797f9;hpb=5ceea06b5395115e0e45d0631e726062eb1b0d13;p=fuel.git

diff --git a/mcp/config/states/maas b/mcp/config/states/maas
index 5e33f3a08..f321b7160 100755
--- a/mcp/config/states/maas
+++ b/mcp/config/states/maas
@@ -14,35 +14,50 @@ 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 ", $1}')
+
 # Wait for MaaS commissioning/deploy to finish, retry on failure
 function maas_fixup() {
   local statuscmd="salt 'mas01*' --out yaml state.apply maas.machines.status"
   local ncount=$(salt --out yaml 'mas01*' pillar.get maas:region:machines | \
     grep -cE '^\s{2}\w+:$')
-  wait_for 180 "${statuscmd} | tee /dev/stderr | " \
+
+  # wait_for has 10sec timeout * 96 = 16 min > 15min for Failed state
+  wait_for 96 "${statuscmd} | tee /dev/stderr | " \
     "grep -Eq '((Deployed|Ready): ${ncount}|status: (Failed|Allocated))'"
   local statusout=$(eval "${statuscmd}")
 
   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 30
+    sleep 10
+  done
+  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 "${fcnodes}" ]; then
+  if [ -n "${fcnodes}" ] || [ -n "${ftnodes}" ]; then
     salt -C 'mas01*' state.apply maas.machines
     return 1
   fi
 
   local fdnodes=$(echo "${statusout}" | \
     grep -Pzo 'status: (Failed deployment|Allocated)\n\s+system_id: \K.+\n')
+  local rnodes=$(echo "${statusout}" | \
+    grep -Pzo 'status: Ready\n\s+system_id: \K.+\n')
   for node_system_id in ${fdnodes}; do
     salt -C 'mas01*' state.apply maas.machines.mark_broken_fixed \
       pillar="{'system_id': '${node_system_id}'}"
-    sleep 30
+    sleep 10
   done
-  if [ -n "${fdnodes}" ]; then
+  if [ -n "${fdnodes}" ] || [ -n "${rnodes}" ]; then
+    salt -C 'mas01*' state.apply maas.machines.storage
     salt -C 'mas01*' state.apply maas.machines.deploy
     return 1
   fi
@@ -69,15 +84,16 @@ salt -C 'mas01*' state.apply maas.cluster
 
 wait_for 10 "salt -C 'mas01*' state.apply maas.region"
 
-salt -C 'mas01*' state.apply maas.machines
-wait_for 10 maas_fixup
+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
-salt -C 'mas01*' state.apply maas.machines.deploy
 wait_for 10 maas_fixup
 
 salt -C 'mas01*' pillar.item\
@@ -85,18 +101,6 @@ 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 10.0 "(for n in ${bm_nodes}; do salt \${n} test.ping 2>/dev/null || exit; done)"
 
-wait_for 10 "salt -C '* and not cfg01* and not mas01*' saltutil.sync_all"
+wait_for 10.0 "salt -C '* and not cfg01* and not mas01*' saltutil.sync_all"