#!/usr/bin/env python3 # # A MongoDB Nagios check script # # Script idea taken from a Tag1 script I found and I modified it a lot # # Main Author # - Mike Zupan # Contributers # - Frank Brandewiede # - Sam Perman # - Shlomo Priymak # - @jhoff909 on github # - @jbraeuer on github # - Dag Stockstad # - @Andor on github # - Steven Richards - Captainkrtek on github # - Max Vernimmen - @mvernimmen-CG / @mvernimmen on github # - Kris Nova - @kris@nivenly.com github.com/kris-nova # - Jan Kantert - firstname@lastname.net # # USAGE # # See the README.md # from __future__ import print_function from __future__ import division import sys import time import optparse import re import os import numbers try: import pymongo except ImportError as e: print(e) sys.exit(2) # As of pymongo v 1.9 the SON API is part of the BSON package, therefore attempt # to import from there and fall back to pymongo in cases of older pymongo if pymongo.version >= "1.9": import bson.son as son else: import pymongo.son as son # # thanks to http://stackoverflow.com/a/1229667/72987 # def optional_arg(arg_default): def func(option, opt_str, value, parser): if parser.rargs and not parser.rargs[0].startswith('-'): val = parser.rargs[0] parser.rargs.pop(0) else: val = arg_default setattr(parser.values, option.dest, val) return func def performance_data(perf_data, params): data = '' if perf_data: data = " |" for p in params: p += (None, None, None, None) param, param_name, warning, critical = p[0:4] data += "%s=%s" % (param_name, str(param)) if warning or critical: warning = warning or 0 critical = critical or 0 data += ";%s;%s" % (warning, critical) data += " " return data def numeric_type(param): return param is None or isinstance(param, numbers.Real) def check_levels(param, warning, critical, message, ok=[]): if (numeric_type(critical) and numeric_type(warning)): if param >= critical: print("CRITICAL - " + message) sys.exit(2) elif param >= warning: print("WARNING - " + message) sys.exit(1) else: print("OK - " + message) sys.exit(0) else: if param in critical: print("CRITICAL - " + message) sys.exit(2) if param in warning: print("WARNING - " + message) sys.exit(1) if param in ok: print("OK - " + message) sys.exit(0) # unexpected param value print("CRITICAL - Unexpected value : %d" % param + "; " + message) return 2 def get_server_status(con): try: set_read_preference(con.admin) data = con.admin.command(pymongo.son_manipulator.SON([('serverStatus', 1)])) except: data = con.admin.command(son.SON([('serverStatus', 1)])) return data def main(argv): p = optparse.OptionParser(conflict_handler="resolve", description="This Nagios plugin checks the health of mongodb.") p.add_option('-H', '--host', action='store', type='string', dest='host', default='127.0.0.1', help='The hostname you want to connect to') p.add_option('-h', '--host-to-check', action='store', type='string', dest='host_to_check', default=None, help='The hostname you want to check (if this is different from the host you are connecting)') p.add_option('-P', '--port', action='store', type='int', dest='port', default=27017, help='The port mongodb is running on') p.add_option('--port-to-check', action='store', type='int', dest='port_to_check', default=None, help='The port you want to check (if this is different from the port you are connecting)') p.add_option('-u', '--user', action='store', type='string', dest='user', default=None, help='The username you want to login as') p.add_option('-p', '--pass', action='store', type='string', dest='passwd', default=None, help='The password you want to use for that user') p.add_option('-W', '--warning', action='store', dest='warning', default=None, help='The warning threshold you want to set') p.add_option('-C', '--critical', action='store', dest='critical', default=None, help='The critical threshold you want to set') p.add_option('-A', '--action', action='store', type='choice', dest='action', default='connect', help='The action you want to take', choices=['connect', 'connections', 'replication_lag', 'replication_lag_percent', 'replset_state', 'memory', 'memory_mapped', 'lock', 'flushing', 'last_flush_time', 'index_miss_ratio', 'databases', 'collections', 'database_size', 'database_indexes', 'collection_documents', 'collection_indexes', 'collection_size', 'collection_storageSize', 'queues', 'oplog', 'journal_commits_in_wl', 'write_data_files', 'journaled', 'opcounters', 'current_lock', 'replica_primary', 'page_faults', 'asserts', 'queries_per_second', 'page_faults', 'chunks_balance', 'connect_primary', 'collection_state', 'row_count', 'replset_quorum']) p.add_option('--max-lag', action='store_true', dest='max_lag', default=False, help='Get max replication lag (for replication_lag action only)') p.add_option('--mapped-memory', action='store_true', dest='mapped_memory', default=False, help='Get mapped memory instead of resident (if resident memory can not be read)') p.add_option('-D', '--perf-data', action='store_true', dest='perf_data', default=False, help='Enable output of Nagios performance data') p.add_option('-d', '--database', action='store', dest='database', default='admin', help='Specify the database to check') p.add_option('--all-databases', action='store_true', dest='all_databases', default=False, help='Check all databases (action database_size)') p.add_option('-s', '--ssl', dest='ssl', default=False, action='callback', callback=optional_arg(True), help='Connect using SSL') p.add_option('-r', '--replicaset', dest='replicaset', default=None, action='callback', callback=optional_arg(True), help='Connect to replicaset') p.add_option('-q', '--querytype', action='store', dest='query_type', default='query', help='The query type to check [query|insert|update|delete|getmore|command] from queries_per_second') p.add_option('-c', '--collection', action='store', dest='collection', default='admin', help='Specify the collection to check') p.add_option('-T', '--time', action='store', type='int', dest='sample_time', default=1, help='Time used to sample number of pages faults') p.add_option('-M', '--mongoversion', action='store', type='choice', dest='mongo_version', default='2', help='The MongoDB version you are talking with, either 2 or 3', choices=['2','3']) p.add_option('-a', '--authdb', action='store', type='string', dest='authdb', default='admin', help='The database you want to authenticate against') p.add_option('--ssl-ca-cert-file', action='store', type='string', dest='ssl_ca_cert_file', default=None, help='Path to Certificate Authority file for SSL') p.add_option('-f', '--ssl-cert-file', action='store', type='string', dest='cert_file', default=None, help='Path to PEM encoded key and cert for client authentication') p.add_option('-m','--auth-mechanism', action='store', type='choice', dest='auth_mechanism', default=None, help='Auth mechanism used for auth with mongodb', choices=['MONGODB-X509','SCRAM-SHA-256','SCRAM-SHA-1']) options, arguments = p.parse_args() host = options.host host_to_check = options.host_to_check if options.host_to_check else options.host port = options.port port_to_check = options.port_to_check if options.port_to_check else options.port user = options.user passwd = options.passwd authdb = options.authdb query_type = options.query_type collection = options.collection sample_time = options.sample_time if (options.action == 'replset_state'): warning = str(options.warning or "") critical = str(options.critical or "") else: warning = float(options.warning or 0) critical = float(options.critical or 0) action = options.action perf_data = options.perf_data max_lag = options.max_lag mongo_version = options.mongo_version database = options.database ssl = options.ssl replicaset = options.replicaset ssl_ca_cert_file = options.ssl_ca_cert_file cert_file = options.cert_file auth_mechanism = options.auth_mechanism if action == 'replica_primary' and replicaset is None: return "replicaset must be passed in when using replica_primary check" elif not action == 'replica_primary' and replicaset: return "passing a replicaset while not checking replica_primary does not work" # # moving the login up here and passing in the connection # start = time.time() err, con = mongo_connect(host, port, ssl, user, passwd, replicaset, authdb, ssl_ca_cert_file, cert_file) if err != 0: return err # Autodetect mongo-version and force pymongo to let us know if it can connect or not. err, mongo_version = check_version(con) if err != 0: return err conn_time = time.time() - start if action == "connections": return check_connections(con, warning, critical, perf_data) elif action == "replication_lag": return check_rep_lag(con, host_to_check, port_to_check, warning, critical, False, perf_data, max_lag, user, passwd) elif action == "replication_lag_percent": return check_rep_lag(con, host_to_check, port_to_check, warning, critical, True, perf_data, max_lag, user, passwd, ssl, ssl_ca_cert_file, cert_file) elif action == "replset_state": return check_replset_state(con, perf_data, warning, critical) elif action == "memory": return check_memory(con, warning, critical, perf_data, options.mapped_memory, host) elif action == "memory_mapped": return check_memory_mapped(con, warning, critical, perf_data) elif action == "queues": return check_queues(con, warning, critical, perf_data) elif action == "lock": return check_lock(con, warning, critical, perf_data, mongo_version) elif action == "current_lock": return check_current_lock(con, host, port, warning, critical, perf_data) elif action == "flushing": return check_flushing(con, warning, critical, True, perf_data) elif action == "last_flush_time": return check_flushing(con, warning, critical, False, perf_data) elif action == "index_miss_ratio": index_miss_ratio(con, warning, critical, perf_data) elif action == "databases": return check_databases(con, warning, critical, perf_data) elif action == "collections": return check_collections(con, warning, critical, perf_data) elif action == "oplog": return check_oplog(con, warning, critical, perf_data) elif action == "journal_commits_in_wl": return check_journal_commits_in_wl(con, warning, critical, perf_data) elif action == "database_size": if options.all_databases: return check_all_databases_size(con, warning, critical, perf_data) else: return check_database_size(con, database, warning, critical, perf_data) elif action == "database_indexes": return check_database_indexes(con, database, warning, critical, perf_data) elif action == "collection_documents": return check_collection_documents(con, database, collection, warning, critical, perf_data) elif action == "collection_indexes": return check_collection_indexes(con, database, collection, warning, critical, perf_data) elif action == "collection_size": return check_collection_size(con, database, collection, warning, critical, perf_data) elif action == "collection_storageSize": return check_collection_storageSize(con, database, collection, warning, critical, perf_data) elif action == "journaled": return check_journaled(con, warning, critical, perf_data) elif action == "write_data_files": return check_write_to_datafiles(con, warning, critical, perf_data) elif action == "opcounters": return check_opcounters(con, host, port, warning, critical, perf_data) elif action == "asserts": return check_asserts(con, host, port, warning, critical, perf_data) elif action == "replica_primary": return check_replica_primary(con, host, warning, critical, perf_data, replicaset, mongo_version) elif action == "queries_per_second": return check_queries_per_second(con, query_type, warning, critical, perf_data, mongo_version) elif action == "page_faults": check_page_faults(con, sample_time, warning, critical, perf_data) elif action == "chunks_balance": chunks_balance(con, database, collection, warning, critical) elif action == "connect_primary": return check_connect_primary(con, warning, critical, perf_data) elif action == "collection_state": return check_collection_state(con, database, collection) elif action == "row_count": return check_row_count(con, database, collection, warning, critical, perf_data) elif action == "replset_quorum": return check_replset_quorum(con, perf_data) else: return check_connect(host, port, warning, critical, perf_data, user, passwd, conn_time) def mongo_connect(host=None, port=None, ssl=False, user=None, passwd=None, replica=None, authdb="admin", ssl_ca_cert_file=None, ssl_cert=None, auth_mechanism=None): from pymongo.errors import ConnectionFailure from pymongo.errors import PyMongoError con_args = dict() if ssl: con_args['ssl'] = ssl if ssl_ca_cert_file: con_args['ssl_ca_certs'] = ssl_ca_cert_file if ssl_cert: con_args['ssl_certfile'] = ssl_cert try: # ssl connection for pymongo > 2.3 if pymongo.version >= "2.3": if replica is None: con = pymongo.MongoClient(host, port, **con_args) else: con = pymongo.MongoClient(host, port, read_preference=pymongo.ReadPreference.SECONDARY, replicaSet=replica, **con_args) else: if replica is None: con = pymongo.Connection(host, port, slave_okay=True, network_timeout=10) else: con = pymongo.Connection(host, port, slave_okay=True, network_timeout=10) # we must authenticate the connection, otherwise we won't be able to perform certain operations if ssl_cert and ssl_ca_cert_file and user and auth_mechanism == 'SCRAM-SHA-256': con.the_database.authenticate(user, mechanism='SCRAM-SHA-256') elif ssl_cert and ssl_ca_cert_file and user and auth_mechanism == 'SCRAM-SHA-1': con.the_database.authenticate(user, mechanism='SCRAM-SHA-1') elif ssl_cert and ssl_ca_cert_file and user and auth_mechanism == 'MONGODB-X509': con.the_database.authenticate(user, mechanism='MONGODB-X509') try: result = con.admin.command("ismaster") except ConnectionFailure: print("CRITICAL - Connection to Mongo server on %s:%s has failed" % (host, port) ) sys.exit(2) if 'arbiterOnly' in result and result['arbiterOnly'] == True: print("OK - State: 7 (Arbiter on port %s)" % (port)) sys.exit(0) if user and passwd: db = con[authdb] try: db.authenticate(user, password=passwd) except PyMongoError: sys.exit("Username/Password incorrect") # Ping to check that the server is responding. con.admin.command("ping") except Exception as e: if isinstance(e, pymongo.errors.AutoReconnect) and str(e).find(" is an arbiter") != -1: # We got a pymongo AutoReconnect exception that tells us we connected to an Arbiter Server # This means: Arbiter is reachable and can answer requests/votes - this is all we need to know from an arbiter print("OK - State: 7 (Arbiter)") sys.exit(0) return exit_with_general_critical(e), None return 0, con def exit_with_general_warning(e): if isinstance(e, SystemExit): return e else: print("WARNING - General MongoDB warning:", e) return 1 def exit_with_general_critical(e): if isinstance(e, SystemExit): return e else: print("CRITICAL - General MongoDB Error:", e) return 2 def set_read_preference(db): if pymongo.version >= "2.2": pymongo.read_preferences.Secondary else: db.read_preference = pymongo.ReadPreference.SECONDARY def check_version(con): try: server_info = con.server_info() except Exception as e: return exit_with_general_critical(e), None return 0, int(server_info['version'].split('.')[0].strip()) def check_connect(host, port, warning, critical, perf_data, user, passwd, conn_time): warning = warning or 3 critical = critical or 6 message = "Connection took %.3f seconds" % conn_time message += performance_data(perf_data, [(conn_time, "connection_time", warning, critical)]) return check_levels(conn_time, warning, critical, message) def check_connections(con, warning, critical, perf_data): warning = warning or 80 critical = critical or 95 try: data = get_server_status(con) current = float(data['connections']['current']) available = float(data['connections']['available']) used_percent = int(float(current / (available + current)) * 100) message = "%i percent (%i of %i connections) used" % (used_percent, current, current + available) message += performance_data(perf_data, [(used_percent, "used_percent", warning, critical), (current, "current_connections"), (available, "available_connections")]) return check_levels(used_percent, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def check_rep_lag(con, host, port, warning, critical, percent, perf_data, max_lag, user, passwd, ssl=None, ssl_ca_cert_file=None, cert_file=None): # Get mongo to tell us replica set member name when connecting locally if "127.0.0.1" == host: if not "me" in list(con.admin.command("ismaster","1").keys()): print("UNKNOWN - This is not replicated MongoDB") return 3 host = con.admin.command("ismaster","1")["me"].split(':')[0] if percent: warning = warning or 50 critical = critical or 75 else: warning = warning or 600 critical = critical or 3600 rs_status = {} slaveDelays = {} try: #set_read_preference(con.admin) # Get replica set status try: rs_status = con.admin.command("replSetGetStatus") except pymongo.errors.OperationFailure as e: if ((e.code == None and str(e).find('failed: not running with --replSet"')) or (e.code == 76 and str(e).find('not running with --replSet"'))): print("UNKNOWN - Not running with replSet") return 3 serverVersion = tuple(con.server_info()['version'].split('.')) if serverVersion >= tuple("2.0.0".split(".")): # # check for version greater then 2.0 # rs_conf = con.local.system.replset.find_one() for member in rs_conf['members']: if member.get('slaveDelay') is not None: slaveDelays[member['host']] = member.get('slaveDelay') else: slaveDelays[member['host']] = 0 # Find the primary and/or the current node primary_node = None host_node = None for member in rs_status["members"]: if member["stateStr"] == "PRIMARY": primary_node = member if member.get('name') == "{0}:{1}".format(host, port): host_node = member # Check if we're in the middle of an election and don't have a primary if primary_node is None: print("WARNING - No primary defined. In an election?") return 1 # Check if we failed to find the current host # below should never happen if host_node is None: print("CRITICAL - Unable to find host '" + host + "' in replica set.") return 2 # Is the specified host the primary? if host_node["stateStr"] == "PRIMARY": if max_lag == False: print("OK - This is the primary.") return 0 else: #get the maximal replication lag data = "" maximal_lag = 0 for member in rs_status['members']: if not member['stateStr'] == "ARBITER": lastSlaveOpTime = member['optimeDate'] replicationLag = abs(primary_node["optimeDate"] - lastSlaveOpTime).seconds - slaveDelays[member['name']] data = data + member['name'] + " lag=%d;" % replicationLag maximal_lag = max(maximal_lag, replicationLag) if percent: err, con = mongo_connect(primary_node['name'].split(':')[0], int(primary_node['name'].split(':')[1]), False, user, passwd) if err != 0: return err primary_timediff = replication_get_time_diff(con) maximal_lag = int(float(maximal_lag) / float(primary_timediff) * 100) message = "Maximal lag is " + str(maximal_lag) + " percents" message += performance_data(perf_data, [(maximal_lag, "replication_lag_percent", warning, critical)]) else: message = "Maximal lag is " + str(maximal_lag) + " seconds" message += performance_data(perf_data, [(maximal_lag, "replication_lag", warning, critical)]) return check_levels(maximal_lag, warning, critical, message) elif host_node["stateStr"] == "ARBITER": print("UNKNOWN - This is an arbiter") return 3 # Find the difference in optime between current node and PRIMARY optime_lag = abs(primary_node["optimeDate"] - host_node["optimeDate"]) if host_node['name'] in slaveDelays: slave_delay = slaveDelays[host_node['name']] elif host_node['name'].endswith(':27017') and host_node['name'][:-len(":27017")] in slaveDelays: slave_delay = slaveDelays[host_node['name'][:-len(":27017")]] else: raise Exception("Unable to determine slave delay for {0}".format(host_node['name'])) try: # work starting from python2.7 lag = optime_lag.total_seconds() except: lag = float(optime_lag.seconds + optime_lag.days * 24 * 3600) if percent: err, con = mongo_connect(primary_node['name'].split(':')[0], int(primary_node['name'].split(':')[1]), ssl, user, passwd, None, None, ssl_ca_cert_file, cert_file) if err != 0: return err primary_timediff = replication_get_time_diff(con) if primary_timediff != 0: lag = int(float(lag) / float(primary_timediff) * 100) else: lag = 0 message = "Lag is " + str(lag) + " percents" message += performance_data(perf_data, [(lag, "replication_lag_percent", warning, critical)]) else: message = "Lag is " + str(lag) + " seconds" message += performance_data(perf_data, [(lag, "replication_lag", warning, critical)]) return check_levels(lag, warning + slaveDelays[host_node['name']], critical + slaveDelays[host_node['name']], message) else: # # less than 2.0 check # # Get replica set status rs_status = con.admin.command("replSetGetStatus") # Find the primary and/or the current node primary_node = None host_node = None for member in rs_status["members"]: if member["stateStr"] == "PRIMARY": primary_node = (member["name"], member["optimeDate"]) if member["name"].split(":")[0].startswith(host): host_node = member # Check if we're in the middle of an election and don't have a primary if primary_node is None: print("WARNING - No primary defined. In an election?") sys.exit(1) # Is the specified host the primary? if host_node["stateStr"] == "PRIMARY": print("OK - This is the primary.") sys.exit(0) # Find the difference in optime between current node and PRIMARY optime_lag = abs(primary_node[1] - host_node["optimeDate"]) lag = optime_lag.seconds if percent: err, con = mongo_connect(primary_node['name'].split(':')[0], int(primary_node['name'].split(':')[1])) if err != 0: return err primary_timediff = replication_get_time_diff(con) lag = int(float(lag) / float(primary_timediff) * 100) message = "Lag is " + str(lag) + " percents" message += performance_data(perf_data, [(lag, "replication_lag_percent", warning, critical)]) else: message = "Lag is " + str(lag) + " seconds" message += performance_data(perf_data, [(lag, "replication_lag", warning, critical)]) return check_levels(lag, warning, critical, message) except Exception as e: return exit_with_general_critical(e) # # Check the memory usage of mongo. Alerting on this may be hard to get right # because it'll try to get as much memory as it can. And that's probably # a good thing. # def check_memory(con, warning, critical, perf_data, mapped_memory, host): # Get the total system memory of this system (This is totally bogus if you # are running this command remotely) and calculate based on that how much # memory used by Mongodb is ok or not. meminfo = open('/proc/meminfo').read() matched = re.search(r'^MemTotal:\s+(\d+)', meminfo) if matched: mem_total_kB = int(matched.groups()[0]) if host != "127.0.0.1" and not warning: # Running remotely and value was not set by user, use hardcoded value warning = 12 else: # running locally or user provided value warning = warning or (mem_total_kB * 0.8) / 1024.0 / 1024.0 if host != "127.0.0.1" and not critical: critical = 16 else: critical = critical or (mem_total_kB * 0.9) / 1024.0 / 1024.0 # debugging #print "mem total: {0}kb, warn: {1}GB, crit: {2}GB".format(mem_total_kB,warning, critical) try: data = get_server_status(con) if not data['mem']['supported'] and not mapped_memory: print("OK - Platform not supported for memory info") return 0 # # convert to gigs # message = "Memory Usage:" try: mem_resident = float(data['mem']['resident']) / 1024.0 message += " %.2fGB resident," % (mem_resident) except: mem_resident = 0 message += " resident unsupported," try: mem_virtual = float(data['mem']['virtual']) / 1024.0 message += " %.2fGB virtual," % mem_virtual except: mem_virtual = 0 message += " virtual unsupported," try: mem_mapped = float(data['mem']['mapped']) / 1024.0 message += " %.2fGB mapped," % mem_mapped except: mem_mapped = 0 message += " mapped unsupported," try: mem_mapped_journal = float(data['mem']['mappedWithJournal']) / 1024.0 message += " %.2fGB mappedWithJournal" % mem_mapped_journal except: mem_mapped_journal = 0 message += performance_data(perf_data, [("%.2f" % mem_resident, "memory_usage", warning, critical), ("%.2f" % mem_mapped, "memory_mapped"), ("%.2f" % mem_virtual, "memory_virtual"), ("%.2f" % mem_mapped_journal, "mappedWithJournal")]) #added for unsupported systems like Solaris if mapped_memory and mem_resident == 0: return check_levels(mem_mapped, warning, critical, message) else: return check_levels(mem_resident, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def check_memory_mapped(con, warning, critical, perf_data): # # These thresholds are basically meaningless, and must be customized to your application # warning = warning or 8 critical = critical or 16 try: data = get_server_status(con) if not data['mem']['supported']: print("OK - Platform not supported for memory info") return 0 # # convert to gigs # message = "Memory Usage:" try: mem_mapped = float(data['mem']['mapped']) / 1024.0 message += " %.2fGB mapped," % mem_mapped except: mem_mapped = -1 message += " mapped unsupported," try: mem_mapped_journal = float(data['mem']['mappedWithJournal']) / 1024.0 message += " %.2fGB mappedWithJournal" % mem_mapped_journal except: mem_mapped_journal = 0 message += performance_data(perf_data, [("%.2f" % mem_mapped, "memory_mapped", warning, critical), ("%.2f" % mem_mapped_journal, "mappedWithJournal")]) if not mem_mapped == -1: return check_levels(mem_mapped, warning, critical, message) else: print("OK - Server does not provide mem.mapped info") return 0 except Exception as e: return exit_with_general_critical(e) # # Return the percentage of the time there was a global Lock # def check_lock(con, warning, critical, perf_data, mongo_version): warning = warning or 10 critical = critical or 30 if mongo_version == 2: try: data = get_server_status(con) lockTime = data['globalLock']['lockTime'] totalTime = data['globalLock']['totalTime'] # # calculate percentage # if lockTime > totalTime: lock_percentage = 0.00 else: lock_percentage = float(lockTime) / float(totalTime) * 100 message = "Lock Percentage: %.2f%%" % lock_percentage message += performance_data(perf_data, [("%.2f" % lock_percentage, "lock_percentage", warning, critical)]) return check_levels(lock_percentage, warning, critical, message) except Exception as e: print("Couldn't get globalLock lockTime info from mongo, are you sure you're not using version 3? See the -M option.") return exit_with_general_critical(e) else: print("OK - MongoDB version 3 doesn't report on global locks") return 0 def check_flushing(con, warning, critical, avg, perf_data): # # These thresholds mean it's taking 5 seconds to perform a background flush to issue a warning # and 10 seconds to issue a critical. # warning = warning or 5000 critical = critical or 15000 try: data = get_server_status(con) try: data['backgroundFlushing'] if avg: flush_time = float(data['backgroundFlushing']['average_ms']) stat_type = "Average" else: flush_time = float(data['backgroundFlushing']['last_ms']) stat_type = "Last" message = "%s Flush Time: %.2fms" % (stat_type, flush_time) message += performance_data(perf_data, [("%.2fms" % flush_time, "%s_flush_time" % stat_type.lower(), warning, critical)]) return check_levels(flush_time, warning, critical, message) except Exception: print("OK - flushing stats not available for this storage engine") return 0 except Exception as e: return exit_with_general_critical(e) def index_miss_ratio(con, warning, critical, perf_data): warning = warning or 10 critical = critical or 30 try: data = get_server_status(con) try: data['indexCounters'] serverVersion = tuple(con.server_info()['version'].split('.')) if serverVersion >= tuple("2.4.0".split(".")): miss_ratio = float(data['indexCounters']['missRatio']) else: miss_ratio = float(data['indexCounters']['btree']['missRatio']) except KeyError: not_supported_msg = "not supported on this platform" try: data['indexCounters'] if 'note' in data['indexCounters']: print("OK - MongoDB says: " + not_supported_msg) return 0 else: print("WARNING - Can't get counter from MongoDB") return 1 except Exception: print("OK - MongoDB says: " + not_supported_msg) return 0 message = "Miss Ratio: %.2f" % miss_ratio message += performance_data(perf_data, [("%.2f" % miss_ratio, "index_miss_ratio", warning, critical)]) return check_levels(miss_ratio, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def check_replset_quorum(con, perf_data): db = con['admin'] warning = 1 critical = 2 primary = 0 try: rs_members = db.command("replSetGetStatus")['members'] for member in rs_members: if member['state'] == 1: primary += 1 if primary == 1: state = 0 message = "Cluster is quorate" else: state = 2 message = "Cluster is not quorate and cannot operate" return check_levels(state, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def check_replset_state(con, perf_data, warning="", critical=""): try: warning = [int(x) for x in warning.split(",")] except: warning = [0, 3, 5] try: critical = [int(x) for x in critical.split(",")] except: critical = [8, 4, -1] ok = list(range(-1, 8)) # should include the range of all posiible values try: worst_state = -2 message = "" try: try: set_read_preference(con.admin) data = con.admin.command(pymongo.son_manipulator.SON([('replSetGetStatus', 1)])) except: data = con.admin.command(son.SON([('replSetGetStatus', 1)])) members = data['members'] my_state = int(data['myState']) worst_state = my_state for member in members: their_state = int(member['state']) message += " %s: %i (%s)" % (member['name'], their_state, state_text(their_state)) if state_is_worse(their_state, worst_state, warning, critical): worst_state = their_state message += performance_data(perf_data, [(my_state, "state")]) except pymongo.errors.OperationFailure as e: if ((e.code == None and str(e).find('failed: not running with --replSet"')) or (e.code == 76 and str(e).find('not running with --replSet"'))): worst_state = -1 return check_levels(worst_state, warning, critical, message, ok) except Exception as e: return exit_with_general_critical(e) def state_is_worse(state, worst_state, warning, critical): if worst_state in critical: return False if worst_state in warning: return state in critical return (state in warning) or (state in critical) def state_text(state): if state == 8: return "Down" elif state == 4: return "Fatal error" elif state == 0: return "Starting up, phase1" elif state == 3: return "Recovering" elif state == 5: return "Starting up, phase2" elif state == 1: return "Primary" elif state == 2: return "Secondary" elif state == 7: return "Arbiter" elif state == -1: return "Not running with replSet" else: return "Unknown state" def check_databases(con, warning, critical, perf_data=None): try: try: set_read_preference(con.admin) data = con.admin.command(pymongo.son_manipulator.SON([('listDatabases', 1)])) except: data = con.admin.command(son.SON([('listDatabases', 1)])) count = len(data['databases']) message = "Number of DBs: %.0f" % count message += performance_data(perf_data, [(count, "databases", warning, critical, message)]) return check_levels(count, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def check_collections(con, warning, critical, perf_data=None): try: try: set_read_preference(con.admin) data = con.admin.command(pymongo.son_manipulator.SON([('listDatabases', 1)])) except: data = con.admin.command(son.SON([('listDatabases', 1)])) count = 0 for db in data['databases']: dbase = con[db['name']] set_read_preference(dbase) count += len(dbase.collection_names()) message = "Number of collections: %.0f" % count message += performance_data(perf_data, [(count, "collections", warning, critical, message)]) return check_levels(count, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def check_all_databases_size(con, warning, critical, perf_data): warning = warning or 100 critical = critical or 1000 try: set_read_preference(con.admin) all_dbs_data = con.admin.command(pymongo.son_manipulator.SON([('listDatabases', 1)])) except: all_dbs_data = con.admin.command(son.SON([('listDatabases', 1)])) total_storage_size = 0 message = "" perf_data_param = [()] for db in all_dbs_data['databases']: database = db['name'] data = con[database].command('dbstats') storage_size = round(data['storageSize'] / 1024 / 1024, 1) message += "; Database %s size: %.0f MB" % (database, storage_size) perf_data_param.append((storage_size, database + "_database_size")) total_storage_size += storage_size perf_data_param[0] = (total_storage_size, "total_size", warning, critical) message += performance_data(perf_data, perf_data_param) message = "Total size: %.0f MB" % total_storage_size + message return check_levels(total_storage_size, warning, critical, message) def check_database_size(con, database, warning, critical, perf_data): warning = warning or 100 critical = critical or 1000 perfdata = "" try: set_read_preference(con.admin) data = con[database].command('dbstats') storage_size = data['storageSize'] // 1024 // 1024 if perf_data: perfdata += " | database_size=%i;%i;%i" % (storage_size, warning, critical) #perfdata += " database=%s" %(database) if storage_size >= critical: print("CRITICAL - Database size: %.0f MB, Database: %s%s" % (storage_size, database, perfdata)) return 2 elif storage_size >= warning: print("WARNING - Database size: %.0f MB, Database: %s%s" % (storage_size, database, perfdata)) return 1 else: print("OK - Database size: %.0f MB, Database: %s%s" % (storage_size, database, perfdata)) return 0 except Exception as e: return exit_with_general_critical(e) def check_database_indexes(con, database, warning, critical, perf_data): # # These thresholds are basically meaningless, and must be customized to your application # warning = warning or 100 critical = critical or 1000 perfdata = "" try: set_read_preference(con.admin) data = con[database].command('dbstats') index_size = data['indexSize'] / 1024 // 1024 if perf_data: perfdata += " | database_indexes=%i;%i;%i" % (index_size, warning, critical) if index_size >= critical: print("CRITICAL - %s indexSize: %.0f MB %s" % (database, index_size, perfdata)) return 2 elif index_size >= warning: print("WARNING - %s indexSize: %.0f MB %s" % (database, index_size, perfdata)) return 1 else: print("OK - %s indexSize: %.0f MB %s" % (database, index_size, perfdata)) return 0 except Exception as e: return exit_with_general_critical(e) def check_collection_documents(con, database, collection, warning, critical, perf_data): perfdata = "" try: set_read_preference(con.admin) data = con[database].command('collstats', collection) documents = data['count'] if perf_data: perfdata += " | collection_documents=%i;%i;%i" % (documents, warning, critical) if documents >= critical: print("CRITICAL - %s.%s documents: %s %s" % (database, collection, documents, perfdata)) return 2 elif documents >= warning: print("WARNING - %s.%s documents: %s %s" % (database, collection, documents, perfdata)) return 1 else: print("OK - %s.%s documents: %s %s" % (database, collection, documents, perfdata)) return 0 except Exception as e: return exit_with_general_critical(e) def check_collection_indexes(con, database, collection, warning, critical, perf_data): # # These thresholds are basically meaningless, and must be customized to your application # warning = warning or 100 critical = critical or 1000 perfdata = "" try: set_read_preference(con.admin) data = con[database].command('collstats', collection) total_index_size = data['totalIndexSize'] / 1024 / 1024 if perf_data: perfdata += " | collection_indexes=%i;%i;%i" % (total_index_size, warning, critical) if total_index_size >= critical: print("CRITICAL - %s.%s totalIndexSize: %.0f MB %s" % (database, collection, total_index_size, perfdata)) return 2 elif total_index_size >= warning: print("WARNING - %s.%s totalIndexSize: %.0f MB %s" % (database, collection, total_index_size, perfdata)) return 1 else: print("OK - %s.%s totalIndexSize: %.0f MB %s" % (database, collection, total_index_size, perfdata)) return 0 except Exception as e: return exit_with_general_critical(e) def check_queues(con, warning, critical, perf_data): warning = warning or 10 critical = critical or 30 try: data = get_server_status(con) total_queues = float(data['globalLock']['currentQueue']['total']) readers_queues = float(data['globalLock']['currentQueue']['readers']) writers_queues = float(data['globalLock']['currentQueue']['writers']) message = "Current queue is : total = %d, readers = %d, writers = %d" % (total_queues, readers_queues, writers_queues) message += performance_data(perf_data, [(total_queues, "total_queues", warning, critical), (readers_queues, "readers_queues"), (writers_queues, "writers_queues")]) return check_levels(total_queues, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def check_collection_size(con, database, collection, warning, critical, perf_data): warning = warning or 100 critical = critical or 1000 perfdata = "" try: set_read_preference(con.admin) data = con[database].command('collstats', collection) size = data['size'] / 1024 / 1024 if perf_data: perfdata += " | collection_size=%i;%i;%i" % (size, warning, critical) if size >= critical: print("CRITICAL - %s.%s size: %.0f MB %s" % (database, collection, size, perfdata)) return 2 elif size >= warning: print("WARNING - %s.%s size: %.0f MB %s" % (database, collection, size, perfdata)) return 1 else: print("OK - %s.%s size: %.0f MB %s" % (database, collection, size, perfdata)) return 0 except Exception as e: return exit_with_general_critical(e) def check_collection_storageSize(con, database, collection, warning, critical, perf_data): warning = warning or 100 critical = critical or 1000 perfdata = "" try: set_read_preference(con.admin) data = con[database].command('collstats', collection) storageSize = data['storageSize'] / 1024 / 1024 if perf_data: perfdata += " | collection_storageSize=%i;%i;%i" % (storageSize, warning, critical) if storageSize >= critical: print("CRITICAL - %s.%s storageSize: %.0f MB %s" % (database, collection, storageSize, perfdata)) return 2 elif storageSize >= warning: print("WARNING - %s.%s storageSize: %.0f MB %s" % (database, collection, storageSize, perfdata)) return 1 else: print("OK - %s.%s storageSize: %.0f MB %s" % (database, collection, storageSize, perfdata)) return 0 except Exception as e: return exit_with_general_critical(e) def check_queries_per_second(con, query_type, warning, critical, perf_data, mongo_version): warning = warning or 250 critical = critical or 500 if query_type not in ['insert', 'query', 'update', 'delete', 'getmore', 'command']: return exit_with_general_critical("The query type of '%s' is not valid" % query_type) try: db = con.local data = get_server_status(con) # grab the count num = int(data['opcounters'][query_type]) # do the math last_count = db.nagios_check.find_one({'check': 'query_counts'}) try: ts = int(time.time()) diff_query = num - last_count['data'][query_type]['count'] diff_ts = ts - last_count['data'][query_type]['ts'] if diff_ts == 0: message = "diff_query = " + str(diff_query) + " diff_ts = " + str(diff_ts) return check_levels(0, warning, critical, message) query_per_sec = float(diff_query) / float(diff_ts) # update the count now if mongo_version == 2: db.nagios_check.update({u'_id': last_count['_id']}, {'$set': {"data.%s" % query_type: {'count': num, 'ts': int(time.time())}}}) else: db.nagios_check.update_one({u'_id': last_count['_id']}, {'$set': {"data.%s" % query_type: {'count': num, 'ts': int(time.time())}}}) message = "Queries / Sec: %f" % query_per_sec message += performance_data(perf_data, [(query_per_sec, "%s_per_sec" % query_type, warning, critical, message)]) except KeyError: # # since it is the first run insert it query_per_sec = 0 message = "First run of check.. no data" if mongo_version == 2: db.nagios_check.update({u'_id': last_count['_id']}, {'$set': {"data.%s" % query_type: {'count': num, 'ts': int(time.time())}}}) else: db.nagios_check.update_one({u'_id': last_count['_id']}, {'$set': {"data.%s" % query_type: {'count': num, 'ts': int(time.time())}}}) except TypeError: # # since it is the first run insert it query_per_sec = 0 message = "First run of check.. no data" if mongo_version == 2: db.nagios_check.insert({'check': 'query_counts', 'data': {query_type: {'count': num, 'ts': int(time.time())}}}) else: db.nagios_check.insert_one({'check': 'query_counts', 'data': {query_type: {'count': num, 'ts': int(time.time())}}}) return check_levels(query_per_sec, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def check_oplog(con, warning, critical, perf_data): """ Checking the oplog time - the time of the log currntly saved in the oplog collection defaults: critical 4 hours warning 24 hours those can be changed as usual with -C and -W parameters""" warning = warning or 24 critical = critical or 4 try: db = con.local ol = db.system.namespaces.find_one({"name": "local.oplog.rs"}) if (db.system.namespaces.find_one({"name": "local.oplog.rs"}) != None): oplog = "oplog.rs" else: ol = db.system.namespaces.find_one({"name": "local.oplog.$main"}) if (db.system.namespaces.find_one({"name": "local.oplog.$main"}) != None): oplog = "oplog.$main" else: message = "neither master/slave nor replica set replication detected" return check_levels(None, warning, critical, message) try: set_read_preference(con.admin) data = con.local.command(pymongo.son_manipulator.SON([('collstats', oplog)])) except: data = con.admin.command(son.SON([('collstats', oplog)])) ol_size = data['size'] ol_storage_size = data['storageSize'] ol_used_storage = int(float(ol_size) / ol_storage_size * 100 + 1) ol = con.local[oplog] firstc = ol.find().sort("$natural", pymongo.ASCENDING).limit(1)[0]['ts'] lastc = ol.find().sort("$natural", pymongo.DESCENDING).limit(1)[0]['ts'] time_in_oplog = (lastc.as_datetime() - firstc.as_datetime()) message = "Oplog saves " + str(time_in_oplog) + " %d%% used" % ol_used_storage try: # work starting from python2.7 hours_in_oplog = time_in_oplog.total_seconds() / 60 / 60 except: hours_in_oplog = float(time_in_oplog.seconds + time_in_oplog.days * 24 * 3600) / 60 / 60 approx_level = hours_in_oplog * 100 / ol_used_storage message += performance_data(perf_data, [("%.2f" % hours_in_oplog, 'oplog_time', warning, critical), ("%.2f " % approx_level, 'oplog_time_100_percent_used')]) return check_levels(-approx_level, -warning, -critical, message) except Exception as e: return exit_with_general_critical(e) def check_journal_commits_in_wl(con, warning, critical, perf_data): """ Checking the number of commits which occurred in the db's write lock. Most commits are performed outside of this lock; committed while in the write lock is undesirable. Under very high write situations it is normal for this value to be nonzero. """ warning = warning or 10 critical = critical or 40 try: data = get_server_status(con) j_commits_in_wl = data['dur']['commitsInWriteLock'] message = "Journal commits in DB write lock : %d" % j_commits_in_wl message += performance_data(perf_data, [(j_commits_in_wl, "j_commits_in_wl", warning, critical)]) return check_levels(j_commits_in_wl, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def check_journaled(con, warning, critical, perf_data): """ Checking the average amount of data in megabytes written to the recovery log in the last four seconds""" warning = warning or 20 critical = critical or 40 try: data = get_server_status(con) journaled = data['dur']['journaledMB'] message = "Journaled : %.2f MB" % journaled message += performance_data(perf_data, [("%.2f" % journaled, "journaled", warning, critical)]) return check_levels(journaled, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def check_write_to_datafiles(con, warning, critical, perf_data): """ Checking the average amount of data in megabytes written to the databases datafiles in the last four seconds. As these writes are already journaled, they can occur lazily, and thus the number indicated here may be lower than the amount physically written to disk.""" warning = warning or 20 critical = critical or 40 try: data = get_server_status(con) writes = data['dur']['writeToDataFilesMB'] message = "Write to data files : %.2f MB" % writes message += performance_data(perf_data, [("%.2f" % writes, "write_to_data_files", warning, critical)]) return check_levels(writes, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def get_opcounters(data, opcounters_name, host, port): try: insert = data[opcounters_name]['insert'] query = data[opcounters_name]['query'] update = data[opcounters_name]['update'] delete = data[opcounters_name]['delete'] getmore = data[opcounters_name]['getmore'] command = data[opcounters_name]['command'] except KeyError as e: return 0, [0] * 100 total_commands = insert + query + update + delete + getmore + command new_vals = [total_commands, insert, query, update, delete, getmore, command] return maintain_delta(new_vals, host, port, opcounters_name) def check_opcounters(con, host, port, warning, critical, perf_data): """ A function to get all opcounters delta per minute. In case of a replication - gets the opcounters+opcountersRepl""" warning = warning or 10000 critical = critical or 15000 data = get_server_status(con) err1, delta_opcounters = get_opcounters(data, 'opcounters', host, port) err2, delta_opcounters_repl = get_opcounters(data, 'opcountersRepl', host, port) if err1 == 0 and err2 == 0: delta = [(x + y) for x, y in zip(delta_opcounters, delta_opcounters_repl)] delta[0] = delta_opcounters[0] # only the time delta shouldn't be summarized per_minute_delta = [int(x / delta[0] * 60) for x in delta[1:]] message = "Test succeeded , old values missing" message = "Opcounters: total=%d,insert=%d,query=%d,update=%d,delete=%d,getmore=%d,command=%d" % tuple(per_minute_delta) message += performance_data(perf_data, ([(per_minute_delta[0], "total", warning, critical), (per_minute_delta[1], "insert"), (per_minute_delta[2], "query"), (per_minute_delta[3], "update"), (per_minute_delta[4], "delete"), (per_minute_delta[5], "getmore"), (per_minute_delta[6], "command")])) return check_levels(per_minute_delta[0], warning, critical, message) else: return exit_with_general_critical("problem reading data from temp file") def check_current_lock(con, host, port, warning, critical, perf_data): """ A function to get current lock percentage and not a global one, as check_lock function does""" warning = warning or 10 critical = critical or 30 data = get_server_status(con) lockTime = float(data['globalLock']['lockTime']) totalTime = float(data['globalLock']['totalTime']) err, delta = maintain_delta([totalTime, lockTime], host, port, "locktime") if err == 0: lock_percentage = delta[2] / delta[1] * 100 # lockTime/totalTime*100 message = "Current Lock Percentage: %.2f%%" % lock_percentage message += performance_data(perf_data, [("%.2f" % lock_percentage, "current_lock_percentage", warning, critical)]) return check_levels(lock_percentage, warning, critical, message) else: return exit_with_general_warning("problem reading data from temp file") def check_page_faults(con, host, port, warning, critical, perf_data): """ A function to get page_faults per second from the system""" warning = warning or 10 critical = critical or 30 data = get_server_status(con) try: page_faults = float(data['extra_info']['page_faults']) except: # page_faults unsupported on the underlaying system return exit_with_general_critical("page_faults unsupported on the underlaying system") err, delta = maintain_delta([page_faults], host, port, "page_faults") if err == 0: page_faults_ps = delta[1] / delta[0] message = "Page faults : %.2f ps" % page_faults_ps message += performance_data(perf_data, [("%.2f" % page_faults_ps, "page_faults_ps", warning, critical)]) return check_levels(page_faults_ps, warning, critical, message) else: return exit_with_general_warning("problem reading data from temp file") def check_asserts(con, host, port, warning, critical, perf_data): """ A function to get asserts from the system""" warning = warning or 1 critical = critical or 10 data = get_server_status(con) asserts = data['asserts'] #{ "regular" : 0, "warning" : 6, "msg" : 0, "user" : 12, "rollovers" : 0 } regular = asserts['regular'] warning_asserts = asserts['warning'] msg = asserts['msg'] user = asserts['user'] rollovers = asserts['rollovers'] err, delta = maintain_delta([regular, warning_asserts, msg, user, rollovers], host, port, "asserts") if err == 0: if delta[5] != 0: #the number of rollovers were increased warning = -1 # no matter the metrics this situation should raise a warning # if this is normal rollover - the warning will not appear again, but if there will be a lot of asserts # the warning will stay for a long period of time # although this is not a usual situation regular_ps = delta[1] / delta[0] warning_ps = delta[2] / delta[0] msg_ps = delta[3] / delta[0] user_ps = delta[4] / delta[0] rollovers_ps = delta[5] / delta[0] total_ps = regular_ps + warning_ps + msg_ps + user_ps message = "Total asserts : %.2f ps" % total_ps message += performance_data(perf_data, [(total_ps, "asserts_ps", warning, critical), (regular_ps, "regular"), (warning_ps, "warning"), (msg_ps, "msg"), (user_ps, "user")]) return check_levels(total_ps, warning, critical, message) else: return exit_with_general_warning("problem reading data from temp file") def get_stored_primary_server_name(db): """ get the stored primary server name from db. """ if "last_primary_server" in db.collection_names(): stored_primary_server = db.last_primary_server.find_one()["server"] else: stored_primary_server = None return stored_primary_server def check_replica_primary(con, host, warning, critical, perf_data, replicaset, mongo_version): """ A function to check if the primary server of a replica set has changed """ if warning is None and critical is None: warning = 1 warning = warning or 2 critical = critical or 2 primary_status = 0 message = "Primary server has not changed" db = con["nagios"] data = get_server_status(con) if replicaset != data['repl'].get('setName'): message = "Replica set requested: %s differs from the one found: %s" % (replicaset, data['repl'].get('setName')) primary_status = 2 return check_levels(primary_status, warning, critical, message) current_primary = data['repl'].get('primary') saved_primary = get_stored_primary_server_name(db) if current_primary is None: current_primary = "None" if saved_primary is None: saved_primary = "None" if current_primary != saved_primary: last_primary_server_record = {"server": current_primary} if mongo_version == 2: db.last_primary_server.update({"_id": "last_primary"}, {"$set": last_primary_server_record}, upsert=True) else: db.last_primary_server.update_one({"_id": "last_primary"}, {"$set": last_primary_server_record}, upsert=True) message = "Primary server has changed from %s to %s" % (saved_primary, current_primary) primary_status = 1 return check_levels(primary_status, warning, critical, message) def check_page_faults(con, sample_time, warning, critical, perf_data): warning = warning or 10 critical = critical or 20 try: try: set_read_preference(con.admin) data1 = con.admin.command(pymongo.son_manipulator.SON([('serverStatus', 1)])) time.sleep(sample_time) data2 = con.admin.command(pymongo.son_manipulator.SON([('serverStatus', 1)])) except: data1 = con.admin.command(son.SON([('serverStatus', 1)])) time.sleep(sample_time) data2 = con.admin.command(son.SON([('serverStatus', 1)])) try: #on linux servers only page_faults = (int(data2['extra_info']['page_faults']) - int(data1['extra_info']['page_faults'])) // sample_time except KeyError: print("WARNING - Can't get extra_info.page_faults counter from MongoDB") sys.exit(1) message = "Page Faults: %i" % (page_faults) message += performance_data(perf_data, [(page_faults, "page_faults", warning, critical)]) check_levels(page_faults, warning, critical, message) except Exception as e: exit_with_general_critical(e) def chunks_balance(con, database, collection, warning, critical): warning = warning or 10 critical = critical or 20 nsfilter = database + "." + collection try: try: set_read_preference(con.admin) col = con.config.chunks nscount = col.find({"ns": nsfilter}).count() shards = col.distinct("shard") except: print("WARNING - Can't get chunks infos from MongoDB") sys.exit(1) if nscount == 0: print("WARNING - Namespace %s is not sharded" % (nsfilter)) sys.exit(1) avgchunksnb = nscount // len(shards) warningnb = avgchunksnb * warning // 100 criticalnb = avgchunksnb * critical // 100 for shard in shards: delta = abs(avgchunksnb - col.find({"ns": nsfilter, "shard": shard}).count()) message = "Namespace: %s, Shard name: %s, Chunk delta: %i" % (nsfilter, shard, delta) if delta >= criticalnb and delta > 0: print("CRITICAL - Chunks not well balanced " + message) sys.exit(2) elif delta >= warningnb and delta > 0: print("WARNING - Chunks not well balanced " + message) sys.exit(1) print("OK - Chunks well balanced across shards") sys.exit(0) except Exception as e: exit_with_general_critical(e) print("OK - Chunks well balanced across shards") sys.exit(0) def check_connect_primary(con, warning, critical, perf_data): warning = warning or 3 critical = critical or 6 try: try: set_read_preference(con.admin) data = con.admin.command(pymongo.son_manipulator.SON([('isMaster', 1)])) except: data = con.admin.command(son.SON([('isMaster', 1)])) if data['ismaster'] == True: print("OK - This server is primary") return 0 phost = data['primary'].split(':')[0] pport = int(data['primary'].split(':')[1]) start = time.time() err, con = mongo_connect(phost, pport) if err != 0: return err pconn_time = time.time() - start pconn_time = round(pconn_time, 0) message = "Connection to primary server " + data['primary'] + " took %i seconds" % pconn_time message += performance_data(perf_data, [(pconn_time, "connection_time", warning, critical)]) return check_levels(pconn_time, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def check_collection_state(con, database, collection): try: con[database][collection].find_one() print("OK - Collection %s.%s is reachable " % (database, collection)) return 0 except Exception as e: return exit_with_general_critical(e) def check_row_count(con, database, collection, warning, critical, perf_data): try: count = con[database][collection].count() message = "Row count: %i" % (count) message += performance_data(perf_data, [(count, "row_count", warning, critical)]) return check_levels(count, warning, critical, message) except Exception as e: return exit_with_general_critical(e) def build_file_name(host, port, action): #done this way so it will work when run independently and from shell module_name = re.match('(.*//*)*(.*)\..*', __file__).group(2) if (port == 27017): return "/tmp/" + module_name + "_data/" + host + "-" + action + ".data" else: return "/tmp/" + module_name + "_data/" + host + "-" + str(port) + "-" + action + ".data" def ensure_dir(f): d = os.path.dirname(f) if not os.path.exists(d): os.makedirs(d) def write_values(file_name, string): f = None try: f = open(file_name, 'w') except IOError as e: #try creating if (e.errno == 2): ensure_dir(file_name) f = open(file_name, 'w') else: raise IOError(e) f.write(string) f.close() return 0 def read_values(file_name): data = None try: f = open(file_name, 'r') data = f.read() f.close() return 0, data except IOError as e: if (e.errno == 2): #no previous data return 1, '' except Exception as e: return 2, None def calc_delta(old, new): delta = [] if (len(old) != len(new)): raise Exception("unequal number of parameters") for i in range(0, len(old)): val = float(new[i]) - float(old[i]) if val < 0: val = new[i] delta.append(val) return 0, delta def maintain_delta(new_vals, host, port, action): file_name = build_file_name(host, port, action) err, data = read_values(file_name) old_vals = data.split(';') new_vals = [str(int(time.time()))] + new_vals delta = None try: err, delta = calc_delta(old_vals, new_vals) except: err = 2 write_res = write_values(file_name, ";" . join(str(x) for x in new_vals)) return err + write_res, delta def replication_get_time_diff(con): col = 'oplog.rs' local = con.local ol = local.system.namespaces.find_one({"name": "local.oplog.$main"}) if ol: col = 'oplog.$main' firstc = local[col].find().sort("$natural", 1).limit(1) lastc = local[col].find().sort("$natural", -1).limit(1) first = next(firstc) last = next(lastc) tfirst = first["ts"] tlast = last["ts"] delta = tlast.time - tfirst.time return delta # # main app # if __name__ == "__main__": sys.exit(main(sys.argv[1:]))