spbtv_redis_failover 1.0.2.1

Sign up to get free protection for your applications and to get access to all the features.
Files changed (53) hide show
  1. checksums.yaml +7 -0
  2. data/.gitignore +19 -0
  3. data/.travis.yml +7 -0
  4. data/.yardopts +6 -0
  5. data/Changes.md +191 -0
  6. data/Gemfile +2 -0
  7. data/LICENSE +22 -0
  8. data/README.md +240 -0
  9. data/Rakefile +9 -0
  10. data/bin/redis_node_manager +7 -0
  11. data/examples/config.yml +17 -0
  12. data/examples/multiple_environments_config.yml +15 -0
  13. data/lib/redis_failover.rb +25 -0
  14. data/lib/redis_failover/cli.rb +142 -0
  15. data/lib/redis_failover/client.rb +517 -0
  16. data/lib/redis_failover/errors.rb +54 -0
  17. data/lib/redis_failover/failover_strategy.rb +25 -0
  18. data/lib/redis_failover/failover_strategy/latency.rb +21 -0
  19. data/lib/redis_failover/manual_failover.rb +52 -0
  20. data/lib/redis_failover/node.rb +190 -0
  21. data/lib/redis_failover/node_manager.rb +741 -0
  22. data/lib/redis_failover/node_snapshot.rb +81 -0
  23. data/lib/redis_failover/node_strategy.rb +34 -0
  24. data/lib/redis_failover/node_strategy/consensus.rb +18 -0
  25. data/lib/redis_failover/node_strategy/majority.rb +18 -0
  26. data/lib/redis_failover/node_strategy/single.rb +17 -0
  27. data/lib/redis_failover/node_watcher.rb +83 -0
  28. data/lib/redis_failover/runner.rb +27 -0
  29. data/lib/redis_failover/util.rb +137 -0
  30. data/lib/redis_failover/version.rb +3 -0
  31. data/misc/redis_failover.png +0 -0
  32. data/spbtv_redis_failover.gemspec +26 -0
  33. data/spec/cli_spec.rb +75 -0
  34. data/spec/client_spec.rb +153 -0
  35. data/spec/failover_strategy/latency_spec.rb +41 -0
  36. data/spec/failover_strategy_spec.rb +17 -0
  37. data/spec/node_manager_spec.rb +136 -0
  38. data/spec/node_snapshot_spec.rb +30 -0
  39. data/spec/node_spec.rb +84 -0
  40. data/spec/node_strategy/consensus_spec.rb +30 -0
  41. data/spec/node_strategy/majority_spec.rb +22 -0
  42. data/spec/node_strategy/single_spec.rb +22 -0
  43. data/spec/node_strategy_spec.rb +22 -0
  44. data/spec/node_watcher_spec.rb +58 -0
  45. data/spec/spec_helper.rb +21 -0
  46. data/spec/support/config/multiple_environments.yml +15 -0
  47. data/spec/support/config/multiple_environments_with_chroot.yml +17 -0
  48. data/spec/support/config/single_environment.yml +7 -0
  49. data/spec/support/config/single_environment_with_chroot.yml +8 -0
  50. data/spec/support/node_manager_stub.rb +87 -0
  51. data/spec/support/redis_stub.rb +105 -0
  52. data/spec/util_spec.rb +21 -0
  53. metadata +207 -0
data/spec/node_spec.rb ADDED
@@ -0,0 +1,84 @@
1
+ require 'spec_helper'
2
+
3
+ module RedisFailover
4
+ describe Node do
5
+ let(:node) { Node.new(:host => 'localhost', :port => '123') }
6
+
7
+ before(:each) do
8
+ node.extend(RedisStubSupport)
9
+ end
10
+
11
+ describe '#initialize' do
12
+ it 'creates a new instance' do
13
+ node.host.should == 'localhost'
14
+ node.port.should == 123
15
+ end
16
+
17
+ it 'reports error if host missing' do
18
+ expect { Node.new }.to raise_error(InvalidNodeError)
19
+ end
20
+ end
21
+
22
+ describe '#ping' do
23
+ it 'responds properly if node is available' do
24
+ expect { node.ping }.to_not raise_error
25
+ end
26
+
27
+ it 'responds properly if node is unavailable' do
28
+ node.redis.make_unavailable!
29
+ expect { node.ping }.to raise_error(NodeUnavailableError)
30
+ end
31
+ end
32
+
33
+ describe '#master?' do
34
+ it 'responds properly if node is master' do
35
+ node.should be_master
36
+ end
37
+
38
+ it 'responds properly if node is not master' do
39
+ node.make_slave!(Node.new(:host => 'masterhost'))
40
+ node.should_not be_master
41
+ end
42
+ end
43
+
44
+
45
+ describe '#slave?' do
46
+ it 'responds properly if node is slave' do
47
+ node.should_not be_slave
48
+ end
49
+
50
+ it 'responds properly if node is not slave' do
51
+ node.make_master!
52
+ node.should be_master
53
+ end
54
+ end
55
+
56
+ describe '#wait' do
57
+ it 'should wait until node dies' do
58
+ thread = Thread.new { node.wait }
59
+ thread.should be_alive
60
+ node.redis.make_unavailable!
61
+ expect { thread.value }.to raise_error
62
+ end
63
+ end
64
+
65
+ describe '#wakeup' do
66
+ it 'should gracefully stop waiting' do
67
+ thread = Thread.new { node.wait }
68
+ thread.should be_alive
69
+ node.wakeup
70
+ sleep 2
71
+ thread.should_not be_alive
72
+ thread.value.should be_nil
73
+ end
74
+ end
75
+
76
+ describe '#perform_operation' do
77
+ it 'raises error for any operation that hangs for too long' do
78
+ expect do
79
+ node.send(:perform_operation) { 1_000_000.times { sleep 0.1 } }
80
+ end.to raise_error(NodeUnavailableError)
81
+ end
82
+ end
83
+ end
84
+ end
@@ -0,0 +1,30 @@
1
+ require 'spec_helper'
2
+
3
+ module RedisFailover
4
+ class NodeStrategy
5
+ NodeStrategy.for(:consensus)
6
+
7
+ describe Consensus do
8
+ let(:node) { Node.new(:host => 'localhost', :port => '123') }
9
+ let(:snapshot) { NodeSnapshot.new(node) }
10
+
11
+ describe '#determine_state' do
12
+ it 'returns the unavailable state if unavailable by all node managers' do
13
+ strategy = NodeStrategy.for(:consensus)
14
+ snapshot.unviewable_by('nm1')
15
+ snapshot.unviewable_by('nm2')
16
+ snapshot.unviewable_by('nm3')
17
+ strategy.determine_state(node, node => snapshot).should == :unavailable
18
+ end
19
+
20
+ it 'returns the available state if unavailable by some node managers' do
21
+ strategy = NodeStrategy.for(:consensus)
22
+ snapshot.unviewable_by('nm1')
23
+ snapshot.unviewable_by('nm2')
24
+ snapshot.viewable_by('nm3', 0)
25
+ strategy.determine_state(node, node => snapshot).should == :available
26
+ end
27
+ end
28
+ end
29
+ end
30
+ end
@@ -0,0 +1,22 @@
1
+ require 'spec_helper'
2
+
3
+ module RedisFailover
4
+ class NodeStrategy
5
+ NodeStrategy.for(:majority)
6
+
7
+ describe Majority do
8
+ let(:node) { Node.new(:host => 'localhost', :port => '123') }
9
+ let(:snapshot) { NodeSnapshot.new(node) }
10
+
11
+ describe '#determine_state' do
12
+ it 'returns the unavailable state if unavailable by the majority of node managers' do
13
+ strategy = NodeStrategy.for(:majority)
14
+ snapshot.viewable_by('nm1', 0)
15
+ snapshot.unviewable_by('nm2')
16
+ snapshot.unviewable_by('nm3')
17
+ strategy.determine_state(node, node => snapshot).should == :unavailable
18
+ end
19
+ end
20
+ end
21
+ end
22
+ end
@@ -0,0 +1,22 @@
1
+ require 'spec_helper'
2
+
3
+ module RedisFailover
4
+ class NodeStrategy
5
+ NodeStrategy.for(:single)
6
+
7
+ describe Single do
8
+ let(:node) { Node.new(:host => 'localhost', :port => '123') }
9
+ let(:snapshot) { NodeSnapshot.new(node) }
10
+
11
+ describe '#determine_state' do
12
+ it 'returns the unavailable state if any node manager reports as down' do
13
+ strategy = NodeStrategy.for(:single)
14
+ snapshot.unviewable_by('nm1')
15
+ snapshot.viewable_by('nm2', 0)
16
+ snapshot.viewable_by('nm3', 0)
17
+ strategy.determine_state(node, node => snapshot).should == :unavailable
18
+ end
19
+ end
20
+ end
21
+ end
22
+ end
@@ -0,0 +1,22 @@
1
+ require 'spec_helper'
2
+
3
+ module RedisFailover
4
+ describe NodeStrategy do
5
+
6
+ describe '.for' do
7
+ it 'creates a new majority strategy instance' do
8
+ s = NodeStrategy.for('majority')
9
+ s.should be_a RedisFailover::NodeStrategy::Majority
10
+ end
11
+
12
+ it 'creates a new consensus strategy instance' do
13
+ s = NodeStrategy.for('consensus')
14
+ s.should be_a RedisFailover::NodeStrategy::Consensus
15
+ end
16
+
17
+ it 'rejects unknown strategies' do
18
+ expect { NodeStrategy.for('foobar') }.to raise_error(RuntimeError)
19
+ end
20
+ end
21
+ end
22
+ end
@@ -0,0 +1,58 @@
1
+ require 'spec_helper'
2
+
3
+ module RedisFailover
4
+ class LightNodeManager
5
+ def initialize
6
+ @node_states = {}
7
+ end
8
+
9
+ def notify_state(node, state, latency = nil)
10
+ @node_states[node] = state
11
+ end
12
+
13
+ def state_for(node)
14
+ @node_states[node]
15
+ end
16
+ end
17
+
18
+ describe NodeWatcher do
19
+ let(:node_manager) { LightNodeManager.new }
20
+ let(:node) { Node.new(:host => 'host', :port => 123).extend(RedisStubSupport) }
21
+
22
+ describe '#watch' do
23
+ context 'node is not syncing with master' do
24
+ it 'properly informs manager of unavailable node' do
25
+ watcher = NodeWatcher.new(node_manager, node, 1)
26
+ watcher.watch
27
+ sleep(3)
28
+ node.redis.make_unavailable!
29
+ sleep(3)
30
+ watcher.shutdown
31
+ node_manager.state_for(node).should == :unavailable
32
+ end
33
+
34
+ it 'properly informs manager of available node' do
35
+ node_manager.notify_state(node, :unavailable)
36
+ watcher = NodeWatcher.new(node_manager, node, 1)
37
+ watcher.watch
38
+ sleep(3)
39
+ watcher.shutdown
40
+ node_manager.state_for(node).should == :available
41
+ end
42
+ end
43
+
44
+ context 'node is syncing with master' do
45
+ it 'properly informs manager of syncing node' do
46
+ node_manager.notify_state(node, :unavailable)
47
+ node.redis.slaveof('masterhost', 9876)
48
+ node.redis.force_sync_with_master(true)
49
+ watcher = NodeWatcher.new(node_manager, node, 1)
50
+ watcher.watch
51
+ sleep(3)
52
+ watcher.shutdown
53
+ node_manager.state_for(node).should == :available
54
+ end
55
+ end
56
+ end
57
+ end
58
+ end
@@ -0,0 +1,21 @@
1
+ require 'rspec'
2
+ require 'ostruct'
3
+ require 'redis_failover'
4
+
5
+ Dir["#{File.dirname(__FILE__)}/support/**/*.rb"].each { |f| require f }
6
+
7
+ class NullObject
8
+ def method_missing(method, *args, &block)
9
+ yield if block_given?
10
+ self
11
+ end
12
+ end
13
+
14
+ module RedisFailover
15
+ Util.logger = NullObject.new
16
+ end
17
+
18
+ def ZK.new(*args); NullObject.new; end
19
+
20
+ RSpec.configure do |config|
21
+ end
@@ -0,0 +1,15 @@
1
+ :development:
2
+ :nodes:
3
+ - localhost:6379
4
+ - localhost:6389
5
+ :zkservers:
6
+ - localhost:2181
7
+
8
+ :staging:
9
+ :nodes:
10
+ - redis01:6379
11
+ - redis02:6379
12
+ :zkservers:
13
+ - zk01:2181
14
+ - zk02:2181
15
+ - zk03:2181
@@ -0,0 +1,17 @@
1
+ :development:
2
+ :nodes:
3
+ - localhost:6379
4
+ - localhost:6389
5
+ :zkservers:
6
+ - localhost:2181
7
+ :chroot: /with/chroot_development
8
+
9
+ :staging:
10
+ :nodes:
11
+ - redis01:6379
12
+ - redis02:6379
13
+ :zkservers:
14
+ - zk01:2181
15
+ - zk02:2181
16
+ - zk03:2181
17
+ :chroot: /with/chroot_staging
@@ -0,0 +1,7 @@
1
+ :nodes:
2
+ - redis01:6379
3
+ - redis02:6379
4
+ :zkservers:
5
+ - zk01:2181
6
+ - zk02:2181
7
+ - zk03:2181
@@ -0,0 +1,8 @@
1
+ :nodes:
2
+ - redis01:6379
3
+ - redis02:6379
4
+ :zkservers:
5
+ - zk01:2181
6
+ - zk02:2181
7
+ - zk03:2181
8
+ :chroot: /with/chroot
@@ -0,0 +1,87 @@
1
+ module RedisFailover
2
+ class NodeManagerStub < NodeManager
3
+ attr_accessor :master
4
+ # HACK - this will go away once we refactor the tests to use a real ZK/Redis server.
5
+ public :current_nodes, :guess_master
6
+
7
+ def discover_nodes
8
+ # only discover nodes once in testing
9
+ return true if @nodes_discovered
10
+
11
+ master = Node.new(:host => 'master')
12
+ slave = Node.new(:host => 'slave')
13
+ [master, slave].each { |node| node.extend(RedisStubSupport) }
14
+ master.make_master!
15
+ slave.make_slave!(master)
16
+ @nodes = [master, slave]
17
+ @unavailable = []
18
+ @master = master
19
+ @slaves = [slave]
20
+ @failover_strategy = Object.new
21
+ @nodes_discovered = true
22
+ end
23
+
24
+ def setup_zk
25
+ @zk = NullObject.new
26
+ end
27
+
28
+ def slaves
29
+ @slaves
30
+ end
31
+
32
+ def start_processing
33
+ @thread = Thread.new { start }
34
+ sleep(1.5)
35
+ end
36
+
37
+ def stop_processing
38
+ @thread.value
39
+ end
40
+
41
+ def force_unavailable(node)
42
+ start_processing
43
+ node.redis.make_unavailable!
44
+ snapshot = OpenStruct.new(
45
+ :node => node,
46
+ :available_count => 0,
47
+ :unavailable_count => 1,
48
+ :node_managers => ['nm'])
49
+ update_master_state(node, node => snapshot)
50
+ stop_processing
51
+ end
52
+
53
+ def force_available(node)
54
+ start_processing
55
+ node.redis.make_available!
56
+ snapshot = OpenStruct.new(
57
+ :node => node,
58
+ :available_count => 1,
59
+ :unavailable_count => 0,
60
+ :node_managers => ['nm'])
61
+ update_master_state(node, node => snapshot)
62
+ stop_processing
63
+ end
64
+
65
+ def force_syncing(node, serve_stale_reads)
66
+ start_processing
67
+ node.redis.force_sync_with_master(serve_stale_reads)
68
+ snapshot = OpenStruct.new(
69
+ :node => node,
70
+ :available_count => 1,
71
+ :unavailable_count => 0,
72
+ :node_managers => ['nm'])
73
+ update_master_state(node, node => snapshot)
74
+ stop_processing
75
+ end
76
+
77
+ def failover_strategy_candidate(snapshots)
78
+ @slaves.pop
79
+ end
80
+
81
+ def delete_path(*args); end
82
+ def create_path(*args); end
83
+ def write_state(*args); end
84
+ def wait_until_master; end
85
+ def current_node_snapshots; {} end
86
+ end
87
+ end
@@ -0,0 +1,105 @@
1
+ require 'ostruct'
2
+
3
+ module RedisFailover
4
+ # Test stub for Redis.
5
+ class RedisStub
6
+ class Proxy
7
+ def initialize(queue, opts = {})
8
+ @info = {'role' => 'master'}
9
+ @config = {'slave-serve-stale-data' => 'yes'}
10
+ @queue = queue
11
+ end
12
+
13
+ def blpop(*args)
14
+ @queue.pop.tap do |value|
15
+ raise value if value
16
+ end
17
+ end
18
+
19
+ def del(*args)
20
+ end
21
+
22
+ def lpush(*args)
23
+ @queue << nil
24
+ end
25
+
26
+ def slaveof(host, port)
27
+ if host == 'no' && port == 'one'
28
+ @info['role'] = 'master'
29
+ @info.delete('master_host')
30
+ @info.delete('master_port')
31
+ else
32
+ @info['role'] = 'slave'
33
+ @info['master_host'] = host
34
+ @info['master_port'] = port.to_s
35
+ end
36
+ end
37
+
38
+ def info
39
+ @info.dup
40
+ end
41
+
42
+ def change_role_to(role)
43
+ @info['role'] = role
44
+ end
45
+
46
+ def config(action, attribute)
47
+ [action, @config[attribute]]
48
+ end
49
+
50
+ def force_sync_with_master(serve_stale_reads)
51
+ @config['slave-serve-stale-data'] = serve_stale_reads ? 'yes' : 'no'
52
+ @info['master_sync_in_progress'] = '1'
53
+ end
54
+
55
+ def force_sync_done
56
+ @info['master_sync_in_progress'] = '0'
57
+ end
58
+ end
59
+
60
+ attr_reader :host, :port, :available
61
+ def initialize(opts = {})
62
+ @host = opts[:host]
63
+ @port = Integer(opts[:port])
64
+ @queue = Queue.new
65
+ @proxy = Proxy.new(@queue, opts)
66
+ @available = true
67
+ end
68
+
69
+ def method_missing(method, *args, &block)
70
+ if @available
71
+ @proxy.send(method, *args, &block)
72
+ else
73
+ raise Errno::ECONNREFUSED
74
+ end
75
+ end
76
+
77
+ def change_role_to(role)
78
+ @proxy.change_role_to(role)
79
+ end
80
+
81
+ def make_available!
82
+ @available = true
83
+ end
84
+
85
+ def make_unavailable!
86
+ @queue << Errno::ECONNREFUSED
87
+ @available = false
88
+ end
89
+
90
+ def to_s
91
+ "#{@host}:#{@port}"
92
+ end
93
+
94
+ def client
95
+ OpenStruct.new(:host => @host, :port => @port)
96
+ end
97
+ end
98
+
99
+ module RedisStubSupport
100
+ def redis
101
+ @redis ||= RedisStub.new(:host => @host, :port => @port)
102
+ end
103
+ alias_method :new_client, :redis
104
+ end
105
+ end