[MaaS] Override failed testing by default

[fuel.git] / mcp / config / states / maas

diff --git a/mcp/config/states/maas b/mcp/config/states/maas
index a6bac89..0034e78 100755 (executable)
--- a/mcp/config/states/maas
+++ b/mcp/config/states/maas
@@ -1,4 +1,5 @@
 #!/bin/bash -e
+# shellcheck disable=SC1090,SC2155
 ##############################################################################
 # Copyright (c) 2017 Mirantis Inc., Enea AB and others.
 # All rights reserved. This program and the accompanying materials
@@ -8,43 +9,54 @@
 ##############################################################################
 
 CI_DEBUG=${CI_DEBUG:-0}; [[ "${CI_DEBUG}" =~ (false|0) ]] || set -x
+ERASE_ENV=${ERASE_ENV:-0}
 
-# shellcheck disable=SC1090
+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 ", $1}')
+
 # Wait for MaaS commissioning/deploy to finish, retry on failure
 function maas_fixup() {
   local statuscmd="salt 'mas01*' --out yaml state.apply maas.machines.status"
-  # shellcheck disable=SC2155
   local ncount=$(salt --out yaml 'mas01*' pillar.get maas:region:machines | \
     grep -cE '^\s{2}\w+:$')
-  wait_for 180 "${statuscmd} | tee /dev/stderr | " \
-    "grep -Eq '((Deployed|Ready): ${ncount}|status:Failed|status:Allocated)'"
-  # shellcheck disable=SC2155
+
+  # wait_for has 10sec timeout * 96 = 16 min > 15min for Failed state
+  wait_for 96 "${statuscmd} | tee /dev/stderr | " \
+    "grep -Eq '((Deployed|Ready): ${ncount}|status: (Failed|Allocated))'"
   local statusout=$(eval "${statuscmd}")
 
-  # shellcheck disable=SC2155
   local fcnodes=$(echo "${statusout}" | \
-    grep -Po '(?<=system_id:)(.*)(?=,status:Failed commissioning)')
+    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
 
-  # shellcheck disable=SC2155
   local fdnodes=$(echo "${statusout}" | \
-    grep -Po '(?<=system_id:)(.*)(?=,status:(Failed deployment|Allocated))')
+    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.deploy
     return 1
   fi
@@ -52,30 +64,41 @@ function maas_fixup() {
   return 0
 }
 
-# MaaS rack/region controller, node commissioning
-salt -C 'mas01*' cmd.run "add-apt-repository ppa:maas/stable"
+# Optionally destroy MaaS machines from a previous run
+if [ "${ERASE_ENV}" -gt 1 ]; then
+  set +e; dnodes=$(salt 'mas01*' --out yaml state.apply maas.machines.status | \
+    grep -Pzo '\s+system_id: \K.+\n'); set -e
+  cleanup_uefi
+  for node_system_id in ${dnodes}; do
+    salt -C 'mas01*' state.apply maas.machines.delete \
+      pillar="{'system_id': '${node_system_id}'}"
+    sleep 10
+  done
+fi
 
+# MaaS rack/region controller, node commissioning
 salt -C 'mas01*' state.apply linux,salt,openssh,ntp
-salt -C 'mas01*' state.apply linux.network.interface
-salt -C 'mas01*' state.apply maas.pxe_nat
 salt -C 'mas01*' state.apply maas.cluster
-salt -C 'cfg01*' state.apply maas.pxe_route
 
 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\
   maas:region:admin:username \
   maas:region:admin:password
 
-wait_for 10 "salt -C '* and not cfg01* and not mas01*' saltutil.sync_all"
+# Check all baremetal nodes are available
+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"