#!/usr/bin/env python
#
# This script will compare the master and slave databases, one table at a time, to determine
# whether they are synchronized or not.  Lines are printed with a leading "x" or "o", inspired
# by the Japanese associate of "X" meaning "no" and "O" meaning "OK".  Any line starting with
# an "x" will follow with details about the table not being in sync.
#
import psycopg2
import time
import sys
import getopt
import signal
import string

# Database Parameters
db_host = 'localhost'
db_name = 'marian'
db_user = 'marian'
db_password = 'marian'

# Miscelaneous variables
PROGNAME = "check_slony_lag"
VERSION = "1.0.0"
opt_timeout = 10
opt_warning = -1
opt_critical = -1
opt_debug = False

# function: debug_print
# ============================================================================
def debug_print(txt):
	"""Lets us turn off print statements that are otherwise unimportant."""
	if opt_debug:
		print txt

# Dictionary of Nagios Plug-in return-values
ERRORS={  'OK':0,
	'WARNING':1,
	'CRITICAL':2,
	'UNKNOWN':3,
	'DEPENDENT':4
};

# Options that have to be handled by defau
# --help, --version, --timeout, --warning, --critical
def usage():
	print "%s: [--help] [--version] [--timeout=n] --warning=n --critical=n" % PROGNAME
	print ""
	print " -h, --help                     This help message"
	print " -d, --debug                    Enable debugging output"
	print " -v, --version                  Version of this program"
	print " -t, --timeout=<secs>           Max amount of time to execute"
	print " -w, --warning=<warn percent>   (required) Warning number of lagged events"
	print " -c, --critical=<crit percent>  (required) Critical number of lagged events"
	print ""
	print "Example:"
	print ""
	print " %s --timeout=20 --warning=80 --critical=90" % PROGNAME
	print ""

try:
	opts, args = getopt.getopt(sys.argv[1:], "hvdt:w:c:", ["help", "version", "debug", "timeout=", "warning=", "critical="])
except getopt.GetoptError, err:
	print str(err)
	usage()
	sys.exit(ERRORS['UNKNOWN'])

for opt, val in opts:
	if opt in ("-v", "--version"):
		print "%s v%s" % (PROGNAME, VERSION)
	elif opt in ("-t", "--timeout"):
		opt_timeout = int(val)
	elif opt in ("-w", "--warning"):
		opt_warning = int(val)
	elif opt in ("-c", "--critical"):
		opt_critical = int(val)
	elif opt in ("-d", "--debug"):
		opt_debug = True
	elif opt in ("-h", "--help"):
		usage()
		sys.exit(ERRORS['UNKNOWN'])
	else:
		assert False, "unknown option"

if opt_critical == -1 or opt_warning == -1:
	print "Error: Make sure you specify a '--warning' AND '--critical'"
	sys.exit(ERRORS['UNKNOWN'])
	
if opt_critical < opt_warning:
	print "Error: The 'warning' value should be less than 'critical' value"
	sys.exit(ERRORS['UNKNOWN'])

def handler(signum, frame):
	print "%s UNKNOWN: Timed out after %d seconds" % (PROGNAME, opt_timeout)
	sys.exit(ERRORS['UNKNOWN'])
	
# Set the signal handler and a 5-second alarm
signal.signal(signal.SIGALRM, handler)
signal.alarm(opt_timeout)

# Connect to the database
db = psycopg2.connect("user=" + db_user + " password=" + db_password + " dbname=" + db_name + " host=" + db_host)

# Prepare our database pointer
db_cursor = db.cursor()

try:
	# Perform a query
	# This is hard-coded for the _slony_erp schema
	db_cursor.execute("select st_lag_num_events from _slony_erp.sl_status")
except psycopg2.ProgrammingError:
	returnVal = ERRORS['CRITICAL']
	returnStr = "%s CRITICAL: SQL Query returned zero records!" % (PROGNAME)
else:
	# Get the results
	db_row = db_cursor.fetchone()
	
	debug_print("db_row: %s" % (db_row,))
	
	if db_row == None:
		# Set counter to 0
		st_lag_num_events = 0
	else:
		# Save the value
		st_lag_num_events = db_row[0]
	
	debug_print("st_lag_num_events: %d" % (st_lag_num_events,))
	
	if st_lag_num_events >= opt_critical:
		returnVal = ERRORS['CRITICAL']
		returnStr = "%s CRITICAL: Replication is lagged by %d events!" % (PROGNAME, st_lag_num_events)
	
	elif st_lag_num_events >= opt_warning:
		returnVal = ERRORS['WARNING']
		returnStr = "%s WARNING: %d events waiting to be replicated." % (PROGNAME, st_lag_num_events)
	
	else:
		returnVal = ERRORS['OK']
		if st_lag_num_events == 0:
			returnStr = "%s OK: No replicated events in queue." % (PROGNAME,)
		elif st_lag_num_events == 1:
			returnStr = "%s OK: Replication is only lagged by 1 event." % (PROGNAME,)
		else:
			returnStr = "%s OK: Replication is only lagged by %d events." % (PROGNAME, st_lag_num_events)

# Print the status of this check
print returnStr

# Exit with our return value
sys.exit(returnVal)