X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=reboot.py;h=337b0b46f359fc173bf4283ced02445510b49873;hb=944d143a6528c4157b71f51ed480aec806cbaa06;hp=495d366d990eae3bd9615f67db293834f291c216;hpb=f38420ff4bc9fb114db5f62b01cc8990dc0a0af2;p=monitor.git

diff --git a/reboot.py b/reboot.py
index 495d366..337b0b4 100755
--- a/reboot.py
+++ b/reboot.py
@@ -8,16 +8,16 @@ import os, sys
 import xml, xmlrpclib
 import errno, time, traceback
 import urllib2
+import urllib
 import threading, popen2
 import array, struct
-#from socket import *
-import socket
 import plc
 import base64
 from subprocess import PIPE, Popen
 import ssh.pxssh as pxssh
 import ssh.pexpect as pexpect
 import socket
+import moncommands 
 
 # Use our versions of telnetlib and pyssh
 sys.path.insert(0, os.path.dirname(sys.argv[0]))
@@ -115,6 +115,7 @@ class Transport:
 	TELNET = 1
 	SSH    = 2
 	HTTP   = 3
+	IPAL   = 4
 
 	TELNET_TIMEOUT = 60
 
@@ -123,10 +124,6 @@ class Transport:
 		self.verbose = verbose
 		self.transport = None
 
-#	def __del__(self):
-#		if self.transport:
-#			self.close()
-
 	def open(self, host, username=None, password=None, prompt="User Name"):
 		transport = None
 
@@ -235,8 +232,11 @@ class PCUControl(Transport,PCUModel,PCURecord):
 			elif '443' in supported_ports and self.portstatus['443'] == "open":
 				type = Transport.HTTP
 			elif '5869' in supported_ports and self.portstatus['5869'] == "open":
-				# For DRAC cards.  not sure how much it's used in the
-				# protocol.. but racadm opens this port.
+				# For DRAC cards. Racadm opens this port.
+				type = Transport.HTTP
+			elif '9100' in supported_ports and self.portstatus['9100'] == "open":
+				type = Transport.IPAL
+			elif '16992' in supported_ports and self.portstatus['16992'] == "open":
 				type = Transport.HTTP
 			else:
 				raise ExceptionPort("Unsupported Port: No transport from open ports")
@@ -275,38 +275,133 @@ class PCUControl(Transport,PCUModel,PCURecord):
 			import traceback
 			traceback.print_exc()
 			return "EOF connection reset" + str(err)
-		#except Exception, err:
-		#	if self.verbose:
-		#		logger.debug("reboot: Exception")
-		#		logger.debug(err)
-		#	if self.transport:
-		#		self.transport.close()
-		#	import traceback
-		#	traceback.print_exc()
-		#	return  "generic exception; unknown problem."
-
 		
 class IPAL(PCUControl):
+	""" 
+		This now uses a proprietary format for communicating with the PCU.  I
+		prefer it to Telnet, and Web access, since it's much lighter weight
+		and, more importantly, IT WORKS!! HHAHHHAHAHAHAHAHA!
+	"""
+
+	def format_msg(self, data, cmd):
+		esc = chr(int('1b',16))
+		return "%c%s%c%s%c" % (esc, self.password, esc, data, cmd) # esc, 'q', chr(4))
+	
+	def recv_noblock(self, s, count):
+		import errno
+
+		try:
+			# TODO: make sleep backoff, before stopping.
+			time.sleep(4)
+			ret = s.recv(count, socket.MSG_DONTWAIT)
+		except socket.error, e:
+			if e[0] == errno.EAGAIN:
+				raise Exception(e[1])
+			else:
+				# TODO: not other exceptions.
+				raise Exception(e)
+		return ret
+
 	def run(self, node_port, dryrun):
-		self.open(self.host)
+		import errno
 
-		# XXX Some iPals require you to hit Enter a few times first
-		self.ifThenSend("Password >", "\r\n\r\n", ExceptionNotFound)
+		power_on = False
 
-		# Login
-		self.ifThenSend("Password >", self.password, ExceptionPassword)
-		self.transport.write("\r\n\r\n")
+		print "open socket"
+		s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+		try:
+			print "connect"
+			s.connect((self.host, 9100))
+		except socket.error, e:
+			s.close()
+			if e[0] == errno.ECONNREFUSED:
+				# cannot connect to remote host
+				raise Exception(e[1])
+			else:
+				# TODO: what other conditions are there?
+				raise Exception(e)
+				
+		# get current status
+		print "Checking status"
+		s.send(self.format_msg("", 'O'))
+		ret = self.recv_noblock(s, 8)
+		print "Current status is '%s'" % ret
+
+		if ret == '':
+			raise Exception("Status returned 'another session already open' %s : %s" % (node_port, ret))
+			
+				
+		if node_port < len(ret):
+			status = ret[node_port]
+			if status == '1':
+				# up
+				power_on = True
+			elif status == '0':
+				# down
+				power_on = False
+			else:
+				raise Exception("Unknown status for PCU socket %s : %s" % (node_port, ret))
+		else:
+			raise Exception("Mismatch between configured port and PCU status: %s %s" % (node_port, ret))
+			
 
-		if not dryrun: # P# - Pulse relay
-			self.ifThenSend("Enter >", 
-							"P%d" % node_port, 
-							ExceptionNotFound)
-		# Get the next prompt
-		self.ifElse("Enter >", ExceptionTimeout)
+		if not dryrun:
+			if power_on:
+				print "Pulsing %s" % node_port
+				s.send(self.format_msg("%s" % node_port, 'P'))
+			else:
+				# NOTE: turn power on ; do not pulse the port.
+				print "Power was off, so turning on ..."
+				s.send(self.format_msg("%s" % node_port, 'E'))
+				#s.send(self.format_msg("%s" % node_port, 'P'))
+
+			print "Receiving response."
+			ret = self.recv_noblock(s, 8)
+			print "Current status is '%s'" % ret
+
+			if node_port < len(ret):
+				status = ret[node_port]
+				if status == '1':
+					# up
+					power_on = True
+				elif status == '0':
+					# down
+					power_on = False
+				else:
+					raise Exception("Unknown status for PCU socket %s : %s" % (node_port, ret))
+			else:
+				raise Exception("Mismatch between configured port and PCU status: %s %s" % (node_port, ret))
 
-		self.close()
+			if power_on:
+				return 0
+			else:
+				return "Failed Power On"
+
+		s.close()
 		return 0
 
+# TELNET version of protocol...
+#		#self.open(self.host)
+#		## XXX Some iPals require you to hit Enter a few times first
+#		#self.ifThenSend("Password >", "\r\n\r\n", ExceptionNotFound)
+#		# Login
+#		self.ifThenSend("Password >", self.password, ExceptionPassword)
+#		self.transport.write("\r\n\r\n")
+#		if not dryrun: # P# - Pulse relay
+#			print "node_port %s" % node_port
+#			self.ifThenSend("Enter >", 
+#							"P7", # % node_port, 
+#							ExceptionNotFound)
+#			print "send newlines"
+#			self.transport.write("\r\n\r\n")
+#			print "after new lines"
+#		# Get the next prompt
+#		print "wait for enter"
+#		self.ifElse("Enter >", ExceptionTimeout)
+#		print "closing "
+#		self.close()
+#		return 0
+
 class APCEurope(PCUControl):
 	def run(self, node_port, dryrun):
 		self.open(self.host, self.username)
@@ -461,6 +556,23 @@ class APC(PCUControl):
 		else:
 			return ret
 
+class IntelAMT(PCUControl):
+	def run(self, node_port, dryrun):
+
+		cmd = moncommands.CMD()
+		#[cmd_str = "IntelAMTSDK/Samples/RemoteControl/remoteControl"
+		cmd_str = "cmdamt/remoteControl"
+
+		if dryrun:
+			# NOTE: -p checks the power state of the host.
+			# TODO: parse the output to find out if it's ok or not.
+			cmd_str += " -p http://%s:16992/RemoteControlService  -user admin -pass '%s' " % (self.host, self.password )
+		else:
+			cmd_str += " -A http://%s:16992/RemoteControlService -user admin -pass '%s' " % (self.host, self.password )
+			
+		print cmd_str
+		return cmd.system(cmd_str, self.TELNET_TIMEOUT)
+
 class DRACRacAdm(PCUControl):
 	def run(self, node_port, dryrun):
 
@@ -511,9 +623,8 @@ class HPiLO(PCUControl):
 		
 class HPiLOHttps(PCUControl):
 	def run(self, node_port, dryrun):
-		import soltesz
 
-		locfg = soltesz.CMD()
+		locfg = moncommands.CMD()
 		cmd = "cmdhttps/locfg.pl -s %s -f %s -u %s -p '%s' | grep 'MESSAGE' | grep -v 'No error'" % (
 					self.host, "iloxml/Get_Network.xml", 
 					self.username, self.password)
@@ -524,7 +635,7 @@ class HPiLOHttps(PCUControl):
 			return sout.strip()
 
 		if not dryrun:
-			locfg = soltesz.CMD()
+			locfg = moncommands.CMD()
 			cmd = "cmdhttps/locfg.pl -s %s -f %s -u %s -p '%s' | grep 'MESSAGE' | grep -v 'No error'" % (
 						self.host, "iloxml/Reset_Server.xml", 
 						self.username, self.password)
@@ -593,7 +704,9 @@ class BayTechCtrlCUnibe(PCUControl):
 
 		# Control Outlets  (5 ,1).........5
 		try:
-			index = s.expect(["Enter Request :"])
+			print s
+			print "Enter Request" in s.before
+			index = s.expect("Enter Request")
 
 			if index == 0:
 				print "3"
@@ -673,7 +786,10 @@ class BayTechCtrlC(PCUControl):
 							print "sending Y"
 							s.send("Y\r\n")
 
-				index = s.expect(["DS-RPC>"])
+				# NOTE: for some reason, the script times out with the
+				# following line.  In manual tests, it works correctly, but
+				# with automated tests, evidently it fails.
+				#index = s.expect(["DS-RPC>"])
 				#print "got prompt back"
 
 			s.close()
@@ -681,7 +797,7 @@ class BayTechCtrlC(PCUControl):
 		except pexpect.EOF:
 			raise ExceptionPrompt("EOF before 'Enter Request' Prompt")
 		except pexpect.TIMEOUT:
-			raise ExceptionPrompt("Timeout before 'Enter Request' Prompt")
+			raise ExceptionPrompt("Timeout before Prompt")
 
 		return 0
 
@@ -770,19 +886,41 @@ class ePowerSwitchGood(PCUControl):
 			# failing here means the User/passwd is wrong (hopefully)
 			raise ExceptionPassword("Incorrect username/password")
 
-		# TODO: after verifying that the user/password is correct, we should
-		# actually reboot the given node.
-
+		# NOTE: after verifying that the user/password is correct, 
+		# 		actually reboot the given node.
 		if not dryrun:
-			# add data to handler,
-			# fetch url one more time on cmd.html, econtrol.html or whatever.
-			pass
+			try:
+				data = urllib.urlencode({'P%d' % node_port : "r"})
+				req = urllib2.Request(self.url + "cmd.html")
+				req.add_header("Authorization", authheader)
+				# add data to handler,
+				f = urllib2.urlopen(req, data)
+				if self.verbose: print f.read()
+			except:
+				import traceback; traceback.print_exc()
 
-		if self.verbose: print f.read()
+				# fetch url one more time on cmd.html, econtrol.html or whatever.
+				# pass
+		else:
+			if self.verbose: print f.read()
 
 		self.close()
 		return 0
 
+class CustomPCU(PCUControl):
+	def run(self, node_port, dryrun):
+		url = "https://www-itec.uni-klu.ac.at/plab-pcu/index.php" 
+
+		if not dryrun:
+			# Turn host off, then on
+			formstr = "plab%s=off" % node_port
+			os.system("curl --user %s:%s --form '%s' --insecure %s" % (self.username, self.password, formstr, url))
+			time.sleep(5)
+			formstr = "plab%s=on" % node_port
+			os.system("curl --user %s:%s --form '%s' --insecure %s" % (self.username, self.password, formstr, url))
+		else:
+			os.system("curl --user %s:%s --insecure %s" % (self.username, self.password, url))
+
 
 class ePowerSwitchOld(PCUControl):
 	def run(self, node_port, dryrun):
@@ -1027,12 +1165,16 @@ def pcu_name(pcu):
 	else:
 		return None
 
-import soltesz
-fb =soltesz.dbLoad("findbadpcus")
+#import database
+from monitor import database
+fb = None
 
 def get_pcu_values(pcu_id):
-	# TODO: obviously, this shouldn't be loaded each time...
-
+	global fb
+	if fb == None:
+		# this shouldn't be loaded each time...
+		fb = database.dbLoad("findbadpcus")
+		
 	try:
 		values = fb['nodes']["id_%s" % pcu_id]['values']
 	except:
@@ -1045,7 +1187,6 @@ def reboot(nodename):
 	
 def reboot_policy(nodename, continue_probe, dryrun):
 	global verbose
-	print "this is a test of reboot_policy()"
 
 	pcu = plc.getpcu(nodename)
 	if not pcu:
@@ -1062,7 +1203,6 @@ def reboot_policy(nodename, continue_probe, dryrun):
 	# Try the PCU first
 	logger.debug("Trying PCU %s %s" % (pcu['hostname'], pcu['model']))
 
-	print "reboot_test"
 	ret = reboot_test(nodename, values, continue_probe, verbose, dryrun)
 
 	if ret != 0:
@@ -1077,12 +1217,13 @@ def reboot_test(nodename, values, continue_probe, verbose, dryrun):
 
 	try:
 		# DataProbe iPal (many sites)
-		if  continue_probe and values['model'].find("Dataprobe IP-41x/IP-81x") >= 0:
-			ipal = IPAL(values, verbose, ['23'])
+		if  continue_probe and values['model'].find("IP-41x_IP-81x") >= 0:
+			ipal = IPAL(values, verbose, ['23', '80', '9100'])
 			rb_ret = ipal.reboot(values[nodename], dryrun)
 				
 		# APC Masterswitch (Berkeley)
-		elif continue_probe and values['model'].find("APC AP79xx/Masterswitch") >= 0:
+		elif continue_probe and ( values['model'].find("AP79xx") >= 0 or \
+								  values['model'].find("Masterswitch") >= 0 ):
 			print values
 
 			# TODO: make a more robust version of APC
@@ -1107,7 +1248,7 @@ def reboot_test(nodename, values, continue_probe, verbose, dryrun):
 				rb_ret = apc.reboot(values[nodename], dryrun)
 
 		# BayTech DS4-RPC
-		elif continue_probe and values['model'].find("Baytech DS4-RPC") >= 0:
+		elif continue_probe and values['model'].find("DS4-RPC") >= 0:
 			if values['pcu_id'] in [1237,1052,1209,1002,1008,1041,1013,1022]:
 				# These  require a 'ctrl-c' to be sent... 
 				baytech = BayTechCtrlC(values, verbose, ['22', '23'])
@@ -1136,7 +1277,7 @@ def reboot_test(nodename, values, continue_probe, verbose, dryrun):
 				rb_ret = baytech.reboot(values[nodename], dryrun)
 
 		# iLO
-		elif continue_probe and values['model'].find("HP iLO") >= 0:
+		elif continue_probe and values['model'].find("ilo") >= 0:
 			try:
 				hpilo = HPiLO(values, verbose, ['22'])
 				rb_ret = hpilo.reboot(0, dryrun)
@@ -1148,7 +1289,7 @@ def reboot_test(nodename, values, continue_probe, verbose, dryrun):
 				rb_ret = hpilo.reboot(0, dryrun)
 
 		# DRAC ssh
-		elif continue_probe and values['model'].find("Dell RAC") >= 0:
+		elif continue_probe and values['model'].find("DRAC") >= 0:
 			# TODO: I don't think DRACRacAdm will throw an exception for the
 			# default method to catch...
 			try:
@@ -1162,20 +1303,26 @@ def reboot_test(nodename, values, continue_probe, verbose, dryrun):
 				wti = WTIIPS4(values, verbose, ['23'])
 				rb_ret = wti.reboot(values[nodename], dryrun)
 
-		# BlackBox PSExxx-xx (e.g. PSE505-FR)
-		elif continue_probe and \
-			(values['model'].find("BlackBox PS5xx") >= 0 or
-			 values['model'].find("ePowerSwitch 1/4/8x") >=0 ):
+		elif continue_probe and values['model'].find("AMT") >= 0:
+				amt = IntelAMT(values, verbose, ['16992'])
+				rb_ret = amt.reboot(values[nodename], dryrun)
 
+		# BlackBox PSExxx-xx (e.g. PSE505-FR)
+		elif continue_probe and values['model'].find("ePowerSwitch") >=0:
 			# TODO: allow a different port than http 80.
 			if values['pcu_id'] in [1089, 1071, 1046, 1035, 1118]:
 				eps = ePowerSwitchGood(values, verbose, ['80'])
 			elif values['pcu_id'] in [1003]:
+				# OLD EPOWER
+				print "OLD EPOWER"
 				eps = ePowerSwitch(values, verbose, ['80'])
 			else:
 				eps = ePowerSwitchGood(values, verbose, ['80'])
 
 			rb_ret = eps.reboot(values[nodename], dryrun)
+		elif continue_probe and values['pcu_id'] in [1122]:
+			custom = CustomPCU(values, verbose, ['80', '443'])
+			custom.reboot(values[nodename], dryrun)
 
 		elif continue_probe:
 			rb_ret = "Unsupported_PCU"