adhd 0.0.1 → 0.1.0
Sign up to get free protection for your applications and to get access to all the features.
- data/README.rdoc +62 -2
- data/VERSION +1 -1
- data/adhd.gemspec +10 -8
- data/bin/adhd +23 -8
- data/bin/adhd_cleanup +57 -0
- data/lib/adhd/adhd_rest_server.rb +229 -0
- data/lib/adhd/config.yml +1 -1
- data/lib/adhd/models/content_doc.rb +17 -0
- data/lib/adhd/models/content_shard.rb +97 -0
- data/lib/adhd/models/node_doc.rb +139 -0
- data/lib/adhd/models/shard_range.rb +202 -0
- data/lib/adhd/node_manager.rb +260 -0
- data/lib/adhd/reactor.rb +194 -12
- data/test/test_adhd.rb +0 -11
- metadata +11 -7
- data/lib/adhd.rb +0 -120
- data/lib/adhd/models.rb +0 -388
- data/lib/adhd/node.rb +0 -13
- data/models.rb +0 -19
data/lib/adhd/config.yml
CHANGED
@@ -0,0 +1,17 @@
|
|
1
|
+
class ContentDoc < CouchRest::ExtendedDocument
|
2
|
+
# NOTE: NO DEFAULT DATABASE IN THE OBJECT -- WE WILL BE STORING A LOT OF
|
3
|
+
# DATABASES OF THIS TYPE.
|
4
|
+
|
5
|
+
|
6
|
+
property :_id
|
7
|
+
property :internal_id
|
8
|
+
property :size_bytes
|
9
|
+
property :filename
|
10
|
+
property :mime_type
|
11
|
+
|
12
|
+
view_by :internal_id
|
13
|
+
|
14
|
+
# A special attachment "File" is expected to exist
|
15
|
+
|
16
|
+
end
|
17
|
+
|
@@ -0,0 +1,97 @@
|
|
1
|
+
|
2
|
+
|
3
|
+
class ContentShard
|
4
|
+
attr_accessor :nodes, :this_shard, :our_node, :this_shard_db
|
5
|
+
|
6
|
+
def initialize(nodesv, this_shardv)
|
7
|
+
@nodes = nodesv
|
8
|
+
@this_shard = this_shardv
|
9
|
+
|
10
|
+
# Work out the rest
|
11
|
+
@our_node = nodesv.our_node
|
12
|
+
@this_shard_db = nodesv.our_node.get_content_db(this_shardv.shard_db_name)
|
13
|
+
|
14
|
+
@last_sync_seq = 0 # @this_shard_db.info['update_seq']
|
15
|
+
end
|
16
|
+
|
17
|
+
def in_shard?(internal_id)
|
18
|
+
internal_id >= this_shard.range_start && internal_id < this_shard.range_end
|
19
|
+
end
|
20
|
+
|
21
|
+
def write_doc(content_doc)
|
22
|
+
# Write a content document to this shard
|
23
|
+
# Make sure it is in this shard
|
24
|
+
if in_shard? content_doc.internal_id
|
25
|
+
this_shard_db.save_doc(content_doc)
|
26
|
+
end
|
27
|
+
end
|
28
|
+
|
29
|
+
def sync
|
30
|
+
# A Shard only pushes with the master of the shard
|
31
|
+
# or the node with the highest is_storage value alive
|
32
|
+
# Shard masters ensure changes are pushed to all
|
33
|
+
|
34
|
+
# NOTE: This method needs serious refactoring
|
35
|
+
# No need to update
|
36
|
+
return false if @this_shard_db.info['update_seq'] == @last_sync_seq
|
37
|
+
|
38
|
+
# Are we the shard master?
|
39
|
+
am_master = (our_node.name == this_shard.master_node)
|
40
|
+
|
41
|
+
if !am_master
|
42
|
+
master_node = Node.by_name(:key => this_shard.master_node).first
|
43
|
+
remote_db = master_node.get_content_db(this_shard.shard_db_name)
|
44
|
+
bool_to = @our_node.replicate_to(this_shard_db, master_node, remote_db)
|
45
|
+
if bool_to
|
46
|
+
@last_sync_seq = @this_shard_db.info['update_seq']
|
47
|
+
return true
|
48
|
+
end
|
49
|
+
end
|
50
|
+
|
51
|
+
# Either we are the master or the master has failed -- we replicate with
|
52
|
+
# all nodes or the first available aside us and master
|
53
|
+
all_good = true
|
54
|
+
this_shard.node_list.each do |node_name|
|
55
|
+
# Push all changes to the other nodes
|
56
|
+
remote_node = Node.by_name(:key => node_name).first
|
57
|
+
remote_db = remote_node.get_content_db(this_shard.shard_db_name)
|
58
|
+
all_good &= @our_node.replicate_to(this_shard_db, remote_node, remote_db)
|
59
|
+
|
60
|
+
if !am_master && bool_to
|
61
|
+
# NOTE: How to build skynet, Note 2
|
62
|
+
# We are doing some "gonzo" replication, here. Our master is
|
63
|
+
# clearly down so we find the second best node; we push our
|
64
|
+
# changes to this node, and now also *replicate from*
|
65
|
+
# that node.
|
66
|
+
@our_node.replicate_from(this_shard_db, remote_node, remote_db)
|
67
|
+
@last_sync_seq = @this_shard_db.info['update_seq']
|
68
|
+
break
|
69
|
+
end
|
70
|
+
end
|
71
|
+
if all_good
|
72
|
+
@last_sync_seq = @this_shard_db.info['update_seq']
|
73
|
+
return true
|
74
|
+
else
|
75
|
+
return false
|
76
|
+
end
|
77
|
+
end
|
78
|
+
|
79
|
+
end
|
80
|
+
|
81
|
+
class ContentDoc < CouchRest::ExtendedDocument
|
82
|
+
# NOTE: NO DEFAULT DATABASE IN THE OBJECT -- WE WILL BE STORING A LOT OF
|
83
|
+
# DATABASES OF THIS TYPE.
|
84
|
+
|
85
|
+
|
86
|
+
property :_id
|
87
|
+
property :internal_id
|
88
|
+
property :size_bytes
|
89
|
+
property :filename
|
90
|
+
property :mime_type
|
91
|
+
|
92
|
+
view_by :internal_id
|
93
|
+
|
94
|
+
# A special attachment "File" is expected to exist
|
95
|
+
|
96
|
+
end
|
97
|
+
|
@@ -0,0 +1,139 @@
|
|
1
|
+
## Key Restrictions ok internal_IDs: must only contain [a-z0-9-]
|
2
|
+
|
3
|
+
class NodeDB
|
4
|
+
|
5
|
+
attr_accessor :local_node_db, :our_node
|
6
|
+
|
7
|
+
def initialize(our_nodev)
|
8
|
+
@our_node = our_nodev
|
9
|
+
|
10
|
+
# Get the address of the CDB from the node
|
11
|
+
@local_node_db = our_nodev.get_node_db
|
12
|
+
end
|
13
|
+
|
14
|
+
def sync
|
15
|
+
# We replicate our state to the management node(s)
|
16
|
+
management_nodes = Node.by_is_management.reverse
|
17
|
+
# NOTE: randomize the order for load balancing here
|
18
|
+
|
19
|
+
# NOTE2: How to build skynet (TODO)
|
20
|
+
# -------------------
|
21
|
+
# If length of management is zero, then chose 3 different random
|
22
|
+
# nodes at each sync, and sync with them in node_name order.
|
23
|
+
# This guarantees that any updates on nodes are communicated in
|
24
|
+
# O(log N) ephocs, at the cost of O(3 * N) connections per epoch.
|
25
|
+
# It also guarantees any new management servers are discovered in
|
26
|
+
# this O(log N) time, creating "jelly fish" or "partition proof"
|
27
|
+
# availability.
|
28
|
+
|
29
|
+
management_nodes.each do |mng_node|
|
30
|
+
remote_db = mng_node.get_node_db
|
31
|
+
bool_from = @our_node.replicate_from(local_node_db, mng_node, remote_db)
|
32
|
+
bool_to = @our_node.replicate_to(local_node_db, mng_node, remote_db)
|
33
|
+
if bool_from && bool_to && !our_node.is_management
|
34
|
+
#puts "Pushed to management"
|
35
|
+
break
|
36
|
+
end
|
37
|
+
#puts "Did not push to management"
|
38
|
+
end
|
39
|
+
end
|
40
|
+
|
41
|
+
def available_node_list
|
42
|
+
# Returns all nodes marked as available
|
43
|
+
all_nodes = Node.by_name
|
44
|
+
return all_nodes.select {|node| node.status == "RUNNING"}
|
45
|
+
end
|
46
|
+
|
47
|
+
def head_management_node
|
48
|
+
management_nodes = Node.by_is_management.reverse
|
49
|
+
hmn = management_nodes.find {|node| node.status == "RUNNING"}
|
50
|
+
return hmn
|
51
|
+
end
|
52
|
+
|
53
|
+
end
|
54
|
+
|
55
|
+
class Node < CouchRest::ExtendedDocument
|
56
|
+
#NODESERVER = CouchRest.new("#{ARGV[1]}")
|
57
|
+
#NODESERVER.default_database = "#{ARGV[0]}_node_db"
|
58
|
+
|
59
|
+
#use_database NODESERVER.default_database
|
60
|
+
|
61
|
+
unique_id :name
|
62
|
+
|
63
|
+
property :name
|
64
|
+
property :url
|
65
|
+
property :is_store
|
66
|
+
property :is_management
|
67
|
+
property :is_directory
|
68
|
+
property :status
|
69
|
+
|
70
|
+
timestamps!
|
71
|
+
|
72
|
+
view_by :name
|
73
|
+
view_by :is_management
|
74
|
+
|
75
|
+
def get_node_db
|
76
|
+
server = CouchRest.new("#{url}")
|
77
|
+
db = server.database!("#{name}_node_db")
|
78
|
+
# puts "Open db #{db}"
|
79
|
+
db
|
80
|
+
end
|
81
|
+
|
82
|
+
def get_shard_db
|
83
|
+
server = CouchRest.new("#{url}")
|
84
|
+
db = server.database!("#{name}_shard_db")
|
85
|
+
# puts "Open db #{db}"
|
86
|
+
db
|
87
|
+
end
|
88
|
+
|
89
|
+
def get_content_db(shard_db_name)
|
90
|
+
server = CouchRest.new("#{url}")
|
91
|
+
db = server.database!("#{name}_#{shard_db_name}_content_db")
|
92
|
+
# puts "Open db #{db}"
|
93
|
+
db
|
94
|
+
end
|
95
|
+
|
96
|
+
# Replicating databases and marking nodes as unavailable
|
97
|
+
# In the future we should hook these into a "replication manager"
|
98
|
+
# for databases. The manager should set up continuous replication across
|
99
|
+
# databases, and only do a replication after some time lapses.
|
100
|
+
|
101
|
+
def replicate_to(local_db, other_node, remote_db)
|
102
|
+
# Do not try to contact unavailable nodes
|
103
|
+
return false if other_node.status == "UNAVAILABLE"
|
104
|
+
# No point replicating to ourselves
|
105
|
+
return false if (name == other_node.name)
|
106
|
+
|
107
|
+
begin
|
108
|
+
# Replicate to other node is possible
|
109
|
+
local_db.replicate_to(remote_db)
|
110
|
+
return true
|
111
|
+
rescue Exception => e
|
112
|
+
# Other node turns out to be unavailable
|
113
|
+
other_node.status = "UNAVAILABLE"
|
114
|
+
other_node.save
|
115
|
+
return false
|
116
|
+
end
|
117
|
+
end
|
118
|
+
|
119
|
+
def replicate_from(local_db, other_node, remote_db)
|
120
|
+
# Do not try to contact unavailable nodes
|
121
|
+
return false if other_node.status == "UNAVAILABLE"
|
122
|
+
# No point replicating to ourselves
|
123
|
+
return false if (name == other_node.name)
|
124
|
+
|
125
|
+
begin
|
126
|
+
# Replicate to other node is possible
|
127
|
+
local_db.replicate_from(remote_db)
|
128
|
+
return true
|
129
|
+
rescue Exception => e
|
130
|
+
# Other node turns out to be unavailable
|
131
|
+
other_node.status = "UNAVAILABLE"
|
132
|
+
other_node.save
|
133
|
+
return false
|
134
|
+
end
|
135
|
+
|
136
|
+
end
|
137
|
+
|
138
|
+
end
|
139
|
+
|
@@ -0,0 +1,202 @@
|
|
1
|
+
class ShardRangeDB
|
2
|
+
|
3
|
+
attr_accessor :nodes, :local_shard_db, :our_node
|
4
|
+
|
5
|
+
def initialize(nodesv)
|
6
|
+
@nodes = nodesv
|
7
|
+
|
8
|
+
# Automatically get our shard_db address from our own node name
|
9
|
+
@our_node = nodesv.our_node
|
10
|
+
@local_shard_db = nodesv.our_node.get_shard_db
|
11
|
+
|
12
|
+
puts "Assign default database for shard ranges (#{@local_shard_db})"
|
13
|
+
ShardRange.use_database @local_shard_db
|
14
|
+
end
|
15
|
+
|
16
|
+
def sync
|
17
|
+
# We replicate our state from the management node(s)
|
18
|
+
# We never push content if we are only storage
|
19
|
+
management_nodes = Node.by_is_management.reverse
|
20
|
+
|
21
|
+
# NOTE: randomize the order for load balancing here
|
22
|
+
|
23
|
+
management_nodes.each do |mng_node|
|
24
|
+
remote_db = mng_node.get_shard_db
|
25
|
+
bool_from = @our_node.replicate_from(local_shard_db, mng_node, remote_db)
|
26
|
+
if our_node.is_management
|
27
|
+
# Push any changes to other management nodes
|
28
|
+
bool_to = @our_node.replicate_to(local_shard_db, mng_node, remote_db)
|
29
|
+
end
|
30
|
+
break if bool_from && !our_node.is_management
|
31
|
+
end
|
32
|
+
end
|
33
|
+
|
34
|
+
def build_shards(number)
|
35
|
+
# Make a large list of possible id boundaries
|
36
|
+
characters = []
|
37
|
+
("0".."9").each do |c|
|
38
|
+
characters << c
|
39
|
+
end
|
40
|
+
("a".."f").each do |c|
|
41
|
+
characters << c
|
42
|
+
end
|
43
|
+
|
44
|
+
# Generate 36 x 36 keys to choose boundaries from
|
45
|
+
all_keys = []
|
46
|
+
characters.each do |c1|
|
47
|
+
characters.each do |c2|
|
48
|
+
characters.each do |c3|
|
49
|
+
all_keys << (c1+c2+c3)
|
50
|
+
end
|
51
|
+
end
|
52
|
+
end
|
53
|
+
|
54
|
+
# Now chose our boundaries
|
55
|
+
num_range_keys = all_keys.length
|
56
|
+
approx_shard_size = (num_range_keys * 1.0) / number
|
57
|
+
|
58
|
+
shard_starts = []
|
59
|
+
(0...number).each do |n|
|
60
|
+
shard_starts << (all_keys[(n * approx_shard_size).floor])
|
61
|
+
end
|
62
|
+
|
63
|
+
shard_ends = shard_starts.clone
|
64
|
+
shard_ends << ("z" * 3)
|
65
|
+
shard_ends.delete_at(0)
|
66
|
+
|
67
|
+
# Finally build them!
|
68
|
+
puts "Build Shards"
|
69
|
+
(0...number).each do |n|
|
70
|
+
puts "Shard #{n}: from #{shard_starts[n]} to #{shard_ends[n]}"
|
71
|
+
shard_name = "sh_#{shard_starts[n]}_to_#{shard_ends[n]}"
|
72
|
+
sr = ShardRange.new
|
73
|
+
sr.range_start = shard_starts[n]
|
74
|
+
sr.range_end = shard_ends[n]
|
75
|
+
sr.shard_db_name = shard_name
|
76
|
+
sr.save
|
77
|
+
end
|
78
|
+
end
|
79
|
+
|
80
|
+
def get_shard(internal_id)
|
81
|
+
# Finds the list of shards within which this ID lives
|
82
|
+
all_shards = ShardRange.by_range_start
|
83
|
+
selected_shards = []
|
84
|
+
all_shards.each do |a_shard| # TODO: linear search is inefficient -- create a view
|
85
|
+
if a_shard.range_start <= internal_id && a_shard.range_end > internal_id
|
86
|
+
selected_shards << a_shard
|
87
|
+
end
|
88
|
+
end
|
89
|
+
selected_shards
|
90
|
+
end
|
91
|
+
|
92
|
+
def get_content_shards
|
93
|
+
# Return the content_shards of our node
|
94
|
+
content_shards = {}
|
95
|
+
ShardRange.by_node(:key => our_node.name).each do |s|
|
96
|
+
|
97
|
+
# Build a content shard object
|
98
|
+
content_shards[s.shard_db_name] = ContentShard.new(nodes, s)
|
99
|
+
end
|
100
|
+
puts "Content shards #{content_shards.length}"
|
101
|
+
content_shards
|
102
|
+
end
|
103
|
+
|
104
|
+
def write_doc_directly(content_doc)
|
105
|
+
# Write a document directly to a nodes content repository
|
106
|
+
success = {:ok => false , :reason => "No available node found"}
|
107
|
+
doc_shard = get_shard(content_doc.internal_id).first
|
108
|
+
doc_shard.get_nodes.each do |node|
|
109
|
+
# Try to write the doc to this node
|
110
|
+
begin
|
111
|
+
remote_node = Node.by_name(:key => node).first
|
112
|
+
remote_ndb = NodeDB.new(remote_node)
|
113
|
+
remote_content_shard = ContentShard.new(remote_ndb, doc_shard)
|
114
|
+
remote_content_shard.this_shard_db.save_doc(content_doc)
|
115
|
+
success = {:ok => true, :doc => content_doc, :db => remote_content_shard.this_shard_db}
|
116
|
+
break
|
117
|
+
rescue RestClient::RequestFailed => rf
|
118
|
+
if rf.http_code == 409
|
119
|
+
puts "Document already there"
|
120
|
+
return {:ok => false , :reason => "Document already in database"}
|
121
|
+
end
|
122
|
+
rescue Exception =>e
|
123
|
+
puts "Could not put doc in node #{node} because of #{rf}"
|
124
|
+
# TODO: change status or chose another management server
|
125
|
+
remote_node.status = "UNAVAILABLE"
|
126
|
+
remote_node.save
|
127
|
+
end
|
128
|
+
end
|
129
|
+
return success
|
130
|
+
end
|
131
|
+
|
132
|
+
def get_doc_directly(internal_id)
|
133
|
+
# Write a document directly to a nodes content repository
|
134
|
+
doc_shard = get_shard(internal_id).first
|
135
|
+
|
136
|
+
# TODO: Randomize the order of nodes for load balancing in retrieval!
|
137
|
+
docx = []
|
138
|
+
doc_shard.get_nodes.each do |node|
|
139
|
+
# Try to write the doc to this node
|
140
|
+
begin
|
141
|
+
remote_node = Node.by_name(:key => node).first
|
142
|
+
remote_ndb = NodeDB.new(remote_node)
|
143
|
+
remote_content_shard = ContentShard.new(remote_ndb, doc_shard)
|
144
|
+
|
145
|
+
docx = ContentDoc.by_internal_id(:key => internal_id, :database => remote_content_shard.this_shard_db)
|
146
|
+
if docx.length > 0
|
147
|
+
return {:ok => true, :doc => docx.first, :db => remote_content_shard.this_shard_db }
|
148
|
+
end
|
149
|
+
rescue
|
150
|
+
puts "Could not put doc in node #{node.name}"
|
151
|
+
# TODO: change status or chose another management server
|
152
|
+
remote_node.status = "UNAVAILABLE"
|
153
|
+
remote_node.save
|
154
|
+
end
|
155
|
+
end
|
156
|
+
return {:ok => false }
|
157
|
+
end
|
158
|
+
|
159
|
+
end
|
160
|
+
|
161
|
+
class ShardRange < CouchRest::ExtendedDocument
|
162
|
+
# SHARDSERVER = CouchRest.new("#{ARGV[1]}")
|
163
|
+
# SHARDSERVER.default_database = "#{ARGV[0]}_shard_db"
|
164
|
+
# use_database SHARDSERVER.default_database
|
165
|
+
|
166
|
+
property :range_start
|
167
|
+
property :range_end
|
168
|
+
property :node_list
|
169
|
+
property :master_node
|
170
|
+
property :shard_db_name
|
171
|
+
|
172
|
+
view_by :range_start
|
173
|
+
|
174
|
+
# View "node" - given a node returns the shards watched
|
175
|
+
# How to use this new
|
176
|
+
#
|
177
|
+
# puts "Which Shards does 'node1' watch?"
|
178
|
+
# ShardRange.by_node(:key => "node1").each do |s|
|
179
|
+
# puts "Shard: #{s.shard_db_name}"
|
180
|
+
# end
|
181
|
+
|
182
|
+
|
183
|
+
view_by :node,
|
184
|
+
:map =>
|
185
|
+
"function(doc) {
|
186
|
+
if (doc['couchrest-type'] == 'ShardRange' && doc.node_list) {
|
187
|
+
doc.node_list.forEach(function(node){
|
188
|
+
emit(node, 1);
|
189
|
+
});
|
190
|
+
}
|
191
|
+
}"
|
192
|
+
|
193
|
+
def get_nodes
|
194
|
+
# Return all nodes, with the master being first
|
195
|
+
all_nodes = node_list.clone
|
196
|
+
all_nodes.delete(master_node)
|
197
|
+
all_nodes = [master_node] + all_nodes
|
198
|
+
all_nodes
|
199
|
+
end
|
200
|
+
|
201
|
+
end
|
202
|
+
|