staugaard-cloudmaster 0.1.3 → 0.1.4
Sign up to get free protection for your applications and to get access to all the features.
- data/VERSION.yml +1 -1
- data/app/active_set_factory.rb +16 -0
- data/app/active_set_none.rb +16 -0
- data/app/active_set_queue.rb +27 -0
- data/app/active_set_s3.rb +25 -0
- data/app/configuration.rb +85 -0
- data/app/default-config.ini +95 -0
- data/app/ec2_image_enumerator.rb +41 -0
- data/app/ec2_instance_enumerator.rb +25 -0
- data/app/instance.rb +146 -0
- data/app/instance_pool.rb +326 -0
- data/app/named_queue.rb +75 -0
- data/app/policy.rb +113 -0
- data/app/policy_daytime.rb +18 -0
- data/app/policy_factory.rb +16 -0
- data/app/policy_fixed.rb +19 -0
- data/app/policy_job.rb +54 -0
- data/app/policy_limit.rb +68 -0
- data/app/policy_manual.rb +36 -0
- data/app/policy_resource.rb +110 -0
- data/app/pool_configuration.rb +172 -0
- data/app/pool_manager.rb +239 -0
- data/app/pool_runner.rb +54 -0
- data/app/reporter.rb +81 -0
- data/app/status_parser_factory.rb +16 -0
- data/app/status_parser_lifeguard.rb +48 -0
- data/app/status_parser_std.rb +11 -0
- metadata +27 -1
@@ -0,0 +1,326 @@
|
|
1
|
+
require 'aws_context'
|
2
|
+
require 'ec2_instance_enumerator'
|
3
|
+
require 'instance'
|
4
|
+
require 'yaml'
|
5
|
+
|
6
|
+
module Cloudmaster
|
7
|
+
|
8
|
+
# Stores and operates on a collection of instances
|
9
|
+
#
|
10
|
+
# Internally, instances are stored as an array of Instance objects.
|
11
|
+
class InstancePool
|
12
|
+
include Enumerable
|
13
|
+
|
14
|
+
# Create an instance pool.
|
15
|
+
# This class knows how to start and stop instances, and how to detect
|
16
|
+
# that new instance have come about, or existing ones have gone away.
|
17
|
+
# The constructor takes:
|
18
|
+
# [config] describes configurable instance properties
|
19
|
+
def initialize(reporter, config)
|
20
|
+
@ec2 = AwsContext.instance.ec2
|
21
|
+
@reporter = reporter
|
22
|
+
@config = config
|
23
|
+
@state_change_time = Clock.now
|
24
|
+
@instances = [] # holds Instance objects
|
25
|
+
end
|
26
|
+
|
27
|
+
# private
|
28
|
+
|
29
|
+
# Create and return options, in a way that is acceptable to EC2.
|
30
|
+
def start_opts
|
31
|
+
groups = @config[:security_groups]
|
32
|
+
# this can throw an exception if groups is not formetteed properly
|
33
|
+
begin
|
34
|
+
groups = eval(groups) if groups.kind_of?(String)
|
35
|
+
rescue
|
36
|
+
groups = [:default]
|
37
|
+
end
|
38
|
+
{:key_name => @config[:key_pair_name],
|
39
|
+
:user_data => YAML.dump(@config[:user_data]),
|
40
|
+
:security_groups => groups,
|
41
|
+
:instance_type => @config[:instance_type]}
|
42
|
+
end
|
43
|
+
|
44
|
+
def max2(a, b)
|
45
|
+
a > b ? a : b
|
46
|
+
end
|
47
|
+
|
48
|
+
# Allows iteration through instances.
|
49
|
+
# So enumeration on InstancePool is implicitly enumeration
|
50
|
+
# on @instances.
|
51
|
+
def each
|
52
|
+
@instances.each {|i| yield i}
|
53
|
+
end
|
54
|
+
|
55
|
+
# return first instance
|
56
|
+
def first
|
57
|
+
@instances.first
|
58
|
+
end
|
59
|
+
|
60
|
+
# Return the number of instances in the pool.
|
61
|
+
def size
|
62
|
+
@instances.size
|
63
|
+
end
|
64
|
+
|
65
|
+
# Create an instance and add to the list.
|
66
|
+
# Return the newly created instance.
|
67
|
+
def add(id, public_dns)
|
68
|
+
new_instance = Instance.new(id, public_dns, @config)
|
69
|
+
@instances << new_instance
|
70
|
+
new_instance
|
71
|
+
end
|
72
|
+
|
73
|
+
# Delete the instance from the list
|
74
|
+
def delete(instance)
|
75
|
+
@instances.delete(instance)
|
76
|
+
end
|
77
|
+
|
78
|
+
# Find an instance given its instance id.
|
79
|
+
def find_by_id(id)
|
80
|
+
find {|i| i.id == id}
|
81
|
+
end
|
82
|
+
|
83
|
+
# Return a list of all instance ids.
|
84
|
+
def id_list
|
85
|
+
map {|i| i.id}
|
86
|
+
end
|
87
|
+
|
88
|
+
# Return the maximum number of instances allowed.
|
89
|
+
def maximum
|
90
|
+
@config[:maximum_number_of_instances].to_i
|
91
|
+
end
|
92
|
+
|
93
|
+
# Return the minimum number of instances allowed.
|
94
|
+
def minimum
|
95
|
+
@config[:minimum_number_of_instances].to_i
|
96
|
+
end
|
97
|
+
|
98
|
+
# Return true if the number of instances is less than the minimum.
|
99
|
+
def less_than_minimum?
|
100
|
+
size < minimum
|
101
|
+
end
|
102
|
+
|
103
|
+
# return true if number of instances is more than maximum
|
104
|
+
def greater_than_maximum?
|
105
|
+
size > maximum
|
106
|
+
end
|
107
|
+
|
108
|
+
# Return count of instance below minimum
|
109
|
+
def below_minimum_count
|
110
|
+
less_than_minimum? ? minimum - size : 0
|
111
|
+
end
|
112
|
+
|
113
|
+
# Return count of instance above maximum
|
114
|
+
def above_maximum_count
|
115
|
+
greater_than_maximum? ? size - maximum : 0
|
116
|
+
end
|
117
|
+
|
118
|
+
# Return a list of instances missing public_dns.
|
119
|
+
def missing_public_dns_instances
|
120
|
+
find_all {|i| i.public_dns.nil? || i.public_dns.empty? }
|
121
|
+
end
|
122
|
+
|
123
|
+
# Return ids of all instances missing a public_dns.
|
124
|
+
def missing_public_dns_ids
|
125
|
+
missing_public_dns_instances.collect {|i| i.id}
|
126
|
+
end
|
127
|
+
|
128
|
+
# Find the instance identified by id and update its public_dns
|
129
|
+
# If there is no dns information, then skip it.
|
130
|
+
def update_public_dns(id, public_dns)
|
131
|
+
return if public_dns.nil? || public_dns.empty?
|
132
|
+
i = find_by_id(id)
|
133
|
+
i.public_dns = public_dns if i
|
134
|
+
end
|
135
|
+
|
136
|
+
# Return instances that have not seen status in watchdog_interval
|
137
|
+
def hung_instances
|
138
|
+
find_all {|i| i.watchdog_time_elapsed?}
|
139
|
+
end
|
140
|
+
|
141
|
+
# Return all instances in active state
|
142
|
+
def active_instances
|
143
|
+
find_all {|i| i.state == :active}
|
144
|
+
end
|
145
|
+
|
146
|
+
# Return all instances in shut_down state
|
147
|
+
def shut_down_instances
|
148
|
+
find_all {|i| i.state == :shut_down}
|
149
|
+
end
|
150
|
+
|
151
|
+
# Return instances that are active and have load <= target_load
|
152
|
+
def active_idle_instances
|
153
|
+
target_load = 0
|
154
|
+
active_instances.find_all {|i| i.load_estimate <= target_load}
|
155
|
+
end
|
156
|
+
|
157
|
+
# Shut down all instances who have a load below the target.
|
158
|
+
# Shut down is not the same as stop -- the instances continue to
|
159
|
+
# provide service, but are no longer allocated new clients.
|
160
|
+
def shut_down_idle_instances
|
161
|
+
target_load = @config[:shut_down_threshold].to_i
|
162
|
+
shut_down_instances.find_all {|i| i.load_estimate <= target_load}
|
163
|
+
end
|
164
|
+
|
165
|
+
# Return instances that are shut_down and have
|
166
|
+
# time_since_state_change > shut_down_interval.
|
167
|
+
def shut_down_timeout_instances
|
168
|
+
shut_down_interval = @config[:shut_down_interval].to_i * 60
|
169
|
+
|
170
|
+
shut_down_instances.find_all {|i| i.time_since_state_change > shut_down_interval}
|
171
|
+
end
|
172
|
+
|
173
|
+
# Return the latest time since any state change of any instance.
|
174
|
+
def state_change_time
|
175
|
+
@state_change_time = inject(@state_change_time) do |latest, instance|
|
176
|
+
max2(latest, instance.state_change_time)
|
177
|
+
end
|
178
|
+
end
|
179
|
+
|
180
|
+
# Return the sum of all the extra capacity of active instances
|
181
|
+
# that have excess capacity (load less than target load).
|
182
|
+
def excess_capacity
|
183
|
+
target_load = @config[:target_upper_load].to_f
|
184
|
+
active_instances.inject(0) do |sum, instance|
|
185
|
+
sum + max2(target_load - instance.load_estimate, 0)
|
186
|
+
end
|
187
|
+
end
|
188
|
+
|
189
|
+
# Return the sum of capacity in excess of the target upper load
|
190
|
+
def over_capacity
|
191
|
+
target_load = @config[:target_upper_load].to_f
|
192
|
+
active_instances.inject(0) do |sum, instance|
|
193
|
+
sum + max2(instance.load_estimate - target_load, 0)
|
194
|
+
end
|
195
|
+
end
|
196
|
+
|
197
|
+
# Return the total load of all active instances
|
198
|
+
def total_load
|
199
|
+
active_instances.inject(0) do |sum, instance|
|
200
|
+
sum + instance.load_estimate
|
201
|
+
end
|
202
|
+
end
|
203
|
+
|
204
|
+
# Update the status of an instance using the contents of the status message.
|
205
|
+
def update_status(msg)
|
206
|
+
id = msg[:instance_id]
|
207
|
+
if instance = find_by_id(id)
|
208
|
+
instance.update_status(msg)
|
209
|
+
else
|
210
|
+
@reporter.error("Received status message from unknown instance: #{id}") unless id == 'unknown'
|
211
|
+
end
|
212
|
+
end
|
213
|
+
|
214
|
+
# Return a YAML encoded representation of the active set.
|
215
|
+
# The active set describes the id, public DNS, and load average of
|
216
|
+
# each active instance in the pool.
|
217
|
+
def active_set
|
218
|
+
message = active_instances.collect do |instance|
|
219
|
+
{ :id => instance.id,
|
220
|
+
:public_dns => instance.public_dns,
|
221
|
+
:load_estimate => instance.load_estimate }
|
222
|
+
end
|
223
|
+
YAML.dump(message)
|
224
|
+
end
|
225
|
+
|
226
|
+
# Return all the instance, sortd by lowest load estimate.
|
227
|
+
def sorted_by_lowest_load
|
228
|
+
@instances.sort do |a,b|
|
229
|
+
# Compare the elapsed lifetime status. If the status differs, instances
|
230
|
+
# that have lived beyond the minimum lifetime will be sorted earlier.
|
231
|
+
if a.minimum_lifetime_elapsed? != b.minimum_lifetime_elapsed?
|
232
|
+
if a.minimum_lifetime_elapsed?
|
233
|
+
-1 # This instance has lived long enough, the other hasn't
|
234
|
+
else
|
235
|
+
1 # The other instance has lived long enough, this one hasn't
|
236
|
+
end
|
237
|
+
else
|
238
|
+
a.load_estimate - b.load_estimate
|
239
|
+
end
|
240
|
+
end
|
241
|
+
end
|
242
|
+
|
243
|
+
# Find all instances for which we don't have a public_dns,
|
244
|
+
# For each one,see of EC2 now has the public DNS. If so, store it.
|
245
|
+
def update_public_dns_all
|
246
|
+
missing_ids = missing_public_dns_ids
|
247
|
+
return if missing_ids.size == 0
|
248
|
+
EC2InstanceEnumerator.new(missing_ids).each do |instance|
|
249
|
+
update_public_dns(instance[:id], instance[:public_dns])
|
250
|
+
end
|
251
|
+
end
|
252
|
+
|
253
|
+
# Return instances that match our ami_id that are either pending
|
254
|
+
# or running.
|
255
|
+
# These instances are as returned by ec2: containing fields such as
|
256
|
+
# id and :public_dns.
|
257
|
+
def our_running_instances
|
258
|
+
EC2InstanceEnumerator.new.find_all do |instance|
|
259
|
+
instance[:image_id] == @config[:ami_id] &&
|
260
|
+
%w[pending running].include?(instance[:state])
|
261
|
+
end
|
262
|
+
end
|
263
|
+
|
264
|
+
# Audit the list of instances based on what is currently known to EC2.
|
265
|
+
# In other words, bring our list of instances into agreement with the instances
|
266
|
+
# EC2 knows about by
|
267
|
+
# (1) adding instances that EC2 knows but that we do not, and
|
268
|
+
# (2) deleting instance that EC2 no longer knows about.
|
269
|
+
# This is used initially to build the instance list, and
|
270
|
+
# periodically thereafter to catch instances started or stopped
|
271
|
+
# outside cloudmaster.
|
272
|
+
def audit_existing_instances
|
273
|
+
running_instances = our_running_instances
|
274
|
+
|
275
|
+
# add running instances that we don't have
|
276
|
+
running_instances.each do |running|
|
277
|
+
if ! find_by_id(running[:id])
|
278
|
+
add(running[:id], running[:public_dns])
|
279
|
+
@reporter.info("Instance discovered #{running[:public_dns]}", running[:id])
|
280
|
+
end
|
281
|
+
end
|
282
|
+
# delete instances that are no longer running
|
283
|
+
each do |instance|
|
284
|
+
if ! running_instances.find {|running| running[:id] == instance.id}
|
285
|
+
delete(instance)
|
286
|
+
@reporter.info("Instance disappeared #{instance.public_dns}", instance.id)
|
287
|
+
end
|
288
|
+
end
|
289
|
+
end
|
290
|
+
|
291
|
+
# Start the given number of instances.
|
292
|
+
# Remember started instances by creating an Instance object and storing it
|
293
|
+
# in the pool.
|
294
|
+
# Return an array of the ones we just started.
|
295
|
+
def start_n_instances(number_to_start)
|
296
|
+
return [] if number_to_start <= 0
|
297
|
+
started_instances = @ec2.run_instances(@config[:ami_id], 1,
|
298
|
+
number_to_start, start_opts)[:instances]
|
299
|
+
started_instances.collect do |started_instance|
|
300
|
+
# the public dns is not available yet
|
301
|
+
add(started_instance[:id], nil)
|
302
|
+
end
|
303
|
+
end
|
304
|
+
|
305
|
+
# Stop the given set of instances.
|
306
|
+
# Remove stopped instance from the pool.
|
307
|
+
# Return an array of stopped instances.
|
308
|
+
def stop_instances(instances_to_stop)
|
309
|
+
instances_to_stop.collect do |instance|
|
310
|
+
@ec2.terminate_instances(instance.id.to_s)
|
311
|
+
delete(instance)
|
312
|
+
instance
|
313
|
+
end
|
314
|
+
end
|
315
|
+
|
316
|
+
# Shut down the given set of instances.
|
317
|
+
# Set the state to shut_down
|
318
|
+
# Return an array of shut down instances.
|
319
|
+
def shut_down(instances_to_shut_down)
|
320
|
+
instances_to_shut_down.collect do |instance|
|
321
|
+
instance.shutdown
|
322
|
+
instance
|
323
|
+
end
|
324
|
+
end
|
325
|
+
end
|
326
|
+
end
|
data/app/named_queue.rb
ADDED
@@ -0,0 +1,75 @@
|
|
1
|
+
require 'aws_context'
|
2
|
+
|
3
|
+
module Cloudmaster
|
4
|
+
|
5
|
+
# Implements a queue specified by name.
|
6
|
+
# The queue name is used to look up the actual SQS queue.
|
7
|
+
# The queue name is used in a substring match, so possibly more than one
|
8
|
+
# string may match.
|
9
|
+
# If no queue is found for given name, or if there are
|
10
|
+
# multiple, the NamedQueue throws an exception.
|
11
|
+
# Implements the read and delete operations.
|
12
|
+
# It also caches queue_depth, so it need not query SQS every time
|
13
|
+
# it is needed.
|
14
|
+
class NamedQueue
|
15
|
+
attr_reader :queue_depth
|
16
|
+
|
17
|
+
# Create a queue given the queue name.
|
18
|
+
# A SQS interface must be supplied: it is used to interact with Amazon.
|
19
|
+
# The queue name is given, and is used to look up the actual queue.
|
20
|
+
# Raises an exception if there is more than one queue matching the given
|
21
|
+
# name.
|
22
|
+
# If there is no queue with the given name, create one.
|
23
|
+
def initialize(queue_name)
|
24
|
+
@sqs = AwsContext.instance.sqs
|
25
|
+
queues = @sqs.list_queues(queue_name)
|
26
|
+
@queue_depth = 0
|
27
|
+
@queue = case queues.length
|
28
|
+
when 0
|
29
|
+
queue = @sqs.create_queue(queue_name)
|
30
|
+
raise "Bad Configuration -- no queue: #{queue_name}" if queue.nil?
|
31
|
+
queue
|
32
|
+
when 1
|
33
|
+
queues.first
|
34
|
+
else
|
35
|
+
raise "Bad configuration -- multiple queues match #{queue_name}"
|
36
|
+
end
|
37
|
+
end
|
38
|
+
|
39
|
+
# Read some messages off a queue.
|
40
|
+
# For some reason, we never receive more than 1.
|
41
|
+
# But we are prepared for more.
|
42
|
+
# return an array of the messages that were read.
|
43
|
+
def read_messages(count = 1)
|
44
|
+
@sqs.receive_messages(@queue, count)
|
45
|
+
end
|
46
|
+
|
47
|
+
# Delete a message from the queue given its id.
|
48
|
+
def delete_message(receipt_handle)
|
49
|
+
@sqs.delete_message(@queue, receipt_handle)
|
50
|
+
end
|
51
|
+
|
52
|
+
# Read and discard all messages on the queue.
|
53
|
+
# Return the number read.
|
54
|
+
def empty_queue
|
55
|
+
n = 0
|
56
|
+
while true
|
57
|
+
msgs = read_messages
|
58
|
+
break if msgs.size == 0
|
59
|
+
n += 1
|
60
|
+
msgs.each {|msg| delete_message(msg[:receipt_handle])}
|
61
|
+
end
|
62
|
+
n
|
63
|
+
end
|
64
|
+
|
65
|
+
# Get qeue depth and store it.
|
66
|
+
# In case of failure, keep the old value.
|
67
|
+
# The last value read is available through the queue_depth attribute.
|
68
|
+
def read_queue_depth
|
69
|
+
attr = 'ApproximateNumberOfMessages'
|
70
|
+
attrs = @sqs.get_queue_attributes(@queue, attr)
|
71
|
+
@queue_depth = attrs[attr] if attrs.has_key?(attr)
|
72
|
+
@queue_depth
|
73
|
+
end
|
74
|
+
end
|
75
|
+
end
|
data/app/policy.rb
ADDED
@@ -0,0 +1,113 @@
|
|
1
|
+
require 'policy_limit'
|
2
|
+
|
3
|
+
module Cloudmaster
|
4
|
+
|
5
|
+
# Provides the common data and behaviors for policies.
|
6
|
+
# This includes storing the configuration and instance collection.
|
7
|
+
# It also includes implementing essential methods such as ensure_limite and
|
8
|
+
# stop_hung_instances.
|
9
|
+
# Finally, this implements helpers that process the queue and image names
|
10
|
+
# which might be desirable to override.
|
11
|
+
class Policy
|
12
|
+
def initialize(reporter, config, instances)
|
13
|
+
@reporter = reporter
|
14
|
+
@config = config
|
15
|
+
@instances = instances
|
16
|
+
@limit_policy = PolicyLimit.new(reporter, config, instances)
|
17
|
+
end
|
18
|
+
|
19
|
+
# Make sure there are at least the minimum instances running.
|
20
|
+
# Also make sure there are no more than maximum number of instances.
|
21
|
+
# If not, start or stop enough to conform.
|
22
|
+
# These actions bypass the creation and termination limits normally in place
|
23
|
+
def ensure_limits
|
24
|
+
n = @limit_policy.adjust_limits
|
25
|
+
case
|
26
|
+
when n > 0
|
27
|
+
start_instances(n)
|
28
|
+
when n < 0
|
29
|
+
stop_instances(-n)
|
30
|
+
end
|
31
|
+
end
|
32
|
+
|
33
|
+
# If instances have not sent status in a long time, they
|
34
|
+
# are probably hung, and should be stopped.
|
35
|
+
def stop_hung_instances
|
36
|
+
if @instances.hung_instances.size > 0
|
37
|
+
@reporter.info("Stopping hung instances #{@instances.hung_instances.size}")
|
38
|
+
stop_instances(@instances.hung_instances)
|
39
|
+
end
|
40
|
+
end
|
41
|
+
|
42
|
+
# Default policy application function
|
43
|
+
# Calculate the number to adjust by, and then start or
|
44
|
+
# stop instances accordingly.
|
45
|
+
def apply
|
46
|
+
n = @limit_policy.adjust(adjust)
|
47
|
+
case
|
48
|
+
when n > 0
|
49
|
+
start_instances(n)
|
50
|
+
when n == 0
|
51
|
+
0
|
52
|
+
when n < 0
|
53
|
+
stop_instances(-n)
|
54
|
+
end
|
55
|
+
end
|
56
|
+
|
57
|
+
# Returns the number of additional instances to start (positive)
|
58
|
+
# or the number to stop (negative)
|
59
|
+
# This must be implemented by derived class.
|
60
|
+
def adjust
|
61
|
+
raise NotImpelemntedError("#{self.class.name}#adjust is not implemented.")
|
62
|
+
end
|
63
|
+
|
64
|
+
protected
|
65
|
+
|
66
|
+
# Start the given number of instances.
|
67
|
+
# Return the number started.
|
68
|
+
def start_instances(number_to_start)
|
69
|
+
started_instances = @instances.start_n_instances(number_to_start)
|
70
|
+
if started_instances.size < number_to_start
|
71
|
+
@reporter.error("Failed to start requested number of instances. (#{started_instances.size} instead of #{number_to_start})")
|
72
|
+
end
|
73
|
+
started_instances.each { |started| @reporter.info("Started instance #{started.id}")}
|
74
|
+
started_instances.size
|
75
|
+
end
|
76
|
+
|
77
|
+
# Stop the given list of instances.
|
78
|
+
# Don't stop ones whose minimum lifetime has not elapsed.
|
79
|
+
# Returns the number stopped.
|
80
|
+
def stop_instances_list(instances_to_stop, force = false)
|
81
|
+
instances_to_stop = instances_to_stop.find_all do |instance|
|
82
|
+
# Don't stop instances before minimum_time
|
83
|
+
force || instance.minimum_time_elapsed?
|
84
|
+
end
|
85
|
+
instances = @instances.stop_instances(instances_to_stop)
|
86
|
+
instances.each {|i| @reporter.info("Terminating instance ", i.id) }
|
87
|
+
instances.size
|
88
|
+
end
|
89
|
+
|
90
|
+
# Stop a given number of instances.
|
91
|
+
# Return the number stopped.
|
92
|
+
def stop_n_instances(number_to_stop)
|
93
|
+
return if number_to_stop <= 0
|
94
|
+
# stop the instances with the lowest load estimate
|
95
|
+
instances_with_lowest_load = @instances.sorted_by_lowest_load
|
96
|
+
stop_instances_list(instances_with_lowest_load[0...number_to_stop])
|
97
|
+
end
|
98
|
+
|
99
|
+
# Stop instances.
|
100
|
+
# If passed a number, it stops that number of instances.
|
101
|
+
# If passed a list of instance ids, it stops those instances.
|
102
|
+
def stop_instances(to_stop, *params)
|
103
|
+
if to_stop.class == Fixnum
|
104
|
+
stop_n_instances(to_stop)
|
105
|
+
elsif to_stop.class == Array
|
106
|
+
stop_instances_list(to_stop, *params)
|
107
|
+
else
|
108
|
+
raise "Bad call -- stop_instances #{to_stop.class}"
|
109
|
+
end
|
110
|
+
end
|
111
|
+
end
|
112
|
+
|
113
|
+
end
|
@@ -0,0 +1,18 @@
|
|
1
|
+
require 'policy'
|
2
|
+
|
3
|
+
module Cloudmaster
|
4
|
+
|
5
|
+
# Provide example daytime policy
|
6
|
+
# This increases the pool size during the day.
|
7
|
+
# This is provided as part of an example on how to create custom policies.
|
8
|
+
class PolicyDaytime < Policy
|
9
|
+
# Define the adjust function increase the pool in the
|
10
|
+
# daytime if it is below daytime minimum
|
11
|
+
def adjust
|
12
|
+
hour = Clock.hour
|
13
|
+
return 0 if hour < 10 || hour > 17
|
14
|
+
return 0 if @instances.size >= @config[:minimum_number_of_instances_daytime].to_i
|
15
|
+
return 1
|
16
|
+
end
|
17
|
+
end
|
18
|
+
end
|
@@ -0,0 +1,16 @@
|
|
1
|
+
# This is a factory for Policy implementations.
|
2
|
+
require 'factory'
|
3
|
+
|
4
|
+
module Cloudmaster
|
5
|
+
class PolicyFactory
|
6
|
+
include Factory
|
7
|
+
def PolicyFactory.create(policy_name, *params)
|
8
|
+
name = policy_name.nil? ? 'default' : policy_name.to_s
|
9
|
+
require 'policy_' + name.downcase
|
10
|
+
class_name = 'Policy' +name.capitalize
|
11
|
+
policy = Factory.create_object_from_string(class_name, *params)
|
12
|
+
raise "Bad configuration -- no policy #{class_name}" unless policy
|
13
|
+
policy
|
14
|
+
end
|
15
|
+
end
|
16
|
+
end
|
data/app/policy_fixed.rb
ADDED
@@ -0,0 +1,19 @@
|
|
1
|
+
require 'policy'
|
2
|
+
|
3
|
+
module Cloudmaster
|
4
|
+
|
5
|
+
# Provide fixed policy.
|
6
|
+
# This never adjusts the size of the pool.
|
7
|
+
# PolicyFixed is still useful, because it still ensures that the number
|
8
|
+
# of instances stay beteen the maximum and minimum (because this is
|
9
|
+
# enforced in the base class apply).
|
10
|
+
class PolicyFixed < Policy
|
11
|
+
# Use everything from the base class.
|
12
|
+
|
13
|
+
# Adjust never changes instances.
|
14
|
+
def adjust
|
15
|
+
0
|
16
|
+
end
|
17
|
+
end
|
18
|
+
|
19
|
+
end
|
data/app/policy_job.rb
ADDED
@@ -0,0 +1,54 @@
|
|
1
|
+
|
2
|
+
require 'policy'
|
3
|
+
|
4
|
+
module Cloudmaster
|
5
|
+
|
6
|
+
# Provides job policy
|
7
|
+
# Instances managed by the job policy take work from a queue and report
|
8
|
+
# status thrugh another queue.
|
9
|
+
class PolicyJob < Policy
|
10
|
+
# Initialize the policy by giving it access to the configuration and
|
11
|
+
# the collection of instances.
|
12
|
+
def initialize(reporter, config, instances)
|
13
|
+
super(reporter, config, instances)
|
14
|
+
@config = config
|
15
|
+
end
|
16
|
+
|
17
|
+
# Adjust the pool size.
|
18
|
+
# Add instances if the queue is getting backed up.
|
19
|
+
# Delete instances if the queue is empty and instances are idle.
|
20
|
+
def adjust
|
21
|
+
depth = @config[:work_queue].queue_depth
|
22
|
+
|
23
|
+
# See if we should add instances
|
24
|
+
if (depth > 0 && @instances.size == 0) || depth >= @config[:start_threshold].to_i
|
25
|
+
additional = 0
|
26
|
+
# need this many more to service the queued work
|
27
|
+
needed = (depth.to_f / @config[:start_threshold].to_f).floor
|
28
|
+
needed = 1 if needed < 1
|
29
|
+
if needed > @instances.size
|
30
|
+
additional = needed - @instances.size
|
31
|
+
end
|
32
|
+
@reporter.info("Job policy need additional #{additional} depth: #{depth} needed #{needed}")
|
33
|
+
return additional
|
34
|
+
end
|
35
|
+
|
36
|
+
# Queue not empty -- don't stop any
|
37
|
+
return 0 if depth > 0
|
38
|
+
|
39
|
+
# See if we should stop some
|
40
|
+
# count how many are active idle and active
|
41
|
+
idle = @instances.active_idle_instances
|
42
|
+
active = @instances.active_instances
|
43
|
+
if idle.size > 0
|
44
|
+
# stop some fraction of the idle instances
|
45
|
+
excess = (idle.size / @config[:idle_threshold].to_f).round
|
46
|
+
@reporter.info("Job policy need fewer #{excess} idle: #{idle.size} active #{active.size}")
|
47
|
+
return -excess
|
48
|
+
end
|
49
|
+
0
|
50
|
+
end
|
51
|
+
|
52
|
+
# uses the default apply
|
53
|
+
end
|
54
|
+
end
|
data/app/policy_limit.rb
ADDED
@@ -0,0 +1,68 @@
|
|
1
|
+
module Cloudmaster
|
2
|
+
|
3
|
+
# This enforces the limits on starting and stopping instances.
|
4
|
+
# It can be used by other policies to make sure thrie own adjust method
|
5
|
+
# does not increase or decrease by more than the configurable start limit or stop limit.
|
6
|
+
# This also provides an exaplanation for its actions.
|
7
|
+
class PolicyLimit
|
8
|
+
def initialize(reporter, config, instances)
|
9
|
+
@reporter = reporter
|
10
|
+
@config = config
|
11
|
+
@instances = instances
|
12
|
+
end
|
13
|
+
|
14
|
+
def max2(a, b)
|
15
|
+
a > b ? a : b
|
16
|
+
end
|
17
|
+
|
18
|
+
# Make sure there are at least the minimum instances running.
|
19
|
+
# Also make sure there are no more than maximum number of instances.
|
20
|
+
# Return the number to start (positive) or stop (negative) to stay within the limits.
|
21
|
+
# Do not enforce start_limit and stop_limit.
|
22
|
+
def adjust_limits
|
23
|
+
if @instances.less_than_minimum?
|
24
|
+
number_to_start = @instances.below_minimum_count
|
25
|
+
@reporter.info("Less than minimum -- start more #{number_to_start}")
|
26
|
+
number_to_start
|
27
|
+
elsif @instances.greater_than_maximum?
|
28
|
+
number_to_stop = @instances.above_maximum_count
|
29
|
+
@reporter.info("Greater than maximum -- stop some #{number_to_stop}")
|
30
|
+
-number_to_stop
|
31
|
+
else
|
32
|
+
0
|
33
|
+
end
|
34
|
+
end
|
35
|
+
|
36
|
+
# After other policies have computed a value for adjust, then this one possibly
|
37
|
+
# modifies the value by ensuring that the start_limit and stop_limit constraints are
|
38
|
+
# honored. It also makes sure that the adjustment makes the instance count
|
39
|
+
# stay between the minimum and maximum.
|
40
|
+
def adjust(n)
|
41
|
+
case
|
42
|
+
when n > 0
|
43
|
+
start_limit = @config[:start_limit].to_i
|
44
|
+
if n > start_limit
|
45
|
+
@reporter.info("Limit start -- requested: #{n} limit: #{start_limit}")
|
46
|
+
n = start_limit
|
47
|
+
end
|
48
|
+
remaining = max2(@instances.maximum - @instances.size, 0)
|
49
|
+
if n > remaining
|
50
|
+
@reporter.info("Limit start -- requested: #{n} remaining: #{remaining}")
|
51
|
+
n = remaining
|
52
|
+
end
|
53
|
+
when n < 0
|
54
|
+
stop_limit = @config[:stop_limit].to_i
|
55
|
+
if -n > stop_limit
|
56
|
+
@reporter.info("Limit stop -- requested: #{-n} limit: #{stop_limit}")
|
57
|
+
n = -stop_limit
|
58
|
+
end
|
59
|
+
remaining = max2(@instances.size - @instances.minimum, 0)
|
60
|
+
if -n > remaining
|
61
|
+
@reporter.info("Limit stop -- requested: #{n} remaining: #{remaining}")
|
62
|
+
n = -remaining
|
63
|
+
end
|
64
|
+
end
|
65
|
+
n
|
66
|
+
end
|
67
|
+
end
|
68
|
+
end
|