FEATURE: Halt the install (sles) with an error message on the console and in the message queue if we are unable to create a RAID. To override, set the attribute 'halt_install_on_error=False'.

Author: bsanders

common/src/stack/storage-config/bin/configure-controllers.py

@@ -8,6 +8,7 @@
 #
 
 import sys
+from subprocess import CalledProcessError
 
 sys.path.append('/tmp')
 from stack_site import *
@@ -71,15 +72,13 @@ def getController():
 ## MAIN
 ##
 
-if 'nukecontroller' in attributes:
-	nukecontroller = str2bool(attributes['nukecontroller'])
-else:
-	nukecontroller = False
+# Halt the install with an error message on the console and in the message queue
+# if we are unable to create a RAID.
+# To override, set the attribute 'halt_install_on_error=False'.
 
-if 'secureerase' in attributes:
-	secureerase = str2bool(attributes['secureerase'])
-else:
-	secureerase = False
+halt_on_error  = str2bool(attributes.get('halt_install_on_error', True))
+nukecontroller = str2bool(attributes.get('nukecontroller', False))
+secureerase    = str2bool(attributes.get('secureerase', False))
 
 #
 # if 'secureerase' is true, then that implies that 'nukecontroller' is true
@@ -181,11 +180,15 @@ def getController():
 		#
 		continue
 
-	if raidlevel == 'hotspare' and arrayid == 'global':
-		ctrl.doGlobalHotSpare(adapter, enclosure, hotspares, options)
-	else:
-		ctrl.doRaid(raidlevel, adapter, enclosure, slots, hotspares,
-			options)
+	try:
+		if raidlevel == 'hotspare' and arrayid == 'global':
+			ctrl.doGlobalHotSpare(adapter, enclosure, hotspares, options, check=halt_on_error)
+		else:
+			ctrl.doRaid(raidlevel, adapter, enclosure, slots, hotspares, options, check=halt_on_error)
+	except CalledProcessError as e:
+		print(' '.join(e.cmd))
+		print(f'output: {e.stdout}')
+		sys.exit(1)
 
 	for slot in slots + hotspares:
 		try:
@@ -235,26 +238,30 @@ def getController():
 			enclosure = ctrl.getEnclosure(adapter)
 
 		if 'arrayid' in o.keys():
-			if o['arrayid'] == '*':
-				if raidlevel == '1':
-					#
-					# special case for arrayid == '*' and raidlevel 1 -- put the remaining
-					# disks into RAID 1 pairs
-					#
-					while len(freeslots[adapter]) > 1:
-						disks = [ freeslots[adapter][0], freeslots[adapter][1] ]
-						ctrl.doRaid(1, adapter, enclosure, disks, [], options)
-
-						freeslots[adapter].remove(disks[0])
-						freeslots[adapter].remove(disks[1])
+			try:
+				if o['arrayid'] == '*':
+					if raidlevel == '1':
+						#
+						# special case for arrayid == '*' and raidlevel 1 -- put the remaining
+						# disks into RAID 1 pairs
+						#
+						while len(freeslots[adapter]) > 1:
+							disks = [ freeslots[adapter][0], freeslots[adapter][1] ]
+							ctrl.doRaid(1, adapter, enclosure, disks, [], options, check=halt_on_error)
+
+							freeslots[adapter].remove(disks[0])
+							freeslots[adapter].remove(disks[1])
+					else:
+						# JBOD Mode for the remainder
+						for slot in freeslots[adapter]:
+							ctrl.doRaid(0, adapter, enclosure, [ slot ],
+								[], options, check=halt_on_error)
+
+				# RAID mode - Single array for remaining disks
 				else:
-					# JBOD Mode for the remainder
-					for slot in freeslots[adapter]:
-						ctrl.doRaid(0, adapter, enclosure, [ slot ],
-							[], options)
-
-			# RAID mode - Single array for remaining disks
-			else:
-				ctrl.doRaid(raidlevel, adapter, enclosure, freeslots[adapter],
-					hotspares, options)
-
+					ctrl.doRaid(raidlevel, adapter, enclosure, freeslots[adapter],
+						hotspares, options, check=halt_on_error)
+			except CalledProcessError as e:
+				print(' '.join(e.cmd))
+				print(f'output: {e.stdout}')
+				sys.exit(1)

common/src/stack/storage-config/lib/controller_hpssacli.py

@@ -6,24 +6,29 @@
 class CLI:
 	debug = 0
 
-	def run(self, args):
+	def run(self, args, check=False):
+		'''
+		check=False was added to allow this subprocess wrapper to halt the install on error.
+		currently, only commands related to creating RAID volumes will ever have check=True
+		other commands may be expected to fail
+		'''
+
 		if not os.path.exists('/opt/stack/sbin/hpssacli'):
 			return []
 
 		cmd = [ '/opt/stack/sbin/hpssacli', 'ctrl' ]
 		cmd.extend(args)
 
-		file = open('/tmp/hpssacli.log', 'a')
-		file.write('cmd: %s\n' % ' '.join(cmd))
+		with open('/tmp/hpssacli.log', 'a') as fi:
+			fi.write('cmd: %s\n' % ' '.join(cmd))
 
-		p = subprocess.run(cmd, stdin=subprocess.PIPE)
-		result = p.stdout.decode()
+			p = subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, encoding='utf-8', check=check)
+			result = p.stdout
 
-		file.write('result:\n')
-		for line in result:
-			file.write('%s' % line)
-		file.write('\n\n')
-		file.close()
+			fi.write('result:\n')
+			for line in result:
+				fi.write('%s' % line)
+			fi.write('\n\n')
 
 		return result
 
@@ -165,9 +170,7 @@ def doSecureErase(self, enclosure, adapter, slot):
 		#
 		return
 
-	def doRaid(self, raidlevel, adapter, enclosure, slots, hotspares,
-			flags):
-
+	def doRaid(self, raidlevel, adapter, enclosure, slots, hotspares, flags, check):
 		drives = []
 		for slot in slots:
 			#
@@ -201,7 +204,7 @@ def doRaid(self, raidlevel, adapter, enclosure, slots, hotspares,
 			f = flags.split()
 			cmd.extend(f)
 
-		result = self.run(cmd)
+		result = self.run(cmd, check=check)
 
 		if hotspares:
 			#
@@ -227,9 +230,9 @@ def doRaid(self, raidlevel, adapter, enclosure, slots, hotspares,
 
 				cmd = [ 'slot=%d' % adapter, 'array', arrayid,
 					'add', 'spares=%s' % ','.join(spares) ]
-				result = self.run(cmd)
+				result = self.run(cmd, check=check)
 
-	def doGlobalHotSpare(self, adapter, enclosure, hotspares, options):
+	def doGlobalHotSpare(self, adapter, enclosure, hotspares, options, check):
 		spares = []
 		for slot in hotspares:
 			enclosure = self.getEnclosure(adapter, slot)
@@ -241,5 +244,11 @@ def doGlobalHotSpare(self, adapter, enclosure, hotspares, options):
 		if options:
 			f = options.split()
 			cmd.extend(f)
-		result = self.run(cmd)
-
+		result = self.run(cmd, check=check)
+
+if __name__ == '__main__':
+	s = CLI()
+	a = s.getAdapter()
+	if a is not None:
+		print(s.getEnclosure(a))
+		print(s.getSlots(a))

common/src/stack/storage-config/lib/controller_megacli.py

@@ -9,19 +9,24 @@ class CLI:
 		'cached', 'direct', 'endskcache', 'disdskcache',
 		'cachedbadbbu', 'nocachedbadbbu' ]
 
-	def run(self, args):
+	def run(self, args, check=False):
+		'''
+		check=False was added to allow this subprocess wrapper to halt the install on error.
+		currently, only commands related to creating RAID volumes will ever have check=True
+		other commands may be expected to fail
+		'''
+
 		cmd = [ '/opt/stack/sbin/MegaCli' ]
 		cmd.extend(args)
-		file = open('/tmp/MegaCli.log', 'a+')
-		file.write('cmd: %s\n' % ' '.join(cmd))
-		file.close()
+		with open('/tmp/MegaCli.log', 'a+') as fi:
+			fi.write('cmd: %s\n' % ' '.join(cmd))
 		cmd.extend(['-AppLogFile','/tmp/MegaCli.log'])
 
 		result = []
 
-		p = subprocess.run(cmd, stdout=subprocess.PIPE)
+		p = subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, encoding='utf-8', check=check)
 
-		for line in p.stdout.decode().splitlines():
+		for line in p.stdout.splitlines():
 			tokens = line.split(':', 1)
 			if len(tokens) != 2:
 				continue
@@ -82,8 +87,7 @@ def getSlots(self, adapter):
 
 		return slots
 
-	def doStrippedRaid(self, raidlevel, adapter, enclosure, slots,
-			hotspares, flags):
+	def doStrippedRaid(self, raidlevel, adapter, enclosure, slots, hotspares, flags, check):
 
 		if raidlevel == '10':
 			cmd = [ '-CfgSpanAdd', '-r10' ]
@@ -139,7 +143,7 @@ def doStrippedRaid(self, raidlevel, adapter, enclosure, slots,
 
 		cmd.append('-force')
 		cmd.append('-a%d' % adapter)
-		results = self.run(cmd)
+		results = self.run(cmd, check=check)
 
 		#
 		# apply any flags that are not able to be set in
@@ -159,7 +163,7 @@ def doStrippedRaid(self, raidlevel, adapter, enclosure, slots,
 			for f in setpropflags:
 				cmd = [ '-LDSetProp', f, '-L%d' % vid, 
 					'-a%d' % adapter ]
-				results = self.run(cmd)
+				results = self.run(cmd, check=check)
 
 		# 
 		# support for dedicated hot spares for 10, 50 and 60
@@ -170,8 +174,7 @@ def doStrippedRaid(self, raidlevel, adapter, enclosure, slots,
 			' '.join(options))
 
 
-	def doRaid(self, raidlevel, adapter, enclosure, slots, hotspares,
-			flags):
+	def doRaid(self, raidlevel, adapter, enclosure, slots, hotspares, flags, check):
 
 		if raidlevel in [ '10', '50', '60' ]:
 			self.doStrippedRaid(raidlevel, adapter, enclosure,
@@ -207,7 +210,7 @@ def doRaid(self, raidlevel, adapter, enclosure, slots, hotspares,
 
 			cmd.append('-force')
 			cmd.append('-a%d' % adapter)
-			results = self.run(cmd)
+			results = self.run(cmd, check=check)
 
 			#
 			# apply any flags that are not able to be set in
@@ -226,10 +229,10 @@ def doRaid(self, raidlevel, adapter, enclosure, slots, hotspares,
 			for f in setpropflags:
 				cmd = [ '-LDSetProp', f, '-L%d' % vid, 
 					'-a%d' % adapter ]
-				results = self.run(cmd)
+				results = self.run(cmd, check=check)
 
 
-	def doGlobalHotSpare(self, adapter, enclosure, hotspares, flags):
+	def doGlobalHotSpare(self, adapter, enclosure, hotspares, flags, check):
 		for hotspare in hotspares:
 			cmd = [ '-PDHSP', '-Set', '-PhysDrv',
 				'[%s:%d]' % (enclosure, hotspare),
@@ -238,5 +241,11 @@ def doGlobalHotSpare(self, adapter, enclosure, hotspares, flags):
 			if flags:
 				cmd.append(flags)
 
-			self.run(cmd)
+			self.run(cmd, check=check)
 
+if __name__ == '__main__':
+	s = CLI()
+	a = s.getAdapter()
+	if a is not None:
+		print(s.getEnclosure(a))
+		print(s.getSlots(a))

common/src/stack/storage-config/lib/controller_storcli.py

@@ -7,7 +7,13 @@
 
 class CLI:
 
-	def run(self, args, json_out = False):
+	def run(self, args, json_out=False, check=False):
+		'''
+		check=False was added to allow this subprocess wrapper to halt the install on error.
+		currently, only commands related to creating RAID volumes will ever have check=True
+		other commands may be expected to fail
+		'''
+
 		cmd = [ '/opt/stack/sbin/storcli' ]
 		cmd.extend(args)
 		try:
@@ -19,12 +25,12 @@ def run(self, args, json_out = False):
 		if json_out:
 			cmd.append('J')
 
-		f = open('/tmp/MegaSAS.log','a')
-		f.write('cmd: %s\n\n'  % ' '.join(cmd))
-		p = subprocess.run(cmd, stdout=subprocess.PIPE)
-		o = p.stdout.decode()
-		f.write('%s\n\n' % o)
-		f.close()
+		with open('/tmp/MegaSAS.log','a') as fi:
+			fi.write('cmd: %s\n\n'  % ' '.join(cmd))
+			p = subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, encoding='utf-8', check=check)
+			o = p.stdout
+			fi.write('%s\n\n' % o)
+
 		if json_out:
 			j = json.loads(o)
 			result = j['Controllers'][0]
@@ -115,8 +121,7 @@ def getSlots(self, adapter):
 
 		return slots
 
-	def doRaid(self, raidlevel, adapter, enclosure, slots, hotspares,
-			flags):
+	def doRaid(self, raidlevel, adapter, enclosure, slots, hotspares, flags, check):
 		args = ['/c%d' % adapter, 'add','vd',
 			'type=r%s' % raidlevel]
 
@@ -174,9 +179,9 @@ def doRaid(self, raidlevel, adapter, enclosure, slots, hotspares,
 			args.append('spares=%s' % ','.join(hs))
 
 		args.append('force')
-		self.run(args)
+		self.run(args, check=check)
 
-	def doGlobalHotSpare(self, adapter, enclosure, hotspares, options):
+	def doGlobalHotSpare(self, adapter, enclosure, hotspares, options, check):
 		if enclosure:
 			loc = '/c%d/e%d' % (adapter, enclosure)
 		else:
@@ -187,11 +192,11 @@ def doGlobalHotSpare(self, adapter, enclosure, hotspares, options):
 			if options:
 				f = options.split()
 				args.extend(f)
-			self.run(args)
+			self.run(args, check=check)
 
 
 if __name__ == '__main__':
-	s = StorCLI()
+	s = CLI()
 	a = s.getAdapter()
 	if a is not None:
 		print(s.getEnclosure(a))

sles/src/stack/images/common/sles-stacki.img-patches/usr/lib/YaST2/startup/First-Stage/F08-stacki

@@ -59,6 +59,17 @@ echo "tracker = ${FRONTEND}:3825" > /tmp/stack.conf
 # configure the hardware disk array controller first
 #
 /opt/stack/bin/configure-controllers.py
+if [ $? -ne 0 ]
+then
+	touch /tmp/wait
+	echo "Unable to complete storage controller configuration."
+	while [ -f /tmp/wait ]
+	do
+		sleep 1
+		/opt/stack/bin/smq-publish -chealth '{"state": "controller config failed - check console or reinstall with halt_install_on_error=False"}'
+	done
+fi
+
 
 udevadm settle --timeout=60