flapjack 0.5.5 → 0.6.23
Sign up to get free protection for your applications and to get access to all the features.
- data/.gitignore +10 -0
- data/.rbenv-version +1 -0
- data/.rspec +10 -0
- data/Gemfile +18 -0
- data/Guardfile +14 -0
- data/README.md +152 -173
- data/Rakefile +53 -150
- data/bin/flapjack +72 -0
- data/bin/flapjack-nagios-receiver +111 -0
- data/bin/flapjack-nagios-receiver-control +15 -0
- data/bin/flapjack-netsaint-parser +0 -2
- data/bin/flapjack-populator +133 -16
- data/bin/install-flapjack-systemwide +2 -2
- data/config.ru +11 -0
- data/dist/etc/init.d/flapjack +46 -0
- data/dist/etc/init.d/flapjack-nagios-receiver +36 -0
- data/doc/GLOSSARY.md +19 -0
- data/etc/flapjack_config.yaml.example +90 -0
- data/features/events.feature +132 -0
- data/features/notifications.feature +57 -0
- data/features/packaging-lintian.feature +5 -3
- data/features/steps/events_steps.rb +164 -0
- data/features/steps/flapjack-importer_steps.rb +2 -5
- data/features/steps/flapjack-worker_steps.rb +13 -6
- data/features/steps/notifications_steps.rb +178 -0
- data/features/steps/packaging-lintian_steps.rb +14 -0
- data/features/steps/time_travel_steps.rb +34 -0
- data/features/support/env.rb +63 -36
- data/flapjack.gemspec +35 -186
- data/lib/flapjack.rb +2 -0
- data/lib/flapjack/api.rb +274 -0
- data/lib/flapjack/api/entity_check_presenter.rb +184 -0
- data/lib/flapjack/api/entity_presenter.rb +66 -0
- data/lib/flapjack/cli/worker_manager.rb +1 -2
- data/lib/flapjack/configuration.rb +11 -0
- data/lib/flapjack/coordinator.rb +288 -0
- data/lib/flapjack/daemonizing.rb +186 -0
- data/lib/flapjack/data/contact.rb +45 -0
- data/lib/flapjack/data/entity.rb +89 -0
- data/lib/flapjack/data/entity_check.rb +396 -0
- data/lib/flapjack/data/event.rb +144 -0
- data/lib/flapjack/data/notification.rb +13 -0
- data/lib/flapjack/executive.rb +289 -0
- data/lib/flapjack/filters/acknowledgement.rb +39 -0
- data/lib/flapjack/filters/{any_parents_failed.rb → base.rb} +6 -4
- data/lib/flapjack/filters/delays.rb +53 -0
- data/lib/flapjack/filters/detect_mass_client_failures.rb +44 -0
- data/lib/flapjack/filters/ok.rb +25 -5
- data/lib/flapjack/filters/scheduled_maintenance.rb +17 -0
- data/lib/flapjack/filters/unscheduled_maintenance.rb +17 -0
- data/lib/flapjack/jabber.rb +294 -0
- data/lib/flapjack/notification/common.rb +23 -0
- data/lib/flapjack/notification/email.rb +107 -0
- data/lib/flapjack/notification/email/alert.html.haml +48 -0
- data/lib/flapjack/notification/email/alert.text.erb +14 -0
- data/lib/flapjack/notification/sms.rb +42 -0
- data/lib/flapjack/notification/sms/messagenet.rb +49 -0
- data/lib/flapjack/notifier_engine.rb +4 -4
- data/lib/flapjack/notifiers/mailer/mailer.rb +6 -7
- data/lib/flapjack/notifiers/xmpp/xmpp.rb +12 -12
- data/lib/flapjack/pagerduty.rb +230 -0
- data/lib/flapjack/patches.rb +108 -19
- data/lib/flapjack/persistence/data_mapper/models/check.rb +5 -3
- data/lib/flapjack/persistence/data_mapper/models/check_template.rb +2 -0
- data/lib/flapjack/persistence/data_mapper/models/event.rb +2 -0
- data/lib/flapjack/persistence/data_mapper/models/node.rb +3 -1
- data/lib/flapjack/persistence/data_mapper/models/related_check.rb +3 -1
- data/lib/flapjack/pikelet.rb +56 -0
- data/lib/flapjack/transports/beanstalkd.rb +1 -1
- data/lib/flapjack/transports/result.rb +6 -6
- data/lib/flapjack/utility.rb +46 -0
- data/lib/flapjack/version.rb +5 -0
- data/lib/flapjack/web.rb +198 -0
- data/lib/flapjack/web/views/acknowledge.haml +55 -0
- data/lib/flapjack/web/views/check.haml +162 -0
- data/lib/flapjack/web/views/index.haml +92 -0
- data/lib/flapjack/web/views/self_stats.haml +56 -0
- data/lib/flapjack/{applications/worker.rb → worker/application.rb} +0 -0
- data/lib/flapjack/worker/cli.rb +49 -0
- data/{spec → spec.old}/check_sandbox/echo +0 -0
- data/{spec → spec.old}/check_sandbox/sandboxed_check +0 -0
- data/{spec → spec.old}/configs/flapjack-notifier-couchdb.ini +0 -0
- data/{spec → spec.old}/configs/flapjack-notifier.ini +0 -0
- data/{spec → spec.old}/configs/recipients.ini +0 -0
- data/{spec → spec.old}/helpers.rb +0 -0
- data/{spec → spec.old}/inifile_spec.rb +0 -0
- data/{spec → spec.old}/mock-notifiers/mock/init.rb +0 -0
- data/{spec → spec.old}/mock-notifiers/mock/mock.rb +0 -0
- data/{spec → spec.old}/notifier-directories/spoons/testmailer/init.rb +0 -0
- data/{spec → spec.old}/notifier_application_spec.rb +0 -0
- data/{spec → spec.old}/notifier_filters_spec.rb +0 -0
- data/{spec → spec.old}/notifier_options_multiplexer_spec.rb +0 -0
- data/{spec → spec.old}/notifier_options_spec.rb +0 -0
- data/{spec → spec.old}/notifier_spec.rb +0 -0
- data/{spec → spec.old}/notifiers/mailer_spec.rb +0 -0
- data/{spec → spec.old}/notifiers/xmpp_spec.rb +0 -0
- data/{spec → spec.old}/persistence/datamapper_spec.rb +0 -0
- data/{spec → spec.old}/persistence/mock_persistence_backend.rb +0 -0
- data/{spec → spec.old}/simple.ini +0 -0
- data/{spec → spec.old}/spec.opts +0 -0
- data/{spec → spec.old}/test-filters/blocker.rb +0 -0
- data/{spec → spec.old}/test-filters/mock.rb +0 -0
- data/{spec → spec.old}/transports/beanstalkd_spec.rb +0 -0
- data/{spec → spec.old}/transports/mock_transport.rb +0 -0
- data/{spec → spec.old}/worker_application_spec.rb +0 -0
- data/{spec → spec.old}/worker_options_spec.rb +0 -0
- data/spec/lib/flapjack/api/entity_check_presenter_spec.rb +117 -0
- data/spec/lib/flapjack/api/entity_presenter_spec.rb +92 -0
- data/spec/lib/flapjack/api_spec.rb +170 -0
- data/spec/lib/flapjack/coordinator_spec.rb +16 -0
- data/spec/lib/flapjack/data/entity_check_spec.rb +398 -0
- data/spec/lib/flapjack/data/entity_spec.rb +71 -0
- data/spec/lib/flapjack/data/event_spec.rb +6 -0
- data/spec/lib/flapjack/executive_spec.rb +59 -0
- data/spec/lib/flapjack/filters/acknowledgement_spec.rb +6 -0
- data/spec/lib/flapjack/filters/delays_spec.rb +6 -0
- data/spec/lib/flapjack/filters/detect_mass_client_failures_spec.rb +6 -0
- data/spec/lib/flapjack/filters/ok_spec.rb +6 -0
- data/spec/lib/flapjack/filters/scheduled_maintenance_spec.rb +6 -0
- data/spec/lib/flapjack/filters/unscheduled_maintenance_spec.rb +6 -0
- data/spec/lib/flapjack/jabber_spec.rb +150 -0
- data/spec/lib/flapjack/notification/email_spec.rb +6 -0
- data/spec/lib/flapjack/notification/sms_spec.rb +6 -0
- data/spec/lib/flapjack/pikelet_spec.rb +28 -0
- data/spec/lib/flapjack/web_spec.rb +188 -0
- data/spec/spec_helper.rb +44 -0
- data/spec/support/profile_all_formatter.rb +44 -0
- data/spec/support/uncolored_doc_formatter.rb +9 -0
- data/tasks/events.rake +85 -0
- data/tmp/acknowledge.rb +14 -0
- data/tmp/create_config_yaml.rb +16 -0
- data/tmp/create_events_failure.rb +33 -0
- data/tmp/create_events_ok.rb +33 -0
- data/tmp/create_events_ok_fail_ack_ok.rb +54 -0
- data/tmp/create_events_ok_failure.rb +40 -0
- data/tmp/create_events_ok_failure_ack.rb +54 -0
- data/tmp/dummy_entities.json +1 -0
- data/tmp/generate_nagios_test_hosts.rb +16 -0
- data/tmp/parse_config_yaml.rb +7 -0
- data/tmp/redis_delete_all_keys.rb +11 -0
- data/tmp/test_entities.json +1 -0
- metadata +482 -221
- data/TODO.md +0 -36
- data/VERSION +0 -1
- data/bin/flapjack-benchmark +0 -50
- data/bin/flapjack-notifier +0 -21
- data/bin/flapjack-notifier-manager +0 -43
- data/bin/flapjack-stats +0 -27
- data/bin/flapjack-worker +0 -13
- data/bin/flapjack-worker-manager +0 -35
- data/dist/etc/init.d/flapjack-notifier +0 -47
- data/dist/etc/init.d/flapjack-workers +0 -44
- data/features/flapjack-notifier-manager.feature +0 -19
- data/features/flapjack-worker-manager.feature +0 -27
- data/features/flapjack-worker.feature +0 -27
- data/features/netsaint-config-converter.feature +0 -126
- data/features/persistence/couch.feature +0 -105
- data/features/persistence/sqlite3.feature +0 -105
- data/features/persistence/steps/couch_steps.rb +0 -25
- data/features/persistence/steps/generic_steps.rb +0 -102
- data/features/persistence/steps/sqlite3_steps.rb +0 -13
- data/features/steps/flapjack-notifier-manager_steps.rb +0 -24
- data/features/steps/flapjack-worker-manager_steps.rb +0 -48
- data/lib/flapjack/applications/notifier.rb +0 -222
- data/lib/flapjack/cli/notifier.rb +0 -108
- data/lib/flapjack/cli/notifier_manager.rb +0 -86
- data/lib/flapjack/cli/worker.rb +0 -51
@@ -0,0 +1,45 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
module Flapjack
|
4
|
+
|
5
|
+
module Data
|
6
|
+
|
7
|
+
class Contact
|
8
|
+
|
9
|
+
# takes a check, looks up contacts that are interested in this check (or in the check's entity)
|
10
|
+
# and returns an array of contact ids
|
11
|
+
def self.find_all_for_entity_check(entity_check, options = {})
|
12
|
+
logger = options[:logger]
|
13
|
+
logger = nil
|
14
|
+
raise "Redis connection not set" unless redis = options[:redis]
|
15
|
+
|
16
|
+
entity = entity_check.entity
|
17
|
+
check = entity_check.check
|
18
|
+
|
19
|
+
if logger
|
20
|
+
logger.debug("contacts for #{entity.id} (#{entity.name}): " + redis.smembers("contacts_for:#{entity.id}").length.to_s)
|
21
|
+
logger.debug("contacts for #{check}: " + redis.smembers("contacts_for:#{check}").length.to_s)
|
22
|
+
end
|
23
|
+
|
24
|
+
union = redis.sunion("contacts_for:#{entity.id}", "contacts_for:#{check}")
|
25
|
+
logger.debug("contacts for union of #{entity.id} and #{check}: " + union.length.to_s) if logger
|
26
|
+
union
|
27
|
+
end
|
28
|
+
|
29
|
+
def self.pagerduty_credentials_for_contact(contact, options = {})
|
30
|
+
logger = options[:logger]
|
31
|
+
raise "Redis connection not set" unless redis = options[:redis]
|
32
|
+
|
33
|
+
service_key = redis.hget("contact_media:#{contact}", 'pagerduty')
|
34
|
+
return nil unless service_key
|
35
|
+
|
36
|
+
deets = redis.hgetall("contact_pagerduty:#{contact}")
|
37
|
+
return deets.merge('service_key' => service_key)
|
38
|
+
|
39
|
+
end
|
40
|
+
|
41
|
+
end
|
42
|
+
|
43
|
+
end
|
44
|
+
|
45
|
+
end
|
@@ -0,0 +1,89 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
module Flapjack
|
4
|
+
|
5
|
+
module Data
|
6
|
+
|
7
|
+
class Entity
|
8
|
+
|
9
|
+
attr_accessor :name, :id
|
10
|
+
|
11
|
+
def self.all(options = {})
|
12
|
+
raise "Redis connection not set" unless redis = options[:redis]
|
13
|
+
redis.keys("entity_id:*").collect {|k|
|
14
|
+
k =~ /^entity_id:(.+)$/; entity_name = $1
|
15
|
+
self.new(:name => entity_name, :id => redis.get("entity_id:#{entity_name}").to_i, :redis => redis)
|
16
|
+
}
|
17
|
+
end
|
18
|
+
|
19
|
+
def self.add(entity, options = {})
|
20
|
+
raise "Redis connection not set" unless redis = options[:redis]
|
21
|
+
raise "Entity name not provided" unless entity['name'] && !entity['name'].empty?
|
22
|
+
|
23
|
+
if entity['id']
|
24
|
+
redis.multi
|
25
|
+
existing_name = redis.hget("entity:#{entity['id']}", 'name')
|
26
|
+
redis.del("entity_id:#{existing_name}") unless existing_name == entity['name']
|
27
|
+
redis.set("entity_id:#{entity['name']}", entity['id'])
|
28
|
+
redis.hset("entity:#{entity['id']}", 'name', entity['name'])
|
29
|
+
|
30
|
+
redis.del("contacts_for:#{entity['id']}")
|
31
|
+
if entity['contacts'] && entity['contacts'].respond_to?(:each)
|
32
|
+
entity['contacts'].each {|contact|
|
33
|
+
redis.sadd("contacts_for:#{entity['id']}", contact)
|
34
|
+
}
|
35
|
+
end
|
36
|
+
redis.exec
|
37
|
+
else
|
38
|
+
# empty string is the redis equivalent of a Ruby nil, i.e. key with
|
39
|
+
# no value
|
40
|
+
redis.set("entity_id:#{entity['name']}", '')
|
41
|
+
end
|
42
|
+
end
|
43
|
+
|
44
|
+
def self.find_by_name(entity_name, options = {})
|
45
|
+
raise "Redis connection not set" unless redis = options[:redis]
|
46
|
+
entity_id = redis.get("entity_id:#{entity_name}")
|
47
|
+
if entity_id.nil?
|
48
|
+
# key doesn't exist
|
49
|
+
return unless options[:create]
|
50
|
+
self.add({'name' => entity_name}, :redis => redis)
|
51
|
+
end
|
52
|
+
self.new(:name => entity_name,
|
53
|
+
:id => (entity_id.nil? || entity_id.empty?) ? nil : entity_id.to_i,
|
54
|
+
:redis => redis)
|
55
|
+
end
|
56
|
+
|
57
|
+
def self.find_by_id(entity_id, options = {})
|
58
|
+
raise "Redis connection not set" unless redis = options[:redis]
|
59
|
+
entity_name = redis.hget("entity:#{entity_id}", 'name')
|
60
|
+
return if entity_name.nil? || entity_name.empty?
|
61
|
+
self.new(:name => entity_name, :id => entity_id, :redis => redis)
|
62
|
+
end
|
63
|
+
|
64
|
+
def check_list
|
65
|
+
@redis.keys("check:#{@name}:*").map {|k| k =~ /^check:#{@name}:(.+)$/; $1}
|
66
|
+
end
|
67
|
+
|
68
|
+
def check_count
|
69
|
+
checks = check_list
|
70
|
+
return if checks.nil?
|
71
|
+
checks.length
|
72
|
+
end
|
73
|
+
|
74
|
+
private
|
75
|
+
|
76
|
+
# NB: initializer should not be used directly -- instead one of the finder methods
|
77
|
+
# above will call it
|
78
|
+
def initialize(options = {})
|
79
|
+
raise "Redis connection not set" unless @redis = options[:redis]
|
80
|
+
raise "Entity name not set" unless @name = options[:name]
|
81
|
+
@id = options[:id]
|
82
|
+
@logger = options[:logger]
|
83
|
+
end
|
84
|
+
|
85
|
+
end
|
86
|
+
|
87
|
+
end
|
88
|
+
|
89
|
+
end
|
@@ -0,0 +1,396 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
require 'yajl/json_gem'
|
4
|
+
|
5
|
+
require 'flapjack/patches'
|
6
|
+
|
7
|
+
require 'flapjack/data/entity'
|
8
|
+
|
9
|
+
module Flapjack
|
10
|
+
|
11
|
+
module Data
|
12
|
+
|
13
|
+
class EntityCheck
|
14
|
+
|
15
|
+
STATE_OK = 'ok'
|
16
|
+
STATE_WARNING = 'warning'
|
17
|
+
STATE_CRITICAL = 'critical'
|
18
|
+
STATE_UNKNOWN = 'unknown'
|
19
|
+
|
20
|
+
attr_accessor :entity, :check
|
21
|
+
|
22
|
+
# TODO probably shouldn't always be creating on query -- work out when this should be happening
|
23
|
+
def self.for_event_id(event_id, options = {})
|
24
|
+
raise "Redis connection not set" unless redis = options[:redis]
|
25
|
+
entity_name, check = event_id.split(':')
|
26
|
+
self.new(Flapjack::Data::Entity.find_by_name(entity_name, :redis => redis, :create => true), check,
|
27
|
+
:redis => redis)
|
28
|
+
end
|
29
|
+
|
30
|
+
# TODO probably shouldn't always be creating on query -- work out when this should be happening
|
31
|
+
def self.for_entity_name(entity_name, check, options = {})
|
32
|
+
raise "Redis connection not set" unless redis = options[:redis]
|
33
|
+
self.new(Flapjack::Data::Entity.find_by_name(entity_name, :redis => redis, :create => true), check,
|
34
|
+
:redis => redis)
|
35
|
+
end
|
36
|
+
|
37
|
+
def self.for_entity_id(entity_id, check, options = {})
|
38
|
+
raise "Redis connection not set" unless redis = options[:redis]
|
39
|
+
self.new(Flapjack::Data::Entity.find_by_id(entity_id, :redis => redis), check,
|
40
|
+
:redis => redis)
|
41
|
+
end
|
42
|
+
|
43
|
+
def self.for_entity(entity, check, options = {})
|
44
|
+
raise "Redis connection not set" unless redis = options[:redis]
|
45
|
+
self.new(entity, check, :redis => redis)
|
46
|
+
end
|
47
|
+
|
48
|
+
def entity_name
|
49
|
+
@entity.name
|
50
|
+
end
|
51
|
+
|
52
|
+
# takes a key "entity:check", returns true if the check is in unscheduled
|
53
|
+
# maintenance
|
54
|
+
def in_unscheduled_maintenance?
|
55
|
+
@redis.exists("#{@key}:unscheduled_maintenance")
|
56
|
+
end
|
57
|
+
|
58
|
+
# returns true if the check is in scheduled maintenance
|
59
|
+
def in_scheduled_maintenance?
|
60
|
+
@redis.exists("#{@key}:scheduled_maintenance")
|
61
|
+
end
|
62
|
+
|
63
|
+
# creates, or modifies, an event object and adds it to the events list in redis
|
64
|
+
# 'type' => 'service',
|
65
|
+
# 'state' => state,
|
66
|
+
# 'summary' => check_output,
|
67
|
+
# 'time' => timestamp
|
68
|
+
def create_event(event)
|
69
|
+
event.merge!('entity' => @entity.name, 'check' => @check)
|
70
|
+
event['time'] = Time.now.to_i if event['time'].nil?
|
71
|
+
@redis.rpush('events', Yajl::Encoder.encode(event))
|
72
|
+
end
|
73
|
+
|
74
|
+
def create_acknowledgement(opts = {})
|
75
|
+
defaults = {
|
76
|
+
'summary' => '...'
|
77
|
+
}
|
78
|
+
options = defaults.merge(opts)
|
79
|
+
|
80
|
+
event = { 'type' => 'action',
|
81
|
+
'state' => 'acknowledgement',
|
82
|
+
'summary' => options['summary'],
|
83
|
+
'duration' => options['duration'],
|
84
|
+
'acknowledgement_id' => options['acknowledgement_id']
|
85
|
+
}
|
86
|
+
create_event(event)
|
87
|
+
end
|
88
|
+
|
89
|
+
# FIXME: need to add summary to summary of existing unscheduled maintenance if there is
|
90
|
+
# one, and extend duration / expiry time, instead of creating a separate unscheduled
|
91
|
+
# outage as we are doing now...
|
92
|
+
def create_unscheduled_maintenance(opts = {})
|
93
|
+
start_time = opts[:start_time] # unix timestamp
|
94
|
+
duration = opts[:duration] # seconds
|
95
|
+
summary = opts[:summary]
|
96
|
+
time_remaining = (start_time + duration) - Time.now.to_i
|
97
|
+
if time_remaining > 0
|
98
|
+
@redis.setex("#{@key}:unscheduled_maintenance", time_remaining, start_time)
|
99
|
+
end
|
100
|
+
@redis.zadd("#{@key}:unscheduled_maintenances", duration, start_time)
|
101
|
+
@redis.set("#{@key}:#{start_time}:unscheduled_maintenance:summary", summary)
|
102
|
+
|
103
|
+
@redis.zadd("#{@key}:sorted_unscheduled_maintenance_timestamps", start_time, start_time)
|
104
|
+
end
|
105
|
+
|
106
|
+
# ends any unscheduled maintenance
|
107
|
+
def end_unscheduled_maintenance(opts = {})
|
108
|
+
defaults = {
|
109
|
+
:end_time => Time.now.to_i
|
110
|
+
}
|
111
|
+
options = defaults.merge(opts)
|
112
|
+
end_time = options[:end_time]
|
113
|
+
|
114
|
+
if (um_start = @redis.get("#{@key}:unscheduled_maintenance"))
|
115
|
+
duration = end_time - um_start.to_i
|
116
|
+
@logger.debug("ending unscheduled downtime for #{@key} at #{Time.at(end_time).to_s}") if @logger
|
117
|
+
@redis.del("#{@key}:unscheduled_maintenance")
|
118
|
+
@redis.zadd("#{@key}:unscheduled_maintenances", duration, um_start)
|
119
|
+
@redis.zadd("#{@key}:sorted_unscheduled_maintenance_timestamps", um_start, um_start)
|
120
|
+
else
|
121
|
+
@logger.debug("end_unscheduled_maintenance called for #{@key} but none found") if @logger
|
122
|
+
end
|
123
|
+
end
|
124
|
+
|
125
|
+
# creates a scheduled maintenance period for a check
|
126
|
+
# TODO: consider adding some validation to the data we're adding in here
|
127
|
+
# eg start_time is a believable unix timestamp (not in the past and not too
|
128
|
+
# far in the future), duration is within some bounds...
|
129
|
+
def create_scheduled_maintenance(opts = {})
|
130
|
+
start_time = opts[:start_time] # unix timestamp
|
131
|
+
duration = opts[:duration] # seconds
|
132
|
+
summary = opts[:summary]
|
133
|
+
@redis.zadd("#{@key}:scheduled_maintenances", duration, start_time)
|
134
|
+
@redis.set("#{@key}:#{start_time}:scheduled_maintenance:summary", summary)
|
135
|
+
|
136
|
+
@redis.zadd("#{@key}:sorted_scheduled_maintenance_timestamps", start_time, start_time)
|
137
|
+
|
138
|
+
# scheduled maintenance periods have changed, revalidate
|
139
|
+
update_scheduled_maintenance(:revalidate => true)
|
140
|
+
end
|
141
|
+
|
142
|
+
# delete a scheduled maintenance
|
143
|
+
def delete_scheduled_maintenance(opts = {})
|
144
|
+
start_time = opts[:start_time]
|
145
|
+
@redis.del("#{@key}:#{start_time}:scheduled_maintenance:summary")
|
146
|
+
@redis.zrem("#{@key}:scheduled_maintenances", start_time)
|
147
|
+
|
148
|
+
@redis.zremrangebyscore("#{@key}:sorted_scheduled_maintenance_timestamps", start_time, start_time)
|
149
|
+
|
150
|
+
# scheduled maintenance periods have changed, revalidate
|
151
|
+
update_scheduled_maintenance(:revalidate => true)
|
152
|
+
end
|
153
|
+
|
154
|
+
# if not in scheduled maintenance, looks in scheduled maintenance list for a check to see if
|
155
|
+
# current state should be set to scheduled maintenance, and sets it as appropriate
|
156
|
+
def update_scheduled_maintenance(opts = {})
|
157
|
+
if opts[:revalidate]
|
158
|
+
@redis.del("#{@key}:scheduled_maintenance")
|
159
|
+
else
|
160
|
+
return if in_scheduled_maintenance?
|
161
|
+
end
|
162
|
+
|
163
|
+
# are we within a scheduled maintenance period?
|
164
|
+
t = Time.now.to_i
|
165
|
+
current_sched_ms = maintenances(nil, nil, :scheduled => true).select {|sm|
|
166
|
+
(sm[:start_time] <= t) && (t < sm[:end_time])
|
167
|
+
}
|
168
|
+
return if current_sched_ms.empty?
|
169
|
+
|
170
|
+
# yes! so set current scheduled maintenance
|
171
|
+
# if multiple scheduled maintenances found, find the end_time furthest in the future
|
172
|
+
most_futuristic = current_sched_ms.max {|sm| sm[:end_time] }
|
173
|
+
start_time = most_futuristic[:start_time]
|
174
|
+
duration = most_futuristic[:duration]
|
175
|
+
@redis.setex("#{@key}:scheduled_maintenance", duration.to_i, start_time)
|
176
|
+
end
|
177
|
+
|
178
|
+
# returns nil if no previous state; this must be considered as a possible
|
179
|
+
# state by classes using this model
|
180
|
+
def state
|
181
|
+
@redis.hget("check:#{@key}", 'state')
|
182
|
+
end
|
183
|
+
|
184
|
+
def update_state(state, options = {})
|
185
|
+
return unless validate_state(state)
|
186
|
+
timestamp = options[:timestamp] || Time.now.to_i
|
187
|
+
client = options[:client]
|
188
|
+
summary = options[:summary]
|
189
|
+
count = options[:count]
|
190
|
+
|
191
|
+
# Note the current state (for speedy lookups)
|
192
|
+
@redis.hset("check:#{@key}", 'state', state)
|
193
|
+
|
194
|
+
# FIXME: rename to last_state_change?
|
195
|
+
@redis.hset("check:#{@key}", 'last_change', timestamp)
|
196
|
+
|
197
|
+
# Retain all state changes for entity:check pair
|
198
|
+
@redis.rpush("#{@key}:states", timestamp)
|
199
|
+
@redis.set("#{@key}:#{timestamp}:state", state)
|
200
|
+
@redis.set("#{@key}:#{timestamp}:summary", summary) if summary
|
201
|
+
@redis.set("#{@key}:#{timestamp}:count", count) if count
|
202
|
+
|
203
|
+
@redis.zadd("#{@key}:sorted_state_timestamps", timestamp, timestamp)
|
204
|
+
|
205
|
+
case state
|
206
|
+
when STATE_WARNING, STATE_CRITICAL
|
207
|
+
@redis.zadd('failed_checks', timestamp, @key)
|
208
|
+
# FIXME: Iterate through a list of tags associated with an entity:check pair, and update counters
|
209
|
+
@redis.zadd("failed_checks:client:#{client}", timestamp, @key) if client
|
210
|
+
else
|
211
|
+
@redis.zrem("failed_checks", @key)
|
212
|
+
# FIXME: Iterate through a list of tags associated with an entity:check pair, and update counters
|
213
|
+
@redis.zrem("failed_checks:client:#{client}", @key) if client
|
214
|
+
end
|
215
|
+
end
|
216
|
+
|
217
|
+
def last_update
|
218
|
+
lu = @redis.hget("check:#{@key}", 'last_update')
|
219
|
+
return unless (lu && lu =~ /^\d+$/)
|
220
|
+
lu.to_i
|
221
|
+
end
|
222
|
+
|
223
|
+
def last_update=(timestamp)
|
224
|
+
@redis.hset("check:#{@key}", 'last_update', timestamp)
|
225
|
+
end
|
226
|
+
|
227
|
+
def last_change
|
228
|
+
lc = @redis.hget("check:#{@key}", 'last_change')
|
229
|
+
return unless (lc && lc =~ /^\d+$/)
|
230
|
+
lc.to_i
|
231
|
+
end
|
232
|
+
|
233
|
+
def last_problem_notification
|
234
|
+
lpn = @redis.get("#{@key}:last_problem_notification")
|
235
|
+
return unless (lpn && lpn =~ /^\d+$/)
|
236
|
+
lpn.to_i
|
237
|
+
end
|
238
|
+
|
239
|
+
def last_recovery_notification
|
240
|
+
lrn = @redis.get("#{@key}:last_recovery_notification")
|
241
|
+
return unless (lrn && lrn =~ /^\d+$/)
|
242
|
+
lrn.to_i
|
243
|
+
end
|
244
|
+
|
245
|
+
def last_acknowledgement_notification
|
246
|
+
lan = @redis.get("#{@key}:last_acknowledgement_notification")
|
247
|
+
return unless (lan && lan =~ /^\d+$/)
|
248
|
+
lan.to_i
|
249
|
+
end
|
250
|
+
|
251
|
+
def event_count_at(timestamp)
|
252
|
+
eca = @redis.get("#{@key}:#{timestamp}:count")
|
253
|
+
return unless (eca && eca =~ /^\d+$/)
|
254
|
+
eca.to_i
|
255
|
+
end
|
256
|
+
|
257
|
+
def failed?
|
258
|
+
[STATE_WARNING, STATE_CRITICAL].include?( state )
|
259
|
+
end
|
260
|
+
|
261
|
+
def ok?
|
262
|
+
[STATE_OK].include?( state )
|
263
|
+
end
|
264
|
+
|
265
|
+
def summary
|
266
|
+
timestamp = @redis.lindex("#{@key}:states", -1)
|
267
|
+
@redis.get("#{@key}:#{timestamp}:summary")
|
268
|
+
end
|
269
|
+
|
270
|
+
# Returns a list of states for this entity check, sorted by timestamp.
|
271
|
+
#
|
272
|
+
# start_time and end_time should be passed as integer timestamps; these timestamps
|
273
|
+
# will be considered inclusively, so, e.g. coverage for a day should go
|
274
|
+
# from midnight to 11:59:59 PM. Pass nil for either end to leave that
|
275
|
+
# side unbounded.
|
276
|
+
def historical_states(start_time, end_time, opts = {})
|
277
|
+
start_time ||= '-inf'
|
278
|
+
end_time ||= '+inf'
|
279
|
+
order = opts[:order]
|
280
|
+
query = (order && 'desc'.eql?(order.downcase)) ? :zrevrangebyscore : :zrangebyscore
|
281
|
+
state_ts = @redis.send(query, "#{@key}:sorted_state_timestamps", start_time, end_time)
|
282
|
+
|
283
|
+
state_data = nil
|
284
|
+
|
285
|
+
@redis.multi do |r|
|
286
|
+
state_data = state_ts.collect {|ts|
|
287
|
+
{:timestamp => ts.to_i,
|
288
|
+
:state => r.get("#{@key}:#{ts}:state"),
|
289
|
+
:summary => r.get("#{@key}:#{ts}:summary")}
|
290
|
+
}
|
291
|
+
end
|
292
|
+
|
293
|
+
# The redis commands in a pipeline block return future objects, which
|
294
|
+
# must be evaluated. This relies on a patch in flapjack/patches.rb to
|
295
|
+
# make the Future objects report their class.
|
296
|
+
state_data.collect {|sd|
|
297
|
+
sd.merge!(sd) {|k,ov,nv|
|
298
|
+
(nv.class == Redis::Future) ? nv.value : nv
|
299
|
+
}
|
300
|
+
}
|
301
|
+
end
|
302
|
+
|
303
|
+
# requires a known state timestamp, i.e. probably one returned via
|
304
|
+
# historical_states. will find the one before that in the sorted set,
|
305
|
+
# if any.
|
306
|
+
def historical_state_before(timestamp)
|
307
|
+
pos = @redis.zrank("#{@key}:sorted_state_timestamps", timestamp)
|
308
|
+
return if pos < 1
|
309
|
+
ts = @redis.zrange("#{@key}:sorted_state_timestamps", pos - 1, pos)
|
310
|
+
return if ts.nil? || ts.empty?
|
311
|
+
{:timestamp => ts.first.to_i,
|
312
|
+
:state => @redis.get("#{@key}:#{ts.first}:state"),
|
313
|
+
:summary => @redis.get("#{@key}:#{ts.first}:summary")}
|
314
|
+
end
|
315
|
+
|
316
|
+
def historical_state_after(timestamp)
|
317
|
+
pos = @redis.zrank("#{@key}:sorted_state_timestamps", timestamp)
|
318
|
+
ts = @redis.zrange("#{@key}:sorted_state_timestamps", pos + 1, pos + 2)
|
319
|
+
return if ts.nil? || ts.empty?
|
320
|
+
{:timestamp => ts.first.to_i,
|
321
|
+
:state => @redis.get("#{@key}:#{ts.first}:state"),
|
322
|
+
:summary => @redis.get("#{@key}:#{ts.first}:summary")}
|
323
|
+
end
|
324
|
+
|
325
|
+
# Returns a list of maintenance periods (either unscheduled or scheduled) for this
|
326
|
+
# entity check, sorted by timestamp.
|
327
|
+
#
|
328
|
+
# start_time and end_time should be passed as integer timestamps; these timestamps
|
329
|
+
# will be considered inclusively, so, e.g. coverage for a day should go
|
330
|
+
# from midnight to 11:59:59 PM. Pass nil for either end to leave that
|
331
|
+
# side unbounded.
|
332
|
+
def maintenances(start_time, end_time, opts = {})
|
333
|
+
sched = opts[:scheduled] ? 'scheduled' : 'unscheduled'
|
334
|
+
|
335
|
+
start_time ||= '-inf'
|
336
|
+
end_time ||= '+inf'
|
337
|
+
order = opts[:order]
|
338
|
+
query = (order && 'desc'.eql?(order.downcase)) ? :zrevrangebyscore : :zrangebyscore
|
339
|
+
maint_ts = @redis.send(query, "#{@key}:sorted_#{sched}_maintenance_timestamps", start_time, end_time)
|
340
|
+
|
341
|
+
maint_data = nil
|
342
|
+
|
343
|
+
@redis.multi do |r|
|
344
|
+
maint_data = maint_ts.collect {|ts|
|
345
|
+
{:start_time => ts.to_i,
|
346
|
+
:duration => r.zscore("#{@key}:#{sched}_maintenances", ts),
|
347
|
+
:summary => r.get("#{@key}:#{ts}:#{sched}_maintenance:summary"),
|
348
|
+
}
|
349
|
+
}
|
350
|
+
end
|
351
|
+
|
352
|
+
# The redis commands in a pipeline block return future objects, which
|
353
|
+
# must be evaluated. This relies on a patch in flapjack/patches.rb to
|
354
|
+
# make the Future objects report their class.
|
355
|
+
maint_data.collect {|md|
|
356
|
+
md.merge!(md) {|k,ov,nv| (nv.class == Redis::Future) ? nv.value : nv }
|
357
|
+
md[:end_time] = (md[:start_time] + md[:duration]).floor
|
358
|
+
md
|
359
|
+
}
|
360
|
+
end
|
361
|
+
|
362
|
+
# returns an array of pagerduty credentials. If more than one contact for this entity_check
|
363
|
+
# has pagerduty credentials then there'll be one hash in the array for each set of
|
364
|
+
# credentials.
|
365
|
+
def pagerduty_credentials(options)
|
366
|
+
creds = []
|
367
|
+
raise "Redis connection not set" unless redis = options[:redis]
|
368
|
+
logger = options[:logger]
|
369
|
+
|
370
|
+
contacts = Flapjack::Data::Contact.find_all_for_entity_check(self, { :redis => redis, :logger => logger })
|
371
|
+
contacts.each {|contact|
|
372
|
+
cred = Flapjack::Data::Contact.pagerduty_credentials_for_contact(contact, { :redis => redis, :logger => logger })
|
373
|
+
creds << cred if cred
|
374
|
+
}
|
375
|
+
creds
|
376
|
+
end
|
377
|
+
|
378
|
+
private
|
379
|
+
|
380
|
+
# Passing around the redis handle like this is a SMELL.
|
381
|
+
def initialize(entity, check, options = {})
|
382
|
+
raise "Redis connection not set" unless @redis = options[:redis]
|
383
|
+
raise "Invalid entity" unless @entity = entity
|
384
|
+
raise "Invalid check" unless @check = check
|
385
|
+
@key = "#{entity.name}:#{check}"
|
386
|
+
end
|
387
|
+
|
388
|
+
def validate_state(state)
|
389
|
+
[STATE_OK, STATE_WARNING, STATE_CRITICAL, STATE_UNKNOWN].include?(state)
|
390
|
+
end
|
391
|
+
|
392
|
+
end
|
393
|
+
|
394
|
+
end
|
395
|
+
|
396
|
+
end
|