X-Git-Url: https://gerrit.opnfv.org/gerrit/gitweb?a=blobdiff_plain;f=mcp%2Fconfig%2Fstates%2Fmaas;h=0034e78f13ad32f42fe9097373489f3d91485030;hb=refs%2Fchanges%2F79%2F52279%2F3;hp=fa5e4465958db45e3a42069fb5512ff466e25131;hpb=95874c0a275b8b8d1cc06d1773943564137b5064;p=fuel.git

diff --git a/mcp/config/states/maas b/mcp/config/states/maas
index fa5e44659..0034e78f1 100755
--- a/mcp/config/states/maas
+++ b/mcp/config/states/maas
@@ -1,72 +1,104 @@
-#!/bin/bash
-function wait_for() {
-  local cmdstr=$@
-  local total_attempts=360
-  local sleep_time=10
-  local attempt=1
-  echo "[NOTE] Waiting for cmd to return success: ${cmdstr}"
-  while ((attempt <= total_attempts)); do
-    eval "${cmdstr}" && break || true
-    echo -n '.'; sleep "${sleep_time}"
-    ((attempt+=1))
+#!/bin/bash -e
+# shellcheck disable=SC1090,SC2155
+##############################################################################
+# Copyright (c) 2017 Mirantis Inc., Enea AB and others.
+# All rights reserved. This program and the accompanying materials
+# are made available under the terms of the Apache License, Version 2.0
+# which accompanies this distribution, and is available at
+# http://www.apache.org/licenses/LICENSE-2.0
+##############################################################################
+
+CI_DEBUG=${CI_DEBUG:-0}; [[ "${CI_DEBUG}" =~ (false|0) ]] || set -x
+ERASE_ENV=${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+:$')
+
+  # wait_for has 10sec timeout * 96 = 16 min > 15min for Failed state
+  wait_for 96 "${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 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 "${fcnodes}" ] || [ -n "${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 10
+  done
+  if [ -n "${fdnodes}" ] || [ -n "${rnodes}" ]; then
+    salt -C 'mas01*' state.apply maas.machines.deploy
+    return 1
+  fi
+
+  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*' cmd.run \
-  "route add -net 192.168.11.0/24 gw ${MAAS_IP:-192.168.10.3}"
 
-wait_for "salt -C 'mas01*' state.apply maas.region"
+wait_for 10 "salt -C 'mas01*' state.apply maas.region"
 
-salt -C 'mas01*' state.apply maas.machines
-# TODO: relax cond, as this is not re-entrant (e.g. nodes already deployed)
-wait_for "salt 'mas01*' --out yaml state.apply maas.machines.status | " \
-         "tee /dev/stderr | fgrep -q 'Ready: 5'"
+if [ -n "${bm_nodes}" ]; then
+  salt -C 'mas01*' state.apply maas.machines
+fi
 
 # cleanup outdated salt keys
-salt-key --out yaml | awk '!/^(minions|- cfg01|- mas01)/ {print $2}' | xargs -I{} salt-key -yd {}
+sleep 30
+salt-key --out yaml | awk '!/^(minions|- cfg01|- mas01)/ {print $2}' | \
+  xargs --no-run-if-empty -I{} salt-key -yd {}
 
 # MaaS node deployment
-salt -C 'mas01*' state.apply maas.machines.deploy
-wait_for "salt 'mas01*' --out yaml state.apply maas.machines.status | " \
-         "tee /dev/stderr | fgrep -q 'Deployed: 5'"
+wait_for 10 maas_fixup
 
 salt -C 'mas01*' pillar.item\
   maas:region:admin:username \
   maas:region:admin:password
 
-# KVM, compute node prereqs (libvirt first), VCP deployment
-salt -C '* and not cfg01* and not mas01*' saltutil.sync_all
-
-salt -C 'kvm*' pkg.install bridge-utils
-salt -C 'kvm*' state.apply linux.network
-salt -C 'kvm*' system.reboot
-wait_for "! salt '*' test.ping | tee /dev/stderr | fgrep -q 'Not connected'"
-
-salt -C '* and not cfg01* and not mas01*' state.apply linux,ntp
-
-salt -C 'kvm*' state.sls libvirt
-
-salt -C '* and not cfg01* and not mas01*' state.apply salt
-salt -C 'kvm*' state.sls salt.control
-
-vcp_nodes=$(salt --out yaml 'kvm01*' pillar.get salt:control:cluster:internal:node | awk '/\s+\w+:$/ {gsub(/:$/, "*"); print $1}')
-
-# Check all vcp nodes are available
-rc=1
-while [ $rc -ne 0 ]; do
-  rc=0
-  for node in $vcp_nodes; do
-    salt "$node" test.ping 2>/dev/null 1>&2 || { rc=$?; break; };
-  done
-done
+# 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)"
 
-salt -C '* and not cfg01* and not mas01*' saltutil.sync_all
-salt -C '* and not cfg01* and not mas01*' state.apply salt
-wait_for "salt -C '* and not cfg01* and not mas01*' state.apply linux,ntp"
+wait_for 10.0 "salt -C '* and not cfg01* and not mas01*' saltutil.sync_all"