tengine_job 0.6.9
Sign up to get free protection for your applications and to get access to all the features.
- data/.document +5 -0
- data/.rspec +1 -0
- data/Gemfile +23 -0
- data/Gemfile.lock +109 -0
- data/README.rdoc +20 -0
- data/Rakefile +42 -0
- data/VERSION +1 -0
- data/examples/0004_retry_one_layer.rb +24 -0
- data/examples/0004_retry_one_layer.sh +38 -0
- data/examples/0005_retry_two_layer.rb +54 -0
- data/examples/0005_retry_two_layer.sh +80 -0
- data/examples/0006_retry_three_layer.rb +58 -0
- data/examples/0006_retry_three_layer.sh +74 -0
- data/examples/0007_simple_jobnet.rb +7 -0
- data/examples/0021_dynamic_env.rb +20 -0
- data/examples/VERSION +1 -0
- data/examples/tengine_job_test.sh +10 -0
- data/lib/tengine/job.rb +94 -0
- data/lib/tengine/job/category.rb +54 -0
- data/lib/tengine/job/connectable.rb +43 -0
- data/lib/tengine/job/drivers/job_control_driver.rb +82 -0
- data/lib/tengine/job/drivers/job_execution_driver.rb +30 -0
- data/lib/tengine/job/drivers/jobnet_control_driver.rb +117 -0
- data/lib/tengine/job/drivers/schedule_driver.rb +30 -0
- data/lib/tengine/job/dsl_binder.rb +12 -0
- data/lib/tengine/job/dsl_evaluator.rb +18 -0
- data/lib/tengine/job/dsl_loader.rb +180 -0
- data/lib/tengine/job/edge.rb +150 -0
- data/lib/tengine/job/element_selector_notation.rb +169 -0
- data/lib/tengine/job/end.rb +32 -0
- data/lib/tengine/job/executable.rb +74 -0
- data/lib/tengine/job/execution.rb +141 -0
- data/lib/tengine/job/expansion.rb +37 -0
- data/lib/tengine/job/fork.rb +6 -0
- data/lib/tengine/job/job.rb +23 -0
- data/lib/tengine/job/jobnet.rb +173 -0
- data/lib/tengine/job/jobnet/builder.rb +150 -0
- data/lib/tengine/job/jobnet/job_state_transition.rb +167 -0
- data/lib/tengine/job/jobnet/jobnet_state_transition.rb +110 -0
- data/lib/tengine/job/jobnet/state_transition.rb +37 -0
- data/lib/tengine/job/jobnet_actual.rb +55 -0
- data/lib/tengine/job/jobnet_template.rb +10 -0
- data/lib/tengine/job/join.rb +6 -0
- data/lib/tengine/job/junction.rb +29 -0
- data/lib/tengine/job/killing.rb +30 -0
- data/lib/tengine/job/mm_compatibility.rb +6 -0
- data/lib/tengine/job/mm_compatibility/connectable.rb +13 -0
- data/lib/tengine/job/name_path.rb +31 -0
- data/lib/tengine/job/root.rb +16 -0
- data/lib/tengine/job/root_jobnet_actual.rb +39 -0
- data/lib/tengine/job/root_jobnet_template.rb +49 -0
- data/lib/tengine/job/script_executable.rb +235 -0
- data/lib/tengine/job/signal.rb +121 -0
- data/lib/tengine/job/start.rb +20 -0
- data/lib/tengine/job/stoppable.rb +15 -0
- data/lib/tengine/job/vertex.rb +172 -0
- data/lib/tengine_job.rb +3 -0
- data/spec/fixtures/rjn_0001_simple_jobnet_builder.rb +42 -0
- data/spec/fixtures/rjn_0002_simple_parallel_jobnet_builder.rb +42 -0
- data/spec/fixtures/rjn_0003_fork_join_jobnet_builder.rb +61 -0
- data/spec/fixtures/rjn_0004_parallel_jobnet_with_finally_fixture.rb +62 -0
- data/spec/fixtures/rjn_0005_retry_two_layer_fixture.rb +153 -0
- data/spec/fixtures/rjn_0008_expansion_fixture.rb +32 -0
- data/spec/fixtures/rjn_0009_tree_sequential_jobnet_builder.rb +174 -0
- data/spec/fixtures/rjn_0010_2jobs_and_1job_parallel_jobnet_builder.rb +39 -0
- data/spec/fixtures/rjn_0011_nested_fork_jobnet_builder.rb +96 -0
- data/spec/fixtures/rjn_0012_nested_and_finally_builder.rb +157 -0
- data/spec/fixtures/rjn_1004_hadoop_job_in_jobnet_fixture.rb +105 -0
- data/spec/fixtures/rjn_means_root_jobnet +0 -0
- data/spec/fixtures/test_credential_fixture.rb +12 -0
- data/spec/fixtures/test_server_fixture.rb +28 -0
- data/spec/mongoid.yml +35 -0
- data/spec/spec_helper.rb +56 -0
- data/spec/sshd/.gitignore +1 -0
- data/spec/sshd/id_rsa +51 -0
- data/spec/sshd/id_rsa.pub +1 -0
- data/spec/sshd/ssh_host_rsa_key +51 -0
- data/spec/sshd/ssh_host_rsa_key.pub +1 -0
- data/spec/sshd/sshd_config +10 -0
- data/spec/sshd/sshd_config.erb +11 -0
- data/spec/sshd/tengine_job_test.sh +6 -0
- data/spec/support/jobnet_fixture_builder.rb +145 -0
- data/spec/support/mongo_index_key_log.rb +91 -0
- data/spec/tengine/job/category_spec.rb +193 -0
- data/spec/tengine/job/connectable_spec.rb +94 -0
- data/spec/tengine/job/drivers/job_controll_driver/connection_error_spec.rb +236 -0
- data/spec/tengine/job/drivers/job_controll_driver/duplicated_job_start_spec.rb +302 -0
- data/spec/tengine/job/drivers/job_controll_driver/expansion_spec.rb +120 -0
- data/spec/tengine/job/drivers/job_controll_driver/stop_spec.rb +159 -0
- data/spec/tengine/job/drivers/job_controll_driver_spec.rb +623 -0
- data/spec/tengine/job/drivers/job_execution_driver_spec.rb +88 -0
- data/spec/tengine/job/drivers/jobnet_control_driver/nested_and_finally_spec.rb +472 -0
- data/spec/tengine/job/drivers/jobnet_control_driver/nested_jobnet_spec.rb +231 -0
- data/spec/tengine/job/drivers/jobnet_control_driver/stop_jobnet_spec.rb +202 -0
- data/spec/tengine/job/drivers/jobnet_control_driver_spec.rb +446 -0
- data/spec/tengine/job/drivers/schedule_driver_spec.rb +202 -0
- data/spec/tengine/job/dsl_binder_spec.rb +36 -0
- data/spec/tengine/job/dsl_loader_spec.rb +403 -0
- data/spec/tengine/job/dsls/0013_hadoop_job_run.rb +29 -0
- data/spec/tengine/job/dsls/0014_join_and_join.rb +19 -0
- data/spec/tengine/job/dsls/0015_fork_and_fork.rb +18 -0
- data/spec/tengine/job/dsls/0016_complex_fork_and_join.rb +20 -0
- data/spec/tengine/job/dsls/0017_finally.rb +15 -0
- data/spec/tengine/job/dsls/0018_expansion.rb +23 -0
- data/spec/tengine/job/dsls/0019_execute_job_on_event.rb +16 -0
- data/spec/tengine/job/dsls/0020_duplicated_jobnet_name.rb +16 -0
- data/spec/tengine/job/dsls/1060_test_dir1/1060_test_dir2/0013_hadoop_job_run.rb +29 -0
- data/spec/tengine/job/dsls/2003_expansion/expansion_5.rb +11 -0
- data/spec/tengine/job/dsls/VERSION +1 -0
- data/spec/tengine/job/dynamic_env_spec.rb +95 -0
- data/spec/tengine/job/edge_spec.rb +241 -0
- data/spec/tengine/job/element_selector_notation_spec.rb +354 -0
- data/spec/tengine/job/examples_spec.rb +62 -0
- data/spec/tengine/job/execution_spec.rb +100 -0
- data/spec/tengine/job/expansion_spec.rb +116 -0
- data/spec/tengine/job/hadoop_job_run_spec.rb +65 -0
- data/spec/tengine/job/job_spec.rb +4 -0
- data/spec/tengine/job/jobnet/1015_complecated_jobnet_spec.rb +72 -0
- data/spec/tengine/job/jobnet_actual_spec.rb +175 -0
- data/spec/tengine/job/jobnet_spec.rb +399 -0
- data/spec/tengine/job/jobnet_template_spec.rb +240 -0
- data/spec/tengine/job/killing_spec.rb +91 -0
- data/spec/tengine/job/reset_spec.rb +958 -0
- data/spec/tengine/job/reset_spec/4056_1_dump.txt +1 -0
- data/spec/tengine/job/root_jobnet_actual_spec.rb +89 -0
- data/spec/tengine/job/root_jobnet_template_spec.rb +248 -0
- data/spec/tengine/job/script_executable_spec.rb +132 -0
- data/spec/tengine/job/stoppable_spec.rb +176 -0
- data/spec/tengine/job/vertex_spec.rb +25 -0
- data/spec/tengine_job_spec.rb +4 -0
- data/tengine_job.gemspec +197 -0
- data/tmp/log/.gitignore +1 -0
- metadata +296 -0
@@ -0,0 +1,120 @@
|
|
1
|
+
# -*- coding: utf-8 -*-
|
2
|
+
require 'spec_helper'
|
3
|
+
require 'tengine/rspec'
|
4
|
+
|
5
|
+
describe 'job_control_driver' do
|
6
|
+
include Tengine::RSpec::Extension
|
7
|
+
|
8
|
+
target_dsl File.expand_path("../../../../../lib/tengine/job/drivers/job_control_driver.rb", File.dirname(__FILE__))
|
9
|
+
driver :job_control_driver
|
10
|
+
|
11
|
+
shared_examples_for "/rjn0008/rjn0001/j11を実行する際の環境変数" do |dsl_version|
|
12
|
+
it "expansionだったジョブネットよりも上位のジョブの情報は出力されない" do
|
13
|
+
@rjn0001 = @root.vertex_by_name_path("/rjn0008/rjn0001")
|
14
|
+
@j11 = @root.vertex_by_name_path("/rjn0008/rjn0001/j11")
|
15
|
+
@root.phase_key = :running
|
16
|
+
@rjn0001.phase_key = :running
|
17
|
+
@j11.phase_key = :ready
|
18
|
+
@j11.prev_edges.each{|edge| edge.phase_key = :transmitting}
|
19
|
+
@root.save!
|
20
|
+
@root.reload
|
21
|
+
tengine.should_not_fire
|
22
|
+
mock_ssh = mock(:ssh)
|
23
|
+
mock_channel = mock(:channel)
|
24
|
+
Net::SSH.should_receive(:start).
|
25
|
+
with("localhost", an_instance_of(Tengine::Resource::Credential), an_instance_of(Hash)).and_yield(mock_ssh)
|
26
|
+
mock_ssh.should_receive(:open_channel).and_yield(mock_channel)
|
27
|
+
mock_channel.should_receive(:exec) do |*args|
|
28
|
+
args.length.should == 1
|
29
|
+
# args.first.should =~ %r<source \/etc\/profile && export MM_ACTUAL_JOB_ID=[0-9a-f]{24} MM_ACTUAL_JOB_ANCESTOR_IDS=\\"[0-9a-f]{24}\\" MM_FULL_ACTUAL_JOB_ANCESTOR_IDS=\\"[0-9a-f]{24}\\" MM_ACTUAL_JOB_NAME_PATH=\\"/rjn0001/j11\\" MM_ACTUAL_JOB_SECURITY_TOKEN= MM_SCHEDULE_ID=[0-9a-f]{24} MM_SCHEDULE_ESTIMATED_TIME= MM_TEMPLATE_JOB_ID=[0-9a-f]{24} MM_TEMPLATE_JOB_ANCESTOR_IDS=\\"[0-9a-f]{24}\\" && tengine_job_agent_run -- \$HOME/j11\.sh>
|
30
|
+
args.first.should =~ %r<source \/etc\/profile>
|
31
|
+
t_rjn1001 = Tengine::Job::RootJobnetTemplate.find_by_name("rjn0001")
|
32
|
+
t_rjn1001.dsl_version.should == dsl_version
|
33
|
+
t_j11 = t_rjn1001.vertex_by_name_path("/rjn0001/j11")
|
34
|
+
args.first.should =~ %r<MM_TEMPLATE_JOB_ID=#{t_j11.id.to_s}>
|
35
|
+
args.first.should_not =~ %r<MM_TEMPLATE_JOB_ANCESTOR_IDS=\"#{@template.id.to_s};#{t_rjn1001.id.to_s}\">
|
36
|
+
args.first.should =~ %r<MM_TEMPLATE_JOB_ANCESTOR_IDS=\"#{t_rjn1001.id.to_s}\">
|
37
|
+
args.first.should =~ %r<job_test j11>
|
38
|
+
end
|
39
|
+
tengine.receive("start.job.job.tengine", :properties => {
|
40
|
+
:execution_id => @execution.id.to_s,
|
41
|
+
:root_jobnet_id => @root.id.to_s,
|
42
|
+
:target_jobnet_id => @rjn0001.id.to_s,
|
43
|
+
:target_job_id => @j11.id.to_s,
|
44
|
+
})
|
45
|
+
@root.reload
|
46
|
+
@rjn0001 = @root.vertex_by_name_path("/rjn0008/rjn0001")
|
47
|
+
@j11 = @root.vertex_by_name_path("/rjn0008/rjn0001/j11")
|
48
|
+
@root.phase_key = :running
|
49
|
+
@rjn0001.phase_key = :running
|
50
|
+
end
|
51
|
+
end
|
52
|
+
|
53
|
+
# in [rjn0008]
|
54
|
+
# (S1) --e1-->(rjn0001)--e2-->(rjn0002)--e3-->(E1)
|
55
|
+
#
|
56
|
+
# in [rjn0001]
|
57
|
+
# (S1) --e1-->(j11)--e2-->(j12)--e3-->(E1)
|
58
|
+
#
|
59
|
+
# in [rjn0002]
|
60
|
+
# |--e2-->(j11)--e4-->|
|
61
|
+
# (S1)--e1-->[F1] [J1]--e6-->(E1)
|
62
|
+
# |--e3-->(j12)--e5-->|
|
63
|
+
context "rjn0008" do
|
64
|
+
before do
|
65
|
+
Tengine::Core::Setting.delete_all
|
66
|
+
Tengine::Core::Setting.create!(:name => "dsl_version", :value => "1")
|
67
|
+
Tengine::Job::Vertex.delete_all
|
68
|
+
Rjn0001SimpleJobnetBuilder.new.create_template
|
69
|
+
Rjn0002SimpleParallelJobnetBuilder.new.create_template
|
70
|
+
builder = Rjn0008ExpansionFixture.new
|
71
|
+
@template = builder.create_template
|
72
|
+
@root = @template.generate
|
73
|
+
@ctx = builder.context
|
74
|
+
@execution = Tengine::Job::Execution.create!({
|
75
|
+
:root_jobnet_id => @root.id,
|
76
|
+
})
|
77
|
+
@base_props = {
|
78
|
+
:execution_id => @execution.id.to_s,
|
79
|
+
:root_jobnet_id => @root.id.to_s,
|
80
|
+
:target_jobnet_id => @root.id.to_s,
|
81
|
+
}
|
82
|
+
end
|
83
|
+
|
84
|
+
it_should_behave_like "/rjn0008/rjn0001/j11を実行する際の環境変数", "1"
|
85
|
+
end
|
86
|
+
|
87
|
+
context "複数のバージョンのデータがある場合" do
|
88
|
+
before do
|
89
|
+
Tengine::Core::Setting.delete_all
|
90
|
+
Tengine::Core::Setting.create!(:name => "dsl_version", :value => "2")
|
91
|
+
Tengine::Job::Vertex.delete_all
|
92
|
+
Rjn0001SimpleJobnetBuilder.new.tap do |f|
|
93
|
+
f.create_template(:dsl_version => "1")
|
94
|
+
f.create_template(:dsl_version => "2")
|
95
|
+
end
|
96
|
+
Rjn0002SimpleParallelJobnetBuilder.new.tap do |f|
|
97
|
+
f.create_template(:dsl_version => "1")
|
98
|
+
f.create_template(:dsl_version => "2")
|
99
|
+
end
|
100
|
+
builder = Rjn0008ExpansionFixture.new
|
101
|
+
builder.create_template(:dsl_version => "1")
|
102
|
+
@template = builder.create_template(:dsl_version => "2")
|
103
|
+
@root = @template.generate
|
104
|
+
@ctx = builder.context
|
105
|
+
@execution = Tengine::Job::Execution.create!({
|
106
|
+
:root_jobnet_id => @root.id,
|
107
|
+
})
|
108
|
+
@base_props = {
|
109
|
+
:execution_id => @execution.id.to_s,
|
110
|
+
:root_jobnet_id => @root.id.to_s,
|
111
|
+
:target_jobnet_id => @root.id.to_s,
|
112
|
+
}
|
113
|
+
end
|
114
|
+
|
115
|
+
it_should_behave_like "/rjn0008/rjn0001/j11を実行する際の環境変数", '2'
|
116
|
+
end
|
117
|
+
|
118
|
+
end
|
119
|
+
|
120
|
+
|
@@ -0,0 +1,159 @@
|
|
1
|
+
# -*- coding: utf-8 -*-
|
2
|
+
require 'spec_helper'
|
3
|
+
require 'tengine/rspec'
|
4
|
+
|
5
|
+
require 'net/ssh'
|
6
|
+
|
7
|
+
describe "<BUG>(tengined複数起動)強制停止すると、ステータスが「強制停止済」ではなく「エラー終了」になる" do
|
8
|
+
include Tengine::RSpec::Extension
|
9
|
+
|
10
|
+
driver_path = File.expand_path("../../../../../lib/tengine/job/drivers/job_control_driver.rb", File.dirname(__FILE__))
|
11
|
+
|
12
|
+
#
|
13
|
+
# in [jn0004]
|
14
|
+
# |--e3-->(j2)--e5-->|
|
15
|
+
# (S1)--e1-->(j1)--e2-->[F1] [J1]--e7-->(j4)--e8-->(E1)
|
16
|
+
# |--e4-->(j3)--e6-->|
|
17
|
+
#
|
18
|
+
# in [jn0004/finally]
|
19
|
+
# (S2) --e9-->(jn0004_f)-e10-->(E2)
|
20
|
+
#
|
21
|
+
# 現象:
|
22
|
+
# j1を強制停止した際に、プロセスが2つ動いている場合、その片方のプロセスAが
|
23
|
+
# stop.job.job.tengineイベントを受け取りSSHでtengine_job_agent_killを実行します。
|
24
|
+
# その実行の戻り値を得るまでの間に、もう片方のプロセスBが、プロセスが終了して発火される
|
25
|
+
# finished.process.job.tengineイベントを処理すると、stop_reasonがuser_stopでなくなってしまっていました。
|
26
|
+
#
|
27
|
+
context "jn0004" do
|
28
|
+
before do
|
29
|
+
Tengine::Resource::Server.delete_all
|
30
|
+
Tengine::Job::Execution.delete_all
|
31
|
+
Tengine::Job::Vertex.delete_all
|
32
|
+
TestCredentialFixture.test_credential1
|
33
|
+
TestServerFixture.test_server1
|
34
|
+
TestServerFixture.test_server2
|
35
|
+
builder = Rjn0004ParallelJobnetWithFinally.new
|
36
|
+
@root = builder.create_actual
|
37
|
+
@ctx = builder.context
|
38
|
+
@execution = Tengine::Job::Execution.create!({
|
39
|
+
:root_jobnet_id => @root.id,
|
40
|
+
})
|
41
|
+
@base_props = {
|
42
|
+
:execution_id => @execution.id.to_s,
|
43
|
+
:root_jobnet_id => @root.id.to_s,
|
44
|
+
:root_jobnet_name_path => @root.name_path,
|
45
|
+
:target_jobnet_id => @root.id.to_s,
|
46
|
+
:target_jobnet_name_path => @root.name_path,
|
47
|
+
}
|
48
|
+
|
49
|
+
# 2つのプロセスの代わりに、2つのカーネルを別のFiberで動かす
|
50
|
+
@bootstrap1 = Tengine::Core::Bootstrap.new(:tengined => { :load_path => driver_path })
|
51
|
+
@bootstrap1.kernel.tap{|k| k.bind; k.evaluate}
|
52
|
+
@tengine1 = Tengine::RSpec::ContextWrapper.new(@bootstrap1.kernel)
|
53
|
+
#
|
54
|
+
@bootstrap2 = Tengine::Core::Bootstrap.new(:tengined => { :load_path => driver_path })
|
55
|
+
@bootstrap2.kernel.tap{|k| k.bind; k.evaluate}
|
56
|
+
@tengine2 = Tengine::RSpec::ContextWrapper.new(@bootstrap2.kernel)
|
57
|
+
end
|
58
|
+
|
59
|
+
before do
|
60
|
+
@pid = "123"
|
61
|
+
@ctx[:e1].phase_key = :transmitted
|
62
|
+
@ctx[:j1].tap do |j|
|
63
|
+
j.phase_key = :running
|
64
|
+
j.executing_pid = @pid
|
65
|
+
end
|
66
|
+
@root.phase_key = :running
|
67
|
+
@root.version = 4
|
68
|
+
@root.save!
|
69
|
+
|
70
|
+
@f1 = Fiber.new do
|
71
|
+
@tengine1.should_not_fire
|
72
|
+
ssh1 = mock(:ssh1)
|
73
|
+
Net::SSH.should_receive(:start).with("localhost",
|
74
|
+
an_instance_of(Tengine::Resource::Credential),
|
75
|
+
an_instance_of(Hash)).once.and_yield(ssh1)
|
76
|
+
channel1 = mock(:channel1)
|
77
|
+
ssh1.stub(:open_channel).and_yield(channel1)
|
78
|
+
channel1.stub(:exec).with(any_args).and_yield(channel1, true)
|
79
|
+
channel1.should_receive(:on_close) do
|
80
|
+
Tengine.logger.debug( ("!" * 100) << "\non_close: Fiber.yield #{Process.pid} #{__FILE__}##{__LINE__}")
|
81
|
+
Fiber.yield
|
82
|
+
end # on_dataが呼び出される前に止める
|
83
|
+
channel1.stub(:on_data)
|
84
|
+
channel1.stub(:on_extended_data)
|
85
|
+
@tengine1.receive("stop.job.job.tengine", :properties => {
|
86
|
+
:stop_reason => "user_stop",
|
87
|
+
:target_job_id => @ctx.vertex(:j1).id.to_s,
|
88
|
+
:target_job_name_path => @ctx.vertex(:j1).name_path,
|
89
|
+
}.update(@base_props))
|
90
|
+
:end
|
91
|
+
end
|
92
|
+
|
93
|
+
@f2 = Fiber.new do
|
94
|
+
@tengine2.should_fire(:"error.job.job.tengine", {
|
95
|
+
:source_name => @ctx[:j1].name_as_resource,
|
96
|
+
:properties=>{
|
97
|
+
:target_job_id => @ctx.vertex(:j1).id.to_s,
|
98
|
+
:target_job_name_path => @ctx.vertex(:j1).name_path,
|
99
|
+
:exit_status=>nil,
|
100
|
+
}.update(@base_props)
|
101
|
+
})
|
102
|
+
@tengine2.receive("finished.process.job.tengine", :properties => {
|
103
|
+
:pid=>17485,
|
104
|
+
:exit_status=>nil,
|
105
|
+
:target_job_id => @ctx.vertex(:j1).id.to_s,
|
106
|
+
:target_job_name_path => @ctx.vertex(:j1).name_path,
|
107
|
+
}.update(@base_props))
|
108
|
+
:end
|
109
|
+
end
|
110
|
+
|
111
|
+
@j1 = @root.element("j1")
|
112
|
+
|
113
|
+
@root.reload
|
114
|
+
@root.version.should == 4
|
115
|
+
Tengine::Job.test_harness_clear
|
116
|
+
end
|
117
|
+
|
118
|
+
it "tengine_job_agent_killの戻り値の前にfinished.process.job.tengineが来ても強制終了となるべき" do
|
119
|
+
# f1-1.
|
120
|
+
Tengine.logger.info("1" * 100)
|
121
|
+
# Tengine::Job.should_receive(:test_harness).with(1, "before yield in update_with_lock").once
|
122
|
+
# Tengine::Job.should_receive(:test_harness).with(2, "after yield in update_with_lock").once{ Fiber.yield}
|
123
|
+
@f1.resume.should_not == :end
|
124
|
+
@root.reload
|
125
|
+
@root.version.should == 5
|
126
|
+
@root.element("j1").tap do |j|
|
127
|
+
j.phase_key.should == :dying
|
128
|
+
j.executing_pid.should == @pid
|
129
|
+
j.stop_reason.should == "user_stop"
|
130
|
+
end
|
131
|
+
|
132
|
+
# f2
|
133
|
+
Tengine.logger.info("2" * 100)
|
134
|
+
# Tengine::Job.should_receive(:test_harness).with(3, "before yield in update_with_lock").once
|
135
|
+
# Tengine::Job.should_receive(:test_harness).with(4, "after yield in update_with_lock").once{ Fiber.yield}
|
136
|
+
@f2.resume.should == :end
|
137
|
+
@root.reload
|
138
|
+
@root.version.should == 6
|
139
|
+
@root.element("j1").tap do |j|
|
140
|
+
j.phase_key.should == :error
|
141
|
+
j.executing_pid.should == @pid
|
142
|
+
j.stop_reason.should == "user_stop"
|
143
|
+
end
|
144
|
+
|
145
|
+
# f1-2.
|
146
|
+
Tengine.logger.info("3" * 100)
|
147
|
+
# Tengine::Job.should_receive(:test_harness).with(5, "after update_with_lock").once{ Fiber.yield}
|
148
|
+
@f1.resume.should == :end
|
149
|
+
@root.reload
|
150
|
+
@root.version.should == 6
|
151
|
+
@root.element("j1").tap do |j|
|
152
|
+
j.phase_key.should == :error
|
153
|
+
j.executing_pid.should == @pid
|
154
|
+
j.stop_reason.should == "user_stop"
|
155
|
+
end
|
156
|
+
end
|
157
|
+
|
158
|
+
end
|
159
|
+
end
|
@@ -0,0 +1,623 @@
|
|
1
|
+
# -*- coding: utf-8 -*-
|
2
|
+
require 'spec_helper'
|
3
|
+
require 'tengine/rspec'
|
4
|
+
|
5
|
+
require 'net/ssh'
|
6
|
+
|
7
|
+
describe 'job_control_driver' do
|
8
|
+
include Tengine::RSpec::Extension
|
9
|
+
|
10
|
+
target_dsl File.expand_path("../../../../lib/tengine/job/drivers/job_control_driver.rb", File.dirname(__FILE__))
|
11
|
+
driver :job_control_driver
|
12
|
+
|
13
|
+
context "rjn0001" do
|
14
|
+
before do
|
15
|
+
Tengine::Job::Vertex.delete_all
|
16
|
+
builder = Rjn0001SimpleJobnetBuilder.new
|
17
|
+
@jobnet = builder.create_actual
|
18
|
+
@ctx = builder.context
|
19
|
+
@execution = Tengine::Job::Execution.create!({
|
20
|
+
:root_jobnet_id => @jobnet.id,
|
21
|
+
})
|
22
|
+
end
|
23
|
+
|
24
|
+
context "ジョブの起動イベントを受け取ったら" do
|
25
|
+
it "通常の場合" do
|
26
|
+
@jobnet.phase_key = :starting
|
27
|
+
@ctx.edge(:e1).phase_key = :transmitting
|
28
|
+
@ctx.vertex(:j11).phase_key = :ready
|
29
|
+
@jobnet.save!
|
30
|
+
@jobnet.reload
|
31
|
+
tengine.should_not_fire
|
32
|
+
mock_ssh = mock(:ssh)
|
33
|
+
mock_channel = mock(:channel)
|
34
|
+
Net::SSH.should_receive(:start).
|
35
|
+
with("localhost", an_instance_of(Tengine::Resource::Credential), an_instance_of(Hash)).and_yield(mock_ssh)
|
36
|
+
mock_ssh.should_receive(:open_channel).and_yield(mock_channel)
|
37
|
+
mock_channel.should_receive(:exec) do |*args|
|
38
|
+
args.length.should == 1
|
39
|
+
# args.first.should =~ %r<source \/etc\/profile && export MM_ACTUAL_JOB_ID=[0-9a-f]{24} MM_ACTUAL_JOB_ANCESTOR_IDS=\\"[0-9a-f]{24}\\" MM_FULL_ACTUAL_JOB_ANCESTOR_IDS=\\"[0-9a-f]{24}\\" MM_ACTUAL_JOB_NAME_PATH=\\"/rjn0001/j11\\" MM_ACTUAL_JOB_SECURITY_TOKEN= MM_SCHEDULE_ID=[0-9a-f]{24} MM_SCHEDULE_ESTIMATED_TIME= MM_TEMPLATE_JOB_ID=[0-9a-f]{24} MM_TEMPLATE_JOB_ANCESTOR_IDS=\\"[0-9a-f]{24}\\" && tengine_job_agent_run -- \$HOME/j11\.sh>
|
40
|
+
args.first.should =~ %r<source \/etc\/profile>
|
41
|
+
args.first.should =~ %r<MM_ACTUAL_JOB_ID=[0-9a-f]{24} MM_ACTUAL_JOB_ANCESTOR_IDS=\"[0-9a-f]{24}\" MM_FULL_ACTUAL_JOB_ANCESTOR_IDS=\"[0-9a-f]{24}\" MM_ACTUAL_JOB_NAME_PATH=\"/rjn0001/j11\" MM_ACTUAL_JOB_SECURITY_TOKEN= MM_SCHEDULE_ID=[0-9a-f]{24} MM_SCHEDULE_ESTIMATED_TIME= MM_TEMPLATE_JOB_ID=[0-9a-f]{24} MM_TEMPLATE_JOB_ANCESTOR_IDS=\"[0-9a-f]{24}\">
|
42
|
+
args.first.should =~ %r<job_test j11>
|
43
|
+
end
|
44
|
+
tengine.receive("start.job.job.tengine", :properties => {
|
45
|
+
:execution_id => @execution.id.to_s,
|
46
|
+
:root_jobnet_id => @jobnet.id.to_s,
|
47
|
+
:root_jobnet_name_path => @jobnet.name_path,
|
48
|
+
:target_jobnet_id => @jobnet.id.to_s,
|
49
|
+
:target_jobnet_name_path => @jobnet.name_path,
|
50
|
+
:target_job_id => @ctx.vertex(:j11).id.to_s,
|
51
|
+
:target_job_name_path => @ctx.vertex(:j11).name_path,
|
52
|
+
})
|
53
|
+
@jobnet.reload
|
54
|
+
@ctx.edge(:e1).phase_key.should == :transmitted
|
55
|
+
@ctx.edge(:e2).phase_key.should == :active
|
56
|
+
@ctx.vertex(:j11).phase_key.should == :starting
|
57
|
+
end
|
58
|
+
|
59
|
+
context "starting直前stopによってinitializedになっている場合" do
|
60
|
+
[:starting, :running].each do |root_phase_key|
|
61
|
+
|
62
|
+
it "ルートが#{root_phase_key}" do
|
63
|
+
@jobnet.phase_key = root_phase_key
|
64
|
+
@ctx[:e1].phase_key = :closing
|
65
|
+
@ctx[:e2].phase_key = :closing
|
66
|
+
@ctx[:e3].phase_key = :closing
|
67
|
+
@ctx[:j11].phase_key = :initialized
|
68
|
+
@jobnet.save!
|
69
|
+
@jobnet.reload
|
70
|
+
tengine.should_fire(:"error.jobnet.job.tengine", {
|
71
|
+
:source_name => @ctx[:root].name_as_resource,
|
72
|
+
:properties=>{
|
73
|
+
:execution_id => @execution.id.to_s,
|
74
|
+
:root_jobnet_id => @jobnet.id.to_s,
|
75
|
+
:root_jobnet_name_path => @jobnet.name_path,
|
76
|
+
:target_jobnet_id => @jobnet.id.to_s,
|
77
|
+
:target_jobnet_name_path => @jobnet.name_path,
|
78
|
+
}
|
79
|
+
})
|
80
|
+
tengine.receive("start.job.job.tengine", :properties => {
|
81
|
+
:execution_id => @execution.id.to_s,
|
82
|
+
:root_jobnet_id => @jobnet.id.to_s,
|
83
|
+
:root_jobnet_name_path => @jobnet.name_path,
|
84
|
+
:target_jobnet_id => @jobnet.id.to_s,
|
85
|
+
:target_jobnet_name_path => @jobnet.name_path,
|
86
|
+
:target_job_id => @ctx.vertex(:j11).id.to_s,
|
87
|
+
:target_job_name_path => @ctx.vertex(:j11).name_path,
|
88
|
+
})
|
89
|
+
@jobnet.reload
|
90
|
+
@ctx.edge(:e1).phase_key.should == :closing
|
91
|
+
@ctx.edge(:e2).phase_key.should == :closed
|
92
|
+
@ctx.edge(:e3).phase_key.should == :closed
|
93
|
+
@ctx.vertex(:j11).phase_key.should == :initialized
|
94
|
+
@jobnet.phase_key.should == :error
|
95
|
+
end
|
96
|
+
end
|
97
|
+
|
98
|
+
end
|
99
|
+
|
100
|
+
it "存在しないスクリプトを実行しようとした場合、標準エラー出力にエラーメッセージが返されるので、それを保持する" do
|
101
|
+
@jobnet.phase_key = :starting
|
102
|
+
@ctx.edge(:e1).phase_key = :transmitting
|
103
|
+
@ctx.vertex(:j11).phase_key = :ready
|
104
|
+
@jobnet.save!
|
105
|
+
@jobnet.reload
|
106
|
+
mock_ssh = mock(:ssh)
|
107
|
+
Net::SSH.stub(:start).with(any_args).and_yield(mock_ssh)
|
108
|
+
mock_channel = mock(:channel)
|
109
|
+
mock_ssh.stub(:open_channel).and_yield(mock_channel)
|
110
|
+
mock_channel.stub(:exec).with(any_args).and_yield(mock_channel, true)
|
111
|
+
mock_channel.stub(:on_data)
|
112
|
+
mock_channel.should_receive(:on_extended_data).and_yield(mock_channel,
|
113
|
+
"session", "[Errno::ENOENT] No such file or directory - /home/goku/unexist_script.sh")
|
114
|
+
mock_channel.stub(:on_close)
|
115
|
+
tengine.should_fire(:"error.job.job.tengine", {
|
116
|
+
:source_name => @ctx[:j11].name_as_resource,
|
117
|
+
:properties=>{
|
118
|
+
:execution_id => @execution.id.to_s,
|
119
|
+
:root_jobnet_id => @jobnet.id.to_s,
|
120
|
+
:root_jobnet_name_path => @jobnet.name_path,
|
121
|
+
:target_jobnet_id => @jobnet.id.to_s,
|
122
|
+
:target_jobnet_name_path => @jobnet.name_path,
|
123
|
+
:target_job_id => @ctx.vertex(:j11).id.to_s,
|
124
|
+
:target_job_name_path => @ctx.vertex(:j11).name_path,
|
125
|
+
:exit_status=>nil,
|
126
|
+
:message=>"Failure to execute /rjn0001/j11 via SSH: [Errno::ENOENT] No such file or directory - /home/goku/unexist_script.sh"
|
127
|
+
}
|
128
|
+
})
|
129
|
+
tengine.receive("start.job.job.tengine", :properties => {
|
130
|
+
:execution_id => @execution.id.to_s,
|
131
|
+
:root_jobnet_id => @jobnet.id.to_s,
|
132
|
+
:root_jobnet_name_path => @jobnet.name_path,
|
133
|
+
:target_jobnet_id => @jobnet.id.to_s,
|
134
|
+
:target_jobnet_name_path => @jobnet.name_path,
|
135
|
+
:target_job_id => @ctx.vertex(:j11).id.to_s,
|
136
|
+
:target_job_name_path => @ctx.vertex(:j11).name_path,
|
137
|
+
})
|
138
|
+
@jobnet.reload
|
139
|
+
@ctx.edge(:e1).phase_key.should == :transmitted
|
140
|
+
@ctx.edge(:e2).phase_key.should == :active
|
141
|
+
@ctx.vertex(:j11).tap do |job|
|
142
|
+
job.phase_key.should == :error
|
143
|
+
job.error_messages.should == [
|
144
|
+
"[Errno::ENOENT] No such file or directory - /home/goku/unexist_script.sh",
|
145
|
+
"Failure to execute /rjn0001/j11 via SSH: [Errno::ENOENT] No such file or directory - /home/goku/unexist_script.sh"
|
146
|
+
]
|
147
|
+
end
|
148
|
+
@jobnet.phase_key.should == :running
|
149
|
+
end
|
150
|
+
|
151
|
+
end
|
152
|
+
|
153
|
+
|
154
|
+
it "PIDを取得できたら" do
|
155
|
+
@ctx.edge(:e1).phase_key = :transmitted
|
156
|
+
@ctx.edge(:e2).phase_key = :active
|
157
|
+
@ctx.vertex(:j11).phase_key = :starting
|
158
|
+
@jobnet.save!
|
159
|
+
@jobnet.reload
|
160
|
+
tengine.should_not_fire
|
161
|
+
mock_event = mock(:event)
|
162
|
+
@pid = "123"
|
163
|
+
signal = Tengine::Job::Signal.new(mock_event)
|
164
|
+
signal.data = {:executing_pid => @pid}
|
165
|
+
@ctx.vertex(:j11).ack(signal) # このメソッド内ではsaveされないので、ここでreloadもしません。
|
166
|
+
@ctx.vertex(:j11).executing_pid.should == @pid
|
167
|
+
@ctx.edge(:e1).phase_key.should == :transmitted
|
168
|
+
@ctx.edge(:e2).phase_key.should == :active
|
169
|
+
@ctx.vertex(:j11).phase_key.should == :running
|
170
|
+
end
|
171
|
+
|
172
|
+
test_error_message1 = "Job process failed. STDOUT and STDERR were redirected to files. You can see them at /home/goku/stdout-1234.log and /home/goku/stderr-1234.log on the server test_server1"
|
173
|
+
{
|
174
|
+
:success => ["0", {}],
|
175
|
+
:error => ["1", {
|
176
|
+
:stdout_log => "/home/goku/stdout-1234.log",
|
177
|
+
:stderr_log => "/home/goku/stderr-1234.log",
|
178
|
+
:message => test_error_message1
|
179
|
+
}]
|
180
|
+
}.each do |phase_key, (exit_status, extra_props)|
|
181
|
+
it "ジョブ実行#{phase_key}の通知" do
|
182
|
+
test_key = "test_key.finished.process.job.tengine"
|
183
|
+
Tengine::Core::Event.delete_all(:conditions => {:key => test_key})
|
184
|
+
Tengine::Core::Event.create!(:event_type_name => "job.heartbeat.tengine", :key => test_key)
|
185
|
+
@jobnet.reload
|
186
|
+
j11 = @jobnet.find_descendant_by_name_path("/rjn0001/j11")
|
187
|
+
j11.executing_pid = "123"
|
188
|
+
j11.phase_key = :running
|
189
|
+
j11.previous_edges.length.should == 1
|
190
|
+
j11.previous_edges.first.phase_key = :transmitted
|
191
|
+
@ctx[:root].save!
|
192
|
+
tengine.should_fire(:"#{phase_key}.job.job.tengine",
|
193
|
+
:source_name => @ctx[:j11].name_as_resource,
|
194
|
+
:properties => {
|
195
|
+
:execution_id => @execution.id.to_s,
|
196
|
+
:root_jobnet_id => @jobnet.id.to_s,
|
197
|
+
:root_jobnet_name_path => @jobnet.name_path,
|
198
|
+
:target_jobnet_id => @jobnet.id.to_s,
|
199
|
+
:target_jobnet_name_path => @jobnet.name_path,
|
200
|
+
:target_job_id => @ctx[:j11].id.to_s,
|
201
|
+
:target_job_name_path => @ctx[:j11].name_path,
|
202
|
+
:exit_status => exit_status
|
203
|
+
})
|
204
|
+
tengine.receive(:"finished.process.job.tengine",
|
205
|
+
:key => test_key,
|
206
|
+
:source_name => @ctx[:j11].name_as_resource,
|
207
|
+
:properties => {
|
208
|
+
:execution_id => @execution.id.to_s,
|
209
|
+
:root_jobnet_id => @jobnet.id.to_s,
|
210
|
+
:root_jobnet_name_path => @jobnet.name_path,
|
211
|
+
:target_jobnet_id => @jobnet.id.to_s,
|
212
|
+
:target_jobnet_name_path => @jobnet.name_path,
|
213
|
+
:target_job_id => @ctx[:j11].id.to_s,
|
214
|
+
:target_job_name_path => @ctx[:j11].name_path,
|
215
|
+
:exit_status => exit_status
|
216
|
+
}.merge(extra_props))
|
217
|
+
@jobnet.reload
|
218
|
+
@ctx.edge(:e1).phase_key.should == :transmitted
|
219
|
+
@ctx.edge(:e2).phase_key.should == :active
|
220
|
+
@ctx.vertex(:j11).tap do |j|
|
221
|
+
j.phase_key.should == phase_key
|
222
|
+
j.exit_status.should == exit_status
|
223
|
+
if phase_key == :error
|
224
|
+
j.error_messages.should == [test_error_message1]
|
225
|
+
end
|
226
|
+
end
|
227
|
+
end
|
228
|
+
end
|
229
|
+
|
230
|
+
it "強制停止" do
|
231
|
+
@pid = "123"
|
232
|
+
@jobnet.reload
|
233
|
+
j11 = @jobnet.find_descendant_by_name_path("/rjn0001/j11")
|
234
|
+
j11.executing_pid = @pid
|
235
|
+
j11.phase_key = :running
|
236
|
+
j11.previous_edges.length.should == 1
|
237
|
+
j11.previous_edges.first.phase_key = :transmitted
|
238
|
+
@ctx[:root].save!
|
239
|
+
|
240
|
+
tengine.should_not_fire
|
241
|
+
mock_ssh = mock(:ssh)
|
242
|
+
mock_channel = mock(:channel)
|
243
|
+
Net::SSH.should_receive(:start).
|
244
|
+
with("localhost", an_instance_of(Tengine::Resource::Credential), an_instance_of(Hash)).and_yield(mock_ssh)
|
245
|
+
mock_ssh.should_receive(:open_channel).and_yield(mock_channel)
|
246
|
+
mock_channel.should_receive(:exec) do |*args|
|
247
|
+
interval = Tengine::Job::Killing::DEFAULT_KILLING_SIGNAL_INTERVAL
|
248
|
+
args.length.should == 1
|
249
|
+
args.first.should =~ %r<source \/etc\/profile>
|
250
|
+
args.first.should =~ %r<tengine_job_agent_kill #{@pid} #{interval} KILL$>
|
251
|
+
end
|
252
|
+
tengine.receive(:"stop.job.job.tengine",
|
253
|
+
:source_name => @ctx[:j11].name_as_resource,
|
254
|
+
:properties => {
|
255
|
+
:execution_id => @execution.id.to_s,
|
256
|
+
:root_jobnet_id => @jobnet.id.to_s,
|
257
|
+
:target_jobnet_id => @jobnet.id.to_s,
|
258
|
+
:target_job_id => @ctx[:j11].id.to_s,
|
259
|
+
})
|
260
|
+
@jobnet.reload
|
261
|
+
@ctx.edge(:e1).phase_key.should == :transmitted
|
262
|
+
@ctx.edge(:e2).phase_key.should == :active
|
263
|
+
@ctx.vertex(:j11).tap do |j|
|
264
|
+
j.phase_key.should == :dying
|
265
|
+
j.exit_status.should == nil
|
266
|
+
end
|
267
|
+
end
|
268
|
+
|
269
|
+
it "強制停止(ジョブネット)" do
|
270
|
+
@pid11 = "11"
|
271
|
+
@pid12 = "12"
|
272
|
+
@jobnet.reload
|
273
|
+
j11 = @jobnet.find_descendant_by_name_path("/rjn0001/j11")
|
274
|
+
j11.executing_pid = @pid11
|
275
|
+
j11.phase_key = :success
|
276
|
+
j11.previous_edges.length.should == 1
|
277
|
+
j11.previous_edges.first.phase_key = :transmitted
|
278
|
+
j12 = @jobnet.find_descendant_by_name_path("/rjn0001/j12")
|
279
|
+
j12.executing_pid = @pid12
|
280
|
+
j12.phase_key = :running
|
281
|
+
j12.previous_edges.length.should == 1
|
282
|
+
j12.previous_edges.first.phase_key = :transmitted
|
283
|
+
@ctx[:root].save!
|
284
|
+
|
285
|
+
# phase_key が success の j11 は fireされない
|
286
|
+
tengine.should_not_fire(:"stop.job.job.tengine")
|
287
|
+
# phase_key が running の j12 は fireされる
|
288
|
+
tengine.should_fire(:"stop.job.job.tengine",
|
289
|
+
:source_name => @ctx[:j12].name_as_resource,
|
290
|
+
:properties => {
|
291
|
+
:stop_reason => "user_stop",
|
292
|
+
:target_jobnet_id => @jobnet.id.to_s,
|
293
|
+
:target_jobnet_name_path => "/rjn0001",
|
294
|
+
:target_job_id => @ctx[:j12].id.to_s,
|
295
|
+
:target_job_name_path => "/rjn0001/j12",
|
296
|
+
:execution_id => @execution.id.to_s,
|
297
|
+
:root_jobnet_id => @jobnet.id.to_s,
|
298
|
+
:root_jobnet_name_path => "/rjn0001",
|
299
|
+
})
|
300
|
+
# jobnet に対して強制停止された
|
301
|
+
tengine.receive(:"stop.jobnet.job.tengine",
|
302
|
+
:source_name => @jobnet.name_as_resource,
|
303
|
+
:properties => {
|
304
|
+
:stop_reason => "user_stop",
|
305
|
+
:target_jobnet_id => @jobnet.id.to_s,
|
306
|
+
:target_jobnet_name_path => "/rjn0001",
|
307
|
+
:execution_id => @execution.id.to_s,
|
308
|
+
:root_jobnet_id => @jobnet.id.to_s,
|
309
|
+
:root_jobnet_name_path => "/rjn0001",
|
310
|
+
})
|
311
|
+
end
|
312
|
+
|
313
|
+
it "強制停止(後続のジョブ)" do
|
314
|
+
@pid11 = "11"
|
315
|
+
@pid12 = "12"
|
316
|
+
@jobnet.reload
|
317
|
+
j11 = @jobnet.find_descendant_by_name_path("/rjn0001/j11")
|
318
|
+
j11.executing_pid = @pid11
|
319
|
+
j11.phase_key = :success
|
320
|
+
j11.previous_edges.length.should == 1
|
321
|
+
j11.previous_edges.first.phase_key = :transmitted
|
322
|
+
j12 = @jobnet.find_descendant_by_name_path("/rjn0001/j12")
|
323
|
+
j12.executing_pid = @pid12
|
324
|
+
j12.phase_key = :running
|
325
|
+
j12.previous_edges.length.should == 1
|
326
|
+
j12.previous_edges.first.phase_key = :transmitted
|
327
|
+
@ctx[:root].save!
|
328
|
+
|
329
|
+
mock_ssh = mock(:ssh)
|
330
|
+
mock_channel = mock(:channel)
|
331
|
+
Net::SSH.should_receive(:start).
|
332
|
+
with("localhost", an_instance_of(Tengine::Resource::Credential), an_instance_of(Hash)).and_yield(mock_ssh)
|
333
|
+
mock_ssh.should_receive(:open_channel).and_yield(mock_channel)
|
334
|
+
mock_channel.should_receive(:exec) do |*args|
|
335
|
+
interval = Tengine::Job::Killing::DEFAULT_KILLING_SIGNAL_INTERVAL
|
336
|
+
args.length.should == 1
|
337
|
+
args.first.should =~ %r<source \/etc\/profile>
|
338
|
+
args.first.should =~ %r<tengine_job_agent_kill #{@pid12} #{interval} KILL$>
|
339
|
+
end
|
340
|
+
|
341
|
+
# job12 に対して強制停止
|
342
|
+
tengine.receive(:"stop.job.job.tengine",
|
343
|
+
:source_name => @ctx[:j12].name_as_resource,
|
344
|
+
:properties => {
|
345
|
+
:stop_reason => "user_stop",
|
346
|
+
:target_jobnet_id => @jobnet.id.to_s,
|
347
|
+
:target_jobnet_name_path => "/rjn0001",
|
348
|
+
:target_job_id => @ctx[:j12].id.to_s,
|
349
|
+
:target_job_name_path => "/rjn0001/j12",
|
350
|
+
:execution_id => @execution.id.to_s,
|
351
|
+
:root_jobnet_id => @jobnet.id.to_s,
|
352
|
+
:root_jobnet_name_path => "/rjn0001",
|
353
|
+
})
|
354
|
+
@jobnet.reload
|
355
|
+
@ctx.edge(:e1).phase_key.should == :transmitted
|
356
|
+
@ctx.edge(:e2).phase_key.should == :transmitted
|
357
|
+
@ctx.edge(:e3).phase_key.should == :active
|
358
|
+
@ctx.vertex(:j11).tap do |j|
|
359
|
+
j.phase_key.should == :success
|
360
|
+
j.stop_reason.should == nil
|
361
|
+
end
|
362
|
+
@ctx.vertex(:j12).tap do |j|
|
363
|
+
j.phase_key.should == :dying
|
364
|
+
j.stop_reason.should == "user_stop"
|
365
|
+
end
|
366
|
+
end
|
367
|
+
|
368
|
+
|
369
|
+
if ENV['PASSWORD']
|
370
|
+
context "実際にSSHで接続", :ssh_actual => true do
|
371
|
+
before do
|
372
|
+
resource_fixture = GokuAtEc2ApNortheast.new
|
373
|
+
credential = resource_fixture.goku_ssh_pw
|
374
|
+
credential.auth_values = {:username => ENV['USER'], :password => ENV['PASSWORD']}
|
375
|
+
credential.save!
|
376
|
+
server = resource_fixture.hadoop_master_node
|
377
|
+
server.local_ipv4 = "127.0.0.1"
|
378
|
+
server.save!
|
379
|
+
end
|
380
|
+
|
381
|
+
it do
|
382
|
+
tengine.should_not_fire
|
383
|
+
tengine.receive("start.job.job.tengine", :properties => {
|
384
|
+
:execution_id => @execution.id.to_s,
|
385
|
+
:root_jobnet_id => @jobnet.id.to_s,
|
386
|
+
:target_jobnet_id => @jobnet.id.to_s,
|
387
|
+
})
|
388
|
+
@jobnet.reload
|
389
|
+
j11 = @jobnet.find_descendant_by_name_path("/rjn0001/j11")
|
390
|
+
j11.executing_pid.should_not be_nil
|
391
|
+
j11.exit_status.should == nil
|
392
|
+
j11.phase_key.should == :running
|
393
|
+
j11.previous_edges.length.should == 1
|
394
|
+
j11.previous_edges.first.phase_key.should == :transmitted
|
395
|
+
end
|
396
|
+
|
397
|
+
end
|
398
|
+
end
|
399
|
+
end
|
400
|
+
|
401
|
+
context "再実行" do
|
402
|
+
context "ジョブを再実行" do
|
403
|
+
{
|
404
|
+
false => "後続も実行",
|
405
|
+
true => "スポット再実行"
|
406
|
+
}.each do |spot, caption|
|
407
|
+
context(caption) do
|
408
|
+
|
409
|
+
before do
|
410
|
+
Tengine::Job::Vertex.delete_all
|
411
|
+
builder = Rjn0001SimpleJobnetBuilder.new
|
412
|
+
@root = builder.create_actual
|
413
|
+
@ctx = builder.context
|
414
|
+
@execution = Tengine::Job::Execution.create!({
|
415
|
+
:root_jobnet_id => @root.id,
|
416
|
+
:spot => spot, :retry => true,
|
417
|
+
:target_actual_ids => [@ctx[:j11].id.to_s]
|
418
|
+
})
|
419
|
+
@root.phase_key = :running
|
420
|
+
@ctx[:j11].phase_key = :success
|
421
|
+
@ctx[:j12].phase_key = :error
|
422
|
+
@ctx[:e1].phase_key = :transmitted
|
423
|
+
@ctx[:e2].phase_key = :transmitted
|
424
|
+
@ctx[:e3].phase_key = :active
|
425
|
+
end
|
426
|
+
|
427
|
+
[:initialized, :success, :error, :stuck].each do |phase_key|
|
428
|
+
it "phase_keyが#{phase_key}ならば再実行できるので、startのイベントを発火する" do
|
429
|
+
@ctx[:j11].phase_key = phase_key
|
430
|
+
@root.save!
|
431
|
+
tengine.should_fire(:"start.job.job.tengine", {
|
432
|
+
:source_name => @ctx[:j11].name_as_resource,
|
433
|
+
:properties=>{
|
434
|
+
:execution_id => @execution.id.to_s,
|
435
|
+
:root_jobnet_name_path => @root.name_path,
|
436
|
+
:root_jobnet_id => @root.id.to_s,
|
437
|
+
:target_jobnet_name_path => @root.name_path,
|
438
|
+
:target_jobnet_id => @root.id.to_s,
|
439
|
+
:target_job_name_path => @ctx.vertex(:j11).name_path,
|
440
|
+
:target_job_id => @ctx.vertex(:j11).id.to_s,
|
441
|
+
}
|
442
|
+
})
|
443
|
+
tengine.receive("restart.job.job.tengine", :properties => {
|
444
|
+
:execution_id => @execution.id.to_s,
|
445
|
+
:root_jobnet_id => @root.id.to_s,
|
446
|
+
:root_jobnet_name_path => @root.name_path,
|
447
|
+
:target_jobnet_id => @root.id.to_s,
|
448
|
+
:target_jobnet_name_path => @root.name_path,
|
449
|
+
:target_job_id => @ctx.vertex(:j11).id.to_s,
|
450
|
+
:target_job_name_path => @ctx.vertex(:j11).name_path,
|
451
|
+
})
|
452
|
+
@root.reload
|
453
|
+
@root.phase_key.should == :running
|
454
|
+
@ctx.edge(:e1).phase_key.should == :transmitted
|
455
|
+
@ctx.vertex(:j11).phase_key.should == :ready
|
456
|
+
if spot
|
457
|
+
@ctx.vertex(:j12).phase_key.should == :error
|
458
|
+
@ctx.edge(:e2).phase_key.should == :transmitted
|
459
|
+
@ctx.edge(:e3).phase_key.should == :active
|
460
|
+
else
|
461
|
+
@ctx.vertex(:j12).phase_key.should == :initialized
|
462
|
+
@ctx.edge(:e2).phase_key.should == :active
|
463
|
+
@ctx.edge(:e3).phase_key.should == :active
|
464
|
+
end
|
465
|
+
end
|
466
|
+
end
|
467
|
+
|
468
|
+
[:ready, :starting, :running, :dying].each do |phase_key|
|
469
|
+
it "phase_keyが#{phase_key}ならば再実行できず、エラーのイベントを発火する" do
|
470
|
+
@ctx[:j11].phase_key = phase_key
|
471
|
+
@root.save!
|
472
|
+
tengine.should_fire("restart.job.job.tengine.error.tengined").with(any_args)
|
473
|
+
Tengine::Core::Kernel.temp_exception_reporter(:except_test) do
|
474
|
+
tengine.receive("restart.job.job.tengine", :properties => {
|
475
|
+
:execution_id => @execution.id.to_s,
|
476
|
+
:root_jobnet_id => @root.id.to_s,
|
477
|
+
:target_jobnet_id => @root.id.to_s,
|
478
|
+
:target_job_id => @ctx.vertex(:j11).id.to_s,
|
479
|
+
})
|
480
|
+
end
|
481
|
+
# 再実行に失敗したのでルートジョブネット以下何も状態は変更されません
|
482
|
+
@root.reload
|
483
|
+
@root.phase_key.should == :running
|
484
|
+
@ctx.edge(:e1).phase_key.should == :transmitted
|
485
|
+
@ctx.vertex(:j11).phase_key.should == phase_key
|
486
|
+
end
|
487
|
+
|
488
|
+
end
|
489
|
+
end
|
490
|
+
end
|
491
|
+
|
492
|
+
end
|
493
|
+
|
494
|
+
end
|
495
|
+
|
496
|
+
|
497
|
+
context "<BUG>同じジョブネットが複数バージョン存在する際、ジョブ実行時にスクリプトに渡される環境変数の「MM_TEMPLATE_JOB_ID」「MM_TEMPLATE_JOB_ANCESTOR_IDS」が実行しているバージョン以外のものがセットされている" do
|
498
|
+
shared_examples_for "最新のバージョンのルートジョブネットを参照する" do |dsl_version|
|
499
|
+
|
500
|
+
it do
|
501
|
+
@root.phase_key = :starting
|
502
|
+
@root.element("prev!j11").phase_key = :transmitting
|
503
|
+
@root.element('j11').phase_key = :ready
|
504
|
+
@root.save!
|
505
|
+
@root.reload
|
506
|
+
tengine.should_not_fire
|
507
|
+
mock_ssh = mock(:ssh)
|
508
|
+
mock_channel = mock(:channel)
|
509
|
+
Net::SSH.should_receive(:start).
|
510
|
+
with("localhost", an_instance_of(Tengine::Resource::Credential), an_instance_of(Hash)).and_yield(mock_ssh)
|
511
|
+
mock_ssh.should_receive(:open_channel).and_yield(mock_channel)
|
512
|
+
mock_channel.should_receive(:exec) do |*args|
|
513
|
+
args.length.should == 1
|
514
|
+
# args.first.should =~ %r<source \/etc\/profile && export MM_ACTUAL_JOB_ID=[0-9a-f]{24} MM_ACTUAL_JOB_ANCESTOR_IDS=\\"[0-9a-f]{24}\\" MM_FULL_ACTUAL_JOB_ANCESTOR_IDS=\\"[0-9a-f]{24}\\" MM_ACTUAL_JOB_NAME_PATH=\\"/rjn0001/j11\\" MM_ACTUAL_JOB_SECURITY_TOKEN= MM_SCHEDULE_ID=[0-9a-f]{24} MM_SCHEDULE_ESTIMATED_TIME= MM_TEMPLATE_JOB_ID=[0-9a-f]{24} MM_TEMPLATE_JOB_ANCESTOR_IDS=\\"[0-9a-f]{24}\\" && tengine_job_agent_run -- \$HOME/j11\.sh>
|
515
|
+
args.first.should =~ %r<source \/etc\/profile>
|
516
|
+
args.first.should =~ %r<MM_ACTUAL_JOB_ID=[0-9a-f]{24} MM_ACTUAL_JOB_ANCESTOR_IDS=\"[0-9a-f]{24}\" MM_FULL_ACTUAL_JOB_ANCESTOR_IDS=\"[0-9a-f]{24}\" MM_ACTUAL_JOB_NAME_PATH=\"/rjn0001/j11\" MM_ACTUAL_JOB_SECURITY_TOKEN= MM_SCHEDULE_ID=[0-9a-f]{24} MM_SCHEDULE_ESTIMATED_TIME= MM_TEMPLATE_JOB_ID=[0-9a-f]{24} MM_TEMPLATE_JOB_ANCESTOR_IDS=\"[0-9a-f]{24}\">
|
517
|
+
@template.dsl_version.should == dsl_version
|
518
|
+
template_job = @template.element("/rjn0001/j11")
|
519
|
+
args.first.should =~ %r<MM_TEMPLATE_JOB_ID=#{template_job.id.to_s}>
|
520
|
+
args.first.should =~ %r<MM_TEMPLATE_JOB_ANCESTOR_IDS=\"#{@template.id.to_s}\">
|
521
|
+
args.first.should =~ %r<job_test j11>
|
522
|
+
end
|
523
|
+
tengine.receive("start.job.job.tengine", :properties => {
|
524
|
+
:execution_id => @execution.id.to_s,
|
525
|
+
:root_jobnet_id => @root.id.to_s,
|
526
|
+
:root_jobnet_name_path => @root.name_path,
|
527
|
+
:target_jobnet_id => @root.id.to_s,
|
528
|
+
:target_jobnet_name_path => @root.name_path,
|
529
|
+
:target_job_id => @root.element('j11').id.to_s,
|
530
|
+
:target_job_name_path => @root.element('j11').name_path,
|
531
|
+
})
|
532
|
+
@root.reload
|
533
|
+
@root.element('prev!j11').phase_key.should == :transmitted
|
534
|
+
@root.element('next!j11').phase_key.should == :active
|
535
|
+
@root.element('j11').phase_key.should == :starting
|
536
|
+
end
|
537
|
+
end
|
538
|
+
|
539
|
+
context "バージョン1つだけ" do
|
540
|
+
before do
|
541
|
+
Tengine::Core::Setting.delete_all
|
542
|
+
Tengine::Core::Setting.create!(:name => "dsl_version", :value => "1")
|
543
|
+
Tengine::Job::Vertex.delete_all
|
544
|
+
Rjn0001SimpleJobnetBuilder.new.tap do |builder|
|
545
|
+
@template = builder.create_template(:dsl_version => "1")
|
546
|
+
@root = @template.generate
|
547
|
+
@ctx = builder.context
|
548
|
+
end
|
549
|
+
@execution = Tengine::Job::Execution.create!({
|
550
|
+
:root_jobnet_id => @root.id,
|
551
|
+
})
|
552
|
+
end
|
553
|
+
it{ @root.template.dsl_version.should == "1" }
|
554
|
+
it_should_behave_like "最新のバージョンのルートジョブネットを参照する", "1"
|
555
|
+
end
|
556
|
+
|
557
|
+
context "バージョン2つ" do
|
558
|
+
before do
|
559
|
+
Tengine::Core::Setting.delete_all
|
560
|
+
Tengine::Core::Setting.create!(:name => "dsl_version", :value => "2")
|
561
|
+
Tengine::Job::Vertex.delete_all
|
562
|
+
Rjn0001SimpleJobnetBuilder.new.tap do |builder|
|
563
|
+
builder.create_template(:dsl_version => "1")
|
564
|
+
@template = builder.create_template(:dsl_version => "2")
|
565
|
+
@root = @template.generate
|
566
|
+
@ctx = builder.context
|
567
|
+
end
|
568
|
+
@execution = Tengine::Job::Execution.create!({
|
569
|
+
:root_jobnet_id => @root.id,
|
570
|
+
})
|
571
|
+
end
|
572
|
+
it{ @root.template.dsl_version.should == "2" }
|
573
|
+
it_should_behave_like "最新のバージョンのルートジョブネットを参照する", "2"
|
574
|
+
end
|
575
|
+
|
576
|
+
context "バージョン10個" do
|
577
|
+
before do
|
578
|
+
Tengine::Core::Setting.delete_all
|
579
|
+
Tengine::Core::Setting.create!(:name => "dsl_version", :value => "10")
|
580
|
+
Tengine::Job::Vertex.delete_all
|
581
|
+
Rjn0001SimpleJobnetBuilder.new.tap do |builder|
|
582
|
+
(1..9).each do |idx|
|
583
|
+
builder.create_template(:dsl_version => idx.to_s)
|
584
|
+
end
|
585
|
+
@template = builder.create_template(:dsl_version => "10")
|
586
|
+
@root = @template.generate
|
587
|
+
@ctx = builder.context
|
588
|
+
end
|
589
|
+
@execution = Tengine::Job::Execution.create!({
|
590
|
+
:root_jobnet_id => @root.id,
|
591
|
+
})
|
592
|
+
end
|
593
|
+
it{ @root.template.dsl_version.should == "10" }
|
594
|
+
it_should_behave_like "最新のバージョンのルートジョブネットを参照する", "10"
|
595
|
+
end
|
596
|
+
end
|
597
|
+
|
598
|
+
context "https://www.pivotaltracker.com/story/show/22624209" do
|
599
|
+
it "stuckにする" do
|
600
|
+
Tengine::Core::Schedule.delete_all
|
601
|
+
Tengine::Job::Vertex.delete_all
|
602
|
+
builder = Rjn0001SimpleJobnetBuilder.new
|
603
|
+
@root = builder.create_actual
|
604
|
+
@ctx = builder.context
|
605
|
+
@execution = Tengine::Job::Execution.create!({
|
606
|
+
:root_jobnet_id => @root.id,
|
607
|
+
})
|
608
|
+
@root.phase_key = :initialized
|
609
|
+
@root.save!
|
610
|
+
EM.run_block do
|
611
|
+
tengine.receive("expired.job.heartbeat.tengine", :properties => {
|
612
|
+
:execution_id => @execution.id.to_s,
|
613
|
+
:root_jobnet_id => @root.id.to_s,
|
614
|
+
:target_job_id => @root.children[1].id.to_s,
|
615
|
+
})
|
616
|
+
end
|
617
|
+
@root.reload
|
618
|
+
@root.children[1].phase_key.should == :stuck
|
619
|
+
@root.phase_key.should_not == :stuck # initialized
|
620
|
+
end
|
621
|
+
end
|
622
|
+
|
623
|
+
end
|