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 +1,66 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
# Formats entity data for presentation by the API methods in Flapjack::API.
|
4
|
+
# Currently this just aggregates all of the check data for an entity, leaving
|
5
|
+
# clients to make any further calculations for themselves.
|
6
|
+
|
7
|
+
require 'sinatra/base'
|
8
|
+
|
9
|
+
require 'flapjack/api/entity_check_presenter'
|
10
|
+
require 'flapjack/data/entity_check'
|
11
|
+
|
12
|
+
module Flapjack
|
13
|
+
|
14
|
+
class API < Sinatra::Base
|
15
|
+
|
16
|
+
class EntityPresenter
|
17
|
+
|
18
|
+
def initialize(entity, options = {})
|
19
|
+
@entity = entity
|
20
|
+
@redis = options[:redis]
|
21
|
+
end
|
22
|
+
|
23
|
+
def outages(start_time, end_time)
|
24
|
+
checks.collect {|c|
|
25
|
+
{:check => c, :outages => check_presenter(c).outages(start_time, end_time)}
|
26
|
+
}
|
27
|
+
end
|
28
|
+
|
29
|
+
def unscheduled_maintenance(start_time, end_time)
|
30
|
+
checks.collect {|c|
|
31
|
+
{:check => c, :unscheduled_maintenance =>
|
32
|
+
check_presenter(c).unscheduled_maintenance(start_time, end_time)}
|
33
|
+
}
|
34
|
+
end
|
35
|
+
|
36
|
+
def scheduled_maintenance(start_time, end_time)
|
37
|
+
checks.collect {|c|
|
38
|
+
{:check => c, :scheduled_maintenance =>
|
39
|
+
check_presenter(c).scheduled_maintenance(start_time, end_time)}
|
40
|
+
}
|
41
|
+
end
|
42
|
+
|
43
|
+
def downtime(start_time, end_time)
|
44
|
+
checks.collect {|c|
|
45
|
+
{:check => c, :downtime =>
|
46
|
+
check_presenter(c).downtime(start_time, end_time)}
|
47
|
+
}
|
48
|
+
end
|
49
|
+
|
50
|
+
private
|
51
|
+
|
52
|
+
def checks
|
53
|
+
@check_list ||= @entity.check_list
|
54
|
+
end
|
55
|
+
|
56
|
+
def check_presenter(check)
|
57
|
+
entity_check = Flapjack::Data::EntityCheck.for_entity(@entity, check,
|
58
|
+
:redis => @redis)
|
59
|
+
presenter = Flapjack::API::EntityCheckPresenter.new(entity_check)
|
60
|
+
end
|
61
|
+
|
62
|
+
end
|
63
|
+
|
64
|
+
end
|
65
|
+
|
66
|
+
end
|
@@ -0,0 +1,288 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
require 'eventmachine'
|
4
|
+
# the redis/synchrony gems need to be required in this particular order, see
|
5
|
+
# the redis-rb README for details
|
6
|
+
require 'hiredis'
|
7
|
+
require 'em-synchrony'
|
8
|
+
require 'redis/connection/synchrony'
|
9
|
+
require 'redis'
|
10
|
+
require 'em-resque'
|
11
|
+
require 'em-resque/worker'
|
12
|
+
require 'thin'
|
13
|
+
|
14
|
+
require 'flapjack/patches'
|
15
|
+
|
16
|
+
require 'flapjack/api'
|
17
|
+
require 'flapjack/daemonizing'
|
18
|
+
require 'flapjack/executive'
|
19
|
+
require 'flapjack/jabber'
|
20
|
+
require 'flapjack/pagerduty'
|
21
|
+
require 'flapjack/notification/email'
|
22
|
+
require 'flapjack/notification/sms'
|
23
|
+
require 'flapjack/web'
|
24
|
+
|
25
|
+
module Flapjack
|
26
|
+
|
27
|
+
class Coordinator
|
28
|
+
|
29
|
+
include Flapjack::Daemonizable
|
30
|
+
|
31
|
+
def initialize(config = {})
|
32
|
+
@config = config
|
33
|
+
@pikelets = []
|
34
|
+
|
35
|
+
@logger = Log4r::Logger.new("flapjack-coordinator")
|
36
|
+
@logger.add(Log4r::StdoutOutputter.new("flapjack-coordinator"))
|
37
|
+
@logger.add(Log4r::SyslogOutputter.new("flapjack-coordinator"))
|
38
|
+
end
|
39
|
+
|
40
|
+
def start(options = {})
|
41
|
+
# FIXME raise error if config not set, or empty
|
42
|
+
|
43
|
+
if options[:daemonize]
|
44
|
+
daemonize
|
45
|
+
else
|
46
|
+
setup
|
47
|
+
end
|
48
|
+
end
|
49
|
+
|
50
|
+
def after_daemonize
|
51
|
+
setup
|
52
|
+
end
|
53
|
+
|
54
|
+
def stop
|
55
|
+
shutdown
|
56
|
+
end
|
57
|
+
|
58
|
+
private
|
59
|
+
|
60
|
+
def setup
|
61
|
+
|
62
|
+
# FIXME: the following is currently repeated in flapjack-populator and
|
63
|
+
# flapjack-nagios-receiver - move to a method in a module and include it
|
64
|
+
redis_host = @config['redis']['host'] || '127.0.0.1'
|
65
|
+
redis_port = @config['redis']['port'] || 6379
|
66
|
+
redis_path = @config['redis']['path'] || nil
|
67
|
+
redis_db = @config['redis']['db'] || 0
|
68
|
+
|
69
|
+
if redis_path
|
70
|
+
@redis_options = { :db => redis_db, :path => redis_path }
|
71
|
+
else
|
72
|
+
@redis_options = { :db => redis_db, :host => redis_host, :port => redis_port }
|
73
|
+
end
|
74
|
+
|
75
|
+
EM.synchrony do
|
76
|
+
|
77
|
+
@logger.debug "config keys: #{@config.keys}"
|
78
|
+
|
79
|
+
pikelet_keys = ['executive', 'jabber_gateway', 'pagerduty_gateway',
|
80
|
+
'email_notifier', 'sms_notifier', 'web', 'api']
|
81
|
+
|
82
|
+
@config.keys.each do |pikelet_type|
|
83
|
+
next unless pikelet_keys.include?(pikelet_type) &&
|
84
|
+
@config[pikelet_type].is_a?(Hash) &&
|
85
|
+
@config[pikelet_type]['enabled']
|
86
|
+
@logger.debug "coordinator is now initialising the #{pikelet_type} pikelet"
|
87
|
+
pikelet_cfg = @config[pikelet_type]
|
88
|
+
|
89
|
+
case pikelet_type
|
90
|
+
when 'executive', 'jabber_gateway', 'pagerduty_gateway'
|
91
|
+
build_pikelet(pikelet_type, pikelet_cfg)
|
92
|
+
when 'web', 'api'
|
93
|
+
build_thin_pikelet(pikelet_type, pikelet_cfg)
|
94
|
+
when 'email_notifier', 'sms_notifier'
|
95
|
+
build_resque_pikelet(pikelet_type, pikelet_cfg)
|
96
|
+
end
|
97
|
+
end
|
98
|
+
|
99
|
+
setup_signals
|
100
|
+
end
|
101
|
+
|
102
|
+
end
|
103
|
+
|
104
|
+
def setup_signals
|
105
|
+
trap('INT') { stop }
|
106
|
+
trap('TERM') { stop }
|
107
|
+
unless RUBY_PLATFORM =~ /mswin/
|
108
|
+
trap('QUIT') { stop }
|
109
|
+
# trap('HUP') { }
|
110
|
+
end
|
111
|
+
end
|
112
|
+
|
113
|
+
def build_pikelet(pikelet_type, pikelet_cfg)
|
114
|
+
pikelet_class = case pikelet_type
|
115
|
+
when 'executive'
|
116
|
+
Flapjack::Executive
|
117
|
+
when 'jabber_gateway'
|
118
|
+
Flapjack::Jabber
|
119
|
+
when 'pagerduty_gateway'
|
120
|
+
Flapjack::Pagerduty
|
121
|
+
end
|
122
|
+
return unless pikelet_class
|
123
|
+
|
124
|
+
f = Fiber.new {
|
125
|
+
begin
|
126
|
+
pikelet = pikelet_class.new
|
127
|
+
@pikelets.detect {|p| p[:type] == pikelet_type}[:instance] = pikelet
|
128
|
+
pikelet.bootstrap(:redis => @redis_options, :config => pikelet_cfg)
|
129
|
+
pikelet.main
|
130
|
+
rescue Exception => e
|
131
|
+
trace = e.backtrace.join("\n")
|
132
|
+
@logger.fatal "#{e.message}\n#{trace}"
|
133
|
+
stop
|
134
|
+
end
|
135
|
+
}
|
136
|
+
@pikelets << {:fiber => f, :type => pikelet_type}
|
137
|
+
f.resume
|
138
|
+
@logger.debug "new fiber created for #{pikelet_type}"
|
139
|
+
end
|
140
|
+
|
141
|
+
def build_thin_pikelet(pikelet_type, pikelet_cfg)
|
142
|
+
pikelet_class = case pikelet_type
|
143
|
+
when 'web'
|
144
|
+
Flapjack::Web
|
145
|
+
when 'api'
|
146
|
+
Flapjack::API
|
147
|
+
end
|
148
|
+
return unless pikelet_class
|
149
|
+
|
150
|
+
port = nil
|
151
|
+
if pikelet_cfg['port']
|
152
|
+
port = pikelet_cfg['port'].to_i
|
153
|
+
end
|
154
|
+
|
155
|
+
port = 3001 if (port.nil? || port <= 0 || port > 65535)
|
156
|
+
|
157
|
+
pikelet_class.class_variable_set('@@redis', build_redis_connection_pool)
|
158
|
+
|
159
|
+
Thin::Logging.silent = true
|
160
|
+
|
161
|
+
pikelet = Thin::Server.new('0.0.0.0', port, pikelet_class, :signals => false)
|
162
|
+
@pikelets << {:instance => pikelet, :type => pikelet_type}
|
163
|
+
pikelet.start
|
164
|
+
@logger.debug "new thin server instance started for #{pikelet_type}"
|
165
|
+
end
|
166
|
+
|
167
|
+
def build_resque_pikelet(pikelet_type, pikelet_cfg)
|
168
|
+
pikelet_class = case pikelet_type
|
169
|
+
when 'email_notifier'
|
170
|
+
Flapjack::Notification::Email
|
171
|
+
when 'sms_notifier'
|
172
|
+
Flapjack::Notification::Sms
|
173
|
+
end
|
174
|
+
return unless pikelet_class
|
175
|
+
|
176
|
+
# set up connection pooling, stop resque errors (ensure that it's only
|
177
|
+
# done once)
|
178
|
+
if (['email_notifier', 'sms_notifier'] & @pikelets.collect {|p| p[:type]}).empty?
|
179
|
+
::Resque.redis = build_redis_connection_pool
|
180
|
+
## NB: can override the default 'resque' namespace like this
|
181
|
+
#::Resque.redis.namespace = 'flapjack'
|
182
|
+
end
|
183
|
+
|
184
|
+
# See https://github.com/mikel/mail/blob/master/lib/mail/mail.rb#L53
|
185
|
+
# & https://github.com/mikel/mail/blob/master/spec/mail/configuration_spec.rb
|
186
|
+
# for details of configuring mail gem. defaults to SMTP, localhost, port 25
|
187
|
+
|
188
|
+
if pikelet_type.eql?('email_notifier')
|
189
|
+
smtp_config = {}
|
190
|
+
|
191
|
+
if pikelet_cfg['smtp_config']
|
192
|
+
smtp_config = pikelet_cfg['smtp_config'].keys.inject({}) do |ret,obj|
|
193
|
+
ret[obj.to_sym] = pikelet_cfg['smtp_config'][obj]
|
194
|
+
ret
|
195
|
+
end
|
196
|
+
end
|
197
|
+
|
198
|
+
Mail.defaults {
|
199
|
+
delivery_method :smtp, {:enable_starttls_auto => false}.merge(smtp_config)
|
200
|
+
}
|
201
|
+
end
|
202
|
+
|
203
|
+
pikelet_class.class_variable_set('@@config', pikelet_cfg)
|
204
|
+
|
205
|
+
f = Fiber.new {
|
206
|
+
begin
|
207
|
+
# TODO error if pikelet_cfg['queue'].nil?
|
208
|
+
pikelet = EM::Resque::Worker.new(pikelet_cfg['queue'])
|
209
|
+
@pikelets.detect {|p| p[:type] == pikelet_type}[:instance] = pikelet
|
210
|
+
# # Use these to debug the resque workers
|
211
|
+
# flapjack_rsq.verbose = true
|
212
|
+
#flapjack_rsq.very_verbose = true
|
213
|
+
pikelet.work(0.1)
|
214
|
+
rescue Exception => e
|
215
|
+
trace = e.backtrace.join("\n")
|
216
|
+
@logger.fatal "#{e.message}\n#{trace}"
|
217
|
+
stop
|
218
|
+
end
|
219
|
+
}
|
220
|
+
@pikelets << {:fiber => f, :type => pikelet_type}
|
221
|
+
f.resume
|
222
|
+
@logger.debug "new fiber created for #{pikelet_type}"
|
223
|
+
end
|
224
|
+
|
225
|
+
def build_redis_connection_pool(options = {})
|
226
|
+
EventMachine::Synchrony::ConnectionPool.new(:size => options[:size] || 5) do
|
227
|
+
::Redis.new(@redis_options.merge(:driver => (options[:driver] || 'synchrony')))
|
228
|
+
end
|
229
|
+
end
|
230
|
+
|
231
|
+
# # TODO rewrite to be less spammy -- print only initial state and changes
|
232
|
+
# def health_check
|
233
|
+
# @pikelets.each do |pik|
|
234
|
+
# if pik[:instance].is_a?(Thin::Server)
|
235
|
+
# s = pik[:instance].backend.size
|
236
|
+
# @logger.debug "thin on port #{pik[:instance].port} - #{s} connections"
|
237
|
+
# elsif pik[:fiber]
|
238
|
+
# @logger.debug "#{pik[:type]}: #{pik[:fiber].alive? ? 'alive' : 'dead'}"
|
239
|
+
# end
|
240
|
+
# end
|
241
|
+
# end
|
242
|
+
|
243
|
+
def shutdown
|
244
|
+
@pikelets.each do |pik|
|
245
|
+
case pik[:instance]
|
246
|
+
when Flapjack::Executive, Flapjack::Jabber, Flapjack::Pagerduty
|
247
|
+
if pik[:fiber] && pik[:fiber].alive?
|
248
|
+
pik[:instance].stop
|
249
|
+
Fiber.new {
|
250
|
+
# this needs to use a separate Redis connection from the pikelet's
|
251
|
+
# one, as that's in the middle of its blpop
|
252
|
+
r = Redis.new(@redis_options.merge(:driver => 'synchrony'))
|
253
|
+
pik[:instance].add_shutdown_event(:redis => r)
|
254
|
+
r.quit
|
255
|
+
}.resume
|
256
|
+
end
|
257
|
+
when EM::Resque::Worker
|
258
|
+
if pik[:fiber] && pik[:fiber].alive?
|
259
|
+
# resque is polling, so we don't need a shutdown object
|
260
|
+
pik[:instance].shutdown
|
261
|
+
end
|
262
|
+
when Thin::Server # web, api
|
263
|
+
# drop from this side, as HTTP keepalive etc. means browsers
|
264
|
+
# keep connections alive for ages, and we'd be hanging around
|
265
|
+
# waiting for them to drop
|
266
|
+
pik[:instance].stop!
|
267
|
+
end
|
268
|
+
end
|
269
|
+
|
270
|
+
fibers = @pikelets.collect {|p| p[:fiber] }.compact
|
271
|
+
thin_pikelets = @pikelets.collect {|p| p[:instance]}.select {|i| i.is_a?(Thin::Server) }
|
272
|
+
|
273
|
+
Fiber.new {
|
274
|
+
loop do
|
275
|
+
# health_check
|
276
|
+
if fibers.any?(&:alive?) || thin_pikelets.any?{|tp| !tp.backend.empty? }
|
277
|
+
EM::Synchrony.sleep 0.25
|
278
|
+
else
|
279
|
+
EM.stop
|
280
|
+
break
|
281
|
+
end
|
282
|
+
end
|
283
|
+
}.resume
|
284
|
+
end
|
285
|
+
|
286
|
+
end
|
287
|
+
|
288
|
+
end
|
@@ -0,0 +1,186 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
# Copied from thin.
|
4
|
+
|
5
|
+
require 'etc'
|
6
|
+
require 'daemons' unless RUBY_PLATFORM =~ /mswin/
|
7
|
+
|
8
|
+
module Process
|
9
|
+
# Returns +true+ the process identified by +pid+ is running.
|
10
|
+
def running?(pid)
|
11
|
+
Process.getpgid(pid) != -1
|
12
|
+
rescue Errno::EPERM
|
13
|
+
true
|
14
|
+
rescue Errno::ESRCH
|
15
|
+
false
|
16
|
+
end
|
17
|
+
module_function :running?
|
18
|
+
end
|
19
|
+
|
20
|
+
module Flapjack
|
21
|
+
# Raised when the pid file already exist starting as a daemon.
|
22
|
+
class PidFileExist < RuntimeError; end
|
23
|
+
|
24
|
+
# Module included in classes that can be turned into a daemon.
|
25
|
+
# Handle stuff like:
|
26
|
+
# * storing the PID in a file
|
27
|
+
# * redirecting output to the log file
|
28
|
+
# * changing process privileges
|
29
|
+
# * killing the process gracefully
|
30
|
+
module Daemonizable
|
31
|
+
attr_accessor :pid_file, :log_file
|
32
|
+
|
33
|
+
def self.included(base)
|
34
|
+
base.extend ClassMethods
|
35
|
+
end
|
36
|
+
|
37
|
+
def pid
|
38
|
+
File.exist?(pid_file) ? open(pid_file).read.to_i : nil
|
39
|
+
end
|
40
|
+
|
41
|
+
# Turns the current script into a daemon process that detaches from the console.
|
42
|
+
def daemonize
|
43
|
+
raise PlatformNotSupported, 'Daemonizing is not supported on Windows' if RUBY_PLATFORM =~ /mswin/
|
44
|
+
raise ArgumentError, 'You must specify a pid_file to daemonize' unless @pid_file
|
45
|
+
|
46
|
+
remove_stale_pid_file
|
47
|
+
|
48
|
+
pwd = Dir.pwd # Current directory is changed during daemonization, so store it
|
49
|
+
|
50
|
+
# HACK we need to create the directory before daemonization to prevent a bug under 1.9
|
51
|
+
# ignoring all signals when the directory is created after daemonization.
|
52
|
+
FileUtils.mkdir_p File.dirname(@pid_file)
|
53
|
+
FileUtils.mkdir_p File.dirname(@log_file)
|
54
|
+
|
55
|
+
Daemonize.daemonize(File.expand_path(@log_file), 'flapjack server')
|
56
|
+
|
57
|
+
Dir.chdir(pwd)
|
58
|
+
|
59
|
+
write_pid_file
|
60
|
+
|
61
|
+
self.after_daemonize if self.respond_to? :after_daemonize
|
62
|
+
|
63
|
+
at_exit do
|
64
|
+
puts ">> Exiting!"
|
65
|
+
remove_pid_file
|
66
|
+
end
|
67
|
+
end
|
68
|
+
|
69
|
+
# Change privileges of the process
|
70
|
+
# to the specified user and group.
|
71
|
+
def change_privilege(user, group=user)
|
72
|
+
puts ">> Changing process privilege to #{user}:#{group}"
|
73
|
+
|
74
|
+
uid, gid = Process.euid, Process.egid
|
75
|
+
target_uid = Etc.getpwnam(user).uid
|
76
|
+
target_gid = Etc.getgrnam(group).gid
|
77
|
+
|
78
|
+
if uid != target_uid || gid != target_gid
|
79
|
+
# Change process ownership
|
80
|
+
Process.initgroups(user, target_gid)
|
81
|
+
Process::GID.change_privilege(target_gid)
|
82
|
+
Process::UID.change_privilege(target_uid)
|
83
|
+
end
|
84
|
+
rescue Errno::EPERM => e
|
85
|
+
puts "Couldn't change user and group to #{user}:#{group}: #{e}"
|
86
|
+
end
|
87
|
+
|
88
|
+
# Register a proc to be called to restart the server.
|
89
|
+
def on_restart(&block)
|
90
|
+
@on_restart = block
|
91
|
+
end
|
92
|
+
|
93
|
+
# Restart the server.
|
94
|
+
def restart
|
95
|
+
if @on_restart
|
96
|
+
puts '>> Restarting ...'
|
97
|
+
stop
|
98
|
+
remove_pid_file
|
99
|
+
@on_restart.call
|
100
|
+
exit!
|
101
|
+
end
|
102
|
+
end
|
103
|
+
|
104
|
+
module ClassMethods
|
105
|
+
# Send a QUIT or INT (if timeout is +0+) signal the process which
|
106
|
+
# PID is stored in +pid_file+.
|
107
|
+
# If the process is still running after +timeout+, KILL signal is
|
108
|
+
# sent.
|
109
|
+
def kill(pid_file, timeout=60)
|
110
|
+
if timeout == 0
|
111
|
+
send_signal('INT', pid_file, timeout)
|
112
|
+
else
|
113
|
+
send_signal('QUIT', pid_file, timeout)
|
114
|
+
end
|
115
|
+
end
|
116
|
+
|
117
|
+
# Restart the server by sending HUP signal.
|
118
|
+
def restart(pid_file)
|
119
|
+
send_signal('HUP', pid_file)
|
120
|
+
end
|
121
|
+
|
122
|
+
# Send a +signal+ to the process which PID is stored in +pid_file+.
|
123
|
+
def send_signal(signal, pid_file, timeout=60)
|
124
|
+
if pid = read_pid_file(pid_file)
|
125
|
+
puts "Sending #{signal} signal to process #{pid} ... "
|
126
|
+
Process.kill(signal, pid)
|
127
|
+
Timeout.timeout(timeout) do
|
128
|
+
sleep 0.1 while Process.running?(pid)
|
129
|
+
end
|
130
|
+
else
|
131
|
+
puts "Can't stop process, no PID found in #{pid_file}"
|
132
|
+
end
|
133
|
+
rescue Timeout::Error
|
134
|
+
puts "Timeout!"
|
135
|
+
force_kill pid_file
|
136
|
+
rescue Interrupt
|
137
|
+
force_kill pid_file
|
138
|
+
rescue Errno::ESRCH # No such process
|
139
|
+
puts "process not found!"
|
140
|
+
force_kill pid_file
|
141
|
+
end
|
142
|
+
|
143
|
+
def force_kill(pid_file)
|
144
|
+
if pid = read_pid_file(pid_file)
|
145
|
+
puts "Sending KILL signal to process #{pid} ... "
|
146
|
+
Process.kill("KILL", pid)
|
147
|
+
File.delete(pid_file) if File.exist?(pid_file)
|
148
|
+
else
|
149
|
+
puts "Can't stop process, no PID found in #{pid_file}"
|
150
|
+
end
|
151
|
+
end
|
152
|
+
|
153
|
+
def read_pid_file(file)
|
154
|
+
if File.file?(file) && pid = File.read(file)
|
155
|
+
pid.to_i
|
156
|
+
else
|
157
|
+
nil
|
158
|
+
end
|
159
|
+
end
|
160
|
+
end
|
161
|
+
|
162
|
+
protected
|
163
|
+
def remove_pid_file
|
164
|
+
File.delete(@pid_file) if @pid_file && File.exists?(@pid_file)
|
165
|
+
end
|
166
|
+
|
167
|
+
def write_pid_file
|
168
|
+
puts ">> Writing PID to #{@pid_file}"
|
169
|
+
open(@pid_file,"w") { |f| f.write(Process.pid) }
|
170
|
+
File.chmod(0644, @pid_file)
|
171
|
+
end
|
172
|
+
|
173
|
+
# If PID file is stale, remove it.
|
174
|
+
def remove_stale_pid_file
|
175
|
+
if File.exist?(@pid_file)
|
176
|
+
if pid && Process.running?(pid)
|
177
|
+
raise PidFileExist, "#{@pid_file} already exists, seems like it's already running (process ID: #{pid}). " +
|
178
|
+
"Stop the process or delete #{@pid_file}."
|
179
|
+
else
|
180
|
+
puts ">> Deleting stale PID file #{@pid_file}"
|
181
|
+
remove_pid_file
|
182
|
+
end
|
183
|
+
end
|
184
|
+
end
|
185
|
+
end
|
186
|
+
end
|