1 | """ |
---|
2 | Algorithms for figuring out happiness, the number of unique nodes the data is |
---|
3 | on. |
---|
4 | |
---|
5 | Ported to Python 3. |
---|
6 | """ |
---|
7 | |
---|
8 | from queue import PriorityQueue |
---|
9 | |
---|
10 | |
---|
11 | def augmenting_path_for(graph): |
---|
12 | """ |
---|
13 | I return an augmenting path, if there is one, from the source node |
---|
14 | to the sink node in the flow network represented by my graph argument. |
---|
15 | If there is no augmenting path, I return False. I assume that the |
---|
16 | source node is at index 0 of graph, and the sink node is at the last |
---|
17 | index. I also assume that graph is a flow network in adjacency list |
---|
18 | form. |
---|
19 | """ |
---|
20 | bfs_tree = bfs(graph, 0) |
---|
21 | if bfs_tree[len(graph) - 1]: |
---|
22 | n = len(graph) - 1 |
---|
23 | path = [] # [(u, v)], where u and v are vertices in the graph |
---|
24 | while n != 0: |
---|
25 | path.insert(0, (bfs_tree[n], n)) |
---|
26 | n = bfs_tree[n] |
---|
27 | return path |
---|
28 | return False |
---|
29 | |
---|
30 | def bfs(graph, s): |
---|
31 | """ |
---|
32 | Perform a BFS on graph starting at s, where graph is a graph in |
---|
33 | adjacency list form, and s is a node in graph. I return the |
---|
34 | predecessor table that the BFS generates. |
---|
35 | """ |
---|
36 | # This is an adaptation of the BFS described in "Introduction to |
---|
37 | # Algorithms", Cormen et al, 2nd ed., p. 532. |
---|
38 | # WHITE vertices are those that we haven't seen or explored yet. |
---|
39 | WHITE = 0 |
---|
40 | # GRAY vertices are those we have seen, but haven't explored yet |
---|
41 | GRAY = 1 |
---|
42 | # BLACK vertices are those we have seen and explored |
---|
43 | BLACK = 2 |
---|
44 | color = [WHITE for i in range(len(graph))] |
---|
45 | predecessor = [None for i in range(len(graph))] |
---|
46 | distance = [-1 for i in range(len(graph))] |
---|
47 | queue = [s] # vertices that we haven't explored yet. |
---|
48 | color[s] = GRAY |
---|
49 | distance[s] = 0 |
---|
50 | while queue: |
---|
51 | n = queue.pop(0) |
---|
52 | for v in graph[n]: |
---|
53 | if color[v] == WHITE: |
---|
54 | color[v] = GRAY |
---|
55 | distance[v] = distance[n] + 1 |
---|
56 | predecessor[v] = n |
---|
57 | queue.append(v) |
---|
58 | color[n] = BLACK |
---|
59 | return predecessor |
---|
60 | |
---|
61 | def residual_network(graph, f): |
---|
62 | """ |
---|
63 | I return the residual network and residual capacity function of the |
---|
64 | flow network represented by my graph and f arguments. graph is a |
---|
65 | flow network in adjacency-list form, and f is a flow in graph. |
---|
66 | """ |
---|
67 | new_graph = [[] for i in range(len(graph))] |
---|
68 | cf = [[0 for s in range(len(graph))] for sh in range(len(graph))] |
---|
69 | for i in range(len(graph)): |
---|
70 | for v in graph[i]: |
---|
71 | if f[i][v] == 1: |
---|
72 | # We add an edge (v, i) with cf[v,i] = 1. This means |
---|
73 | # that we can remove 1 unit of flow from the edge (i, v) |
---|
74 | new_graph[v].append(i) |
---|
75 | cf[v][i] = 1 |
---|
76 | cf[i][v] = -1 |
---|
77 | else: |
---|
78 | # We add the edge (i, v), since we're not using it right |
---|
79 | # now. |
---|
80 | new_graph[i].append(v) |
---|
81 | cf[i][v] = 1 |
---|
82 | cf[v][i] = -1 |
---|
83 | return (new_graph, cf) |
---|
84 | |
---|
85 | |
---|
86 | def calculate_happiness(mappings): |
---|
87 | """ |
---|
88 | :param mappings: a dict mapping 'share' -> 'peer' |
---|
89 | |
---|
90 | :returns: the happiness, which is the number of unique peers we've |
---|
91 | placed shares on. |
---|
92 | """ |
---|
93 | unique_peers = set(mappings.values()) |
---|
94 | assert None not in unique_peers |
---|
95 | return len(unique_peers) |
---|
96 | |
---|
97 | |
---|
98 | def _calculate_mappings(peers, shares, servermap=None): |
---|
99 | """ |
---|
100 | Given a set of peers, a set of shares, and a dictionary of server -> |
---|
101 | set(shares), determine how the uploader should allocate shares. If a |
---|
102 | servermap is supplied, determine which existing allocations should be |
---|
103 | preserved. If servermap is None, calculate the maximum matching of the |
---|
104 | bipartite graph (U, V, E) such that: |
---|
105 | |
---|
106 | U = peers |
---|
107 | V = shares |
---|
108 | E = peers x shares |
---|
109 | |
---|
110 | Returns a dictionary {share -> set(peer)}, indicating that the share |
---|
111 | should be placed on each peer in the set. If a share's corresponding |
---|
112 | value is None, the share can be placed on any server. Note that the set |
---|
113 | of peers should only be one peer when returned, but it is possible to |
---|
114 | duplicate shares by adding additional servers to the set. |
---|
115 | """ |
---|
116 | peer_to_index, index_to_peer = _reindex(peers, 1) |
---|
117 | share_to_index, index_to_share = _reindex(shares, len(peers) + 1) |
---|
118 | shareIndices = [share_to_index[s] for s in shares] |
---|
119 | if servermap: |
---|
120 | graph = _servermap_flow_graph(peers, shares, servermap) |
---|
121 | else: |
---|
122 | peerIndices = [peer_to_index[peer] for peer in peers] |
---|
123 | graph = _flow_network(peerIndices, shareIndices) |
---|
124 | max_graph = _compute_maximum_graph(graph, shareIndices) |
---|
125 | return _convert_mappings(index_to_peer, index_to_share, max_graph) |
---|
126 | |
---|
127 | |
---|
128 | def _compute_maximum_graph(graph, shareIndices): |
---|
129 | """ |
---|
130 | This is an implementation of the Ford-Fulkerson method for finding |
---|
131 | a maximum flow in a flow network applied to a bipartite graph. |
---|
132 | Specifically, it is the Edmonds-Karp algorithm, since it uses a |
---|
133 | BFS to find the shortest augmenting path at each iteration, if one |
---|
134 | exists. |
---|
135 | |
---|
136 | The implementation here is an adapation of an algorithm described in |
---|
137 | "Introduction to Algorithms", Cormen et al, 2nd ed., pp 658-662. |
---|
138 | """ |
---|
139 | |
---|
140 | if graph == []: |
---|
141 | return {} |
---|
142 | |
---|
143 | dim = len(graph) |
---|
144 | flow_function = [[0 for sh in range(dim)] for s in range(dim)] |
---|
145 | residual_graph, residual_function = residual_network(graph, flow_function) |
---|
146 | |
---|
147 | while augmenting_path_for(residual_graph): |
---|
148 | path = augmenting_path_for(residual_graph) |
---|
149 | # Delta is the largest amount that we can increase flow across |
---|
150 | # all of the edges in path. Because of the way that the residual |
---|
151 | # function is constructed, f[u][v] for a particular edge (u, v) |
---|
152 | # is the amount of unused capacity on that edge. Taking the |
---|
153 | # minimum of a list of those values for each edge in the |
---|
154 | # augmenting path gives us our delta. |
---|
155 | delta = min(residual_function[u][v] for (u, v) in path) |
---|
156 | for (u, v) in path: |
---|
157 | flow_function[u][v] += delta |
---|
158 | flow_function[v][u] -= delta |
---|
159 | residual_graph, residual_function = residual_network(graph,flow_function) |
---|
160 | |
---|
161 | new_mappings = {} |
---|
162 | for shareIndex in shareIndices: |
---|
163 | peer = residual_graph[shareIndex] |
---|
164 | if peer == [dim - 1]: |
---|
165 | new_mappings.setdefault(shareIndex, None) |
---|
166 | else: |
---|
167 | new_mappings.setdefault(shareIndex, peer[0]) |
---|
168 | |
---|
169 | return new_mappings |
---|
170 | |
---|
171 | |
---|
172 | def _extract_ids(mappings): |
---|
173 | shares = set() |
---|
174 | peers = set() |
---|
175 | for share in mappings: |
---|
176 | if mappings[share] == None: |
---|
177 | pass |
---|
178 | else: |
---|
179 | shares.add(share) |
---|
180 | for item in mappings[share]: |
---|
181 | peers.add(item) |
---|
182 | return (peers, shares) |
---|
183 | |
---|
184 | def _distribute_homeless_shares(mappings, homeless_shares, peers_to_shares): |
---|
185 | """ |
---|
186 | Shares which are not mapped to a peer in the maximum spanning graph |
---|
187 | still need to be placed on a server. This function attempts to |
---|
188 | distribute those homeless shares as evenly as possible over the |
---|
189 | available peers. If possible a share will be placed on the server it was |
---|
190 | originally on, signifying the lease should be renewed instead. |
---|
191 | """ |
---|
192 | #print("mappings, homeless_shares, peers_to_shares %s %s %s" % (mappings, homeless_shares, peers_to_shares)) |
---|
193 | servermap_peerids = set([key for key in peers_to_shares]) |
---|
194 | servermap_shareids = set() |
---|
195 | for key in sorted(peers_to_shares.keys()): |
---|
196 | # XXX maybe sort? |
---|
197 | for share in peers_to_shares[key]: |
---|
198 | servermap_shareids.add(share) |
---|
199 | |
---|
200 | # First check to see if the leases can be renewed. |
---|
201 | to_distribute = set() |
---|
202 | for share in homeless_shares: |
---|
203 | if share in servermap_shareids: |
---|
204 | for peerid in peers_to_shares: |
---|
205 | if share in peers_to_shares[peerid]: |
---|
206 | mappings[share] = set([peerid]) |
---|
207 | break |
---|
208 | else: |
---|
209 | to_distribute.add(share) |
---|
210 | # This builds a priority queue of peers with the number of shares |
---|
211 | # each peer holds as the priority. |
---|
212 | priority = {} |
---|
213 | pQueue = PriorityQueue() |
---|
214 | for peerid in servermap_peerids: |
---|
215 | priority.setdefault(peerid, 0) |
---|
216 | for share in mappings: |
---|
217 | if mappings[share] is not None: |
---|
218 | for peer in mappings[share]: |
---|
219 | if peer in servermap_peerids: |
---|
220 | priority[peer] += 1 |
---|
221 | if priority == {}: |
---|
222 | return |
---|
223 | for peerid in priority: |
---|
224 | pQueue.put((priority[peerid], peerid)) |
---|
225 | # Distribute the shares to peers with the lowest priority. |
---|
226 | for share in to_distribute: |
---|
227 | peer = pQueue.get() |
---|
228 | mappings[share] = set([peer[1]]) |
---|
229 | pQueue.put((peer[0]+1, peer[1])) |
---|
230 | |
---|
231 | def _convert_mappings(index_to_peer, index_to_share, maximum_graph): |
---|
232 | """ |
---|
233 | Now that a maximum spanning graph has been found, convert the indexes |
---|
234 | back to their original ids so that the client can pass them to the |
---|
235 | uploader. |
---|
236 | """ |
---|
237 | |
---|
238 | converted_mappings = {} |
---|
239 | for share in maximum_graph: |
---|
240 | peer = maximum_graph[share] |
---|
241 | if peer == None: |
---|
242 | converted_mappings.setdefault(index_to_share[share], None) |
---|
243 | else: |
---|
244 | converted_mappings.setdefault(index_to_share[share], set([index_to_peer[peer]])) |
---|
245 | return converted_mappings |
---|
246 | |
---|
247 | |
---|
248 | def _servermap_flow_graph(peers, shares, servermap): |
---|
249 | """ |
---|
250 | Generates a flow network of peerIndices to shareIndices from a server map |
---|
251 | of 'peer' -> ['shares']. According to Wikipedia, "a flow network is a |
---|
252 | directed graph where each edge has a capacity and each edge receives a flow. |
---|
253 | The amount of flow on an edge cannot exceed the capacity of the edge." This |
---|
254 | is necessary because in order to find the maximum spanning, the Edmonds-Karp algorithm |
---|
255 | converts the problem into a maximum flow problem. |
---|
256 | """ |
---|
257 | if servermap == {}: |
---|
258 | return [] |
---|
259 | |
---|
260 | peer_to_index, index_to_peer = _reindex(peers, 1) |
---|
261 | share_to_index, index_to_share = _reindex(shares, len(peers) + 1) |
---|
262 | graph = [] |
---|
263 | indexedShares = [] |
---|
264 | sink_num = len(peers) + len(shares) + 1 |
---|
265 | graph.append([peer_to_index[peer] for peer in peers]) |
---|
266 | #print("share_to_index %s" % share_to_index) |
---|
267 | #print("servermap %s" % servermap) |
---|
268 | for peer in peers: |
---|
269 | if peer in servermap: |
---|
270 | for s in servermap[peer]: |
---|
271 | if s in share_to_index: |
---|
272 | indexedShares.append(share_to_index[s]) |
---|
273 | graph.insert(peer_to_index[peer], indexedShares) |
---|
274 | for share in shares: |
---|
275 | graph.insert(share_to_index[share], [sink_num]) |
---|
276 | graph.append([]) |
---|
277 | return graph |
---|
278 | |
---|
279 | |
---|
280 | def _reindex(items, base): |
---|
281 | """ |
---|
282 | I take an iteratble of items and give each item an index to be used in |
---|
283 | the construction of a flow network. Indices for these items start at base |
---|
284 | and continue to base + len(items) - 1. |
---|
285 | |
---|
286 | I return two dictionaries: ({item: index}, {index: item}) |
---|
287 | """ |
---|
288 | item_to_index = {} |
---|
289 | index_to_item = {} |
---|
290 | for item in items: |
---|
291 | item_to_index.setdefault(item, base) |
---|
292 | index_to_item.setdefault(base, item) |
---|
293 | base += 1 |
---|
294 | return (item_to_index, index_to_item) |
---|
295 | |
---|
296 | |
---|
297 | def _flow_network(peerIndices, shareIndices): |
---|
298 | """ |
---|
299 | Given set of peerIndices and a set of shareIndices, I create a flow network |
---|
300 | to be used by _compute_maximum_graph. The return value is a two |
---|
301 | dimensional list in the form of a flow network, where each index represents |
---|
302 | a node, and the corresponding list represents all of the nodes it is connected |
---|
303 | to. |
---|
304 | |
---|
305 | This function is similar to allmydata.util.happinessutil.flow_network_for, but |
---|
306 | we connect every peer with all shares instead of reflecting a supplied servermap. |
---|
307 | """ |
---|
308 | graph = [] |
---|
309 | # The first entry in our flow network is the source. |
---|
310 | # Connect the source to every server. |
---|
311 | graph.append(peerIndices) |
---|
312 | sink_num = len(peerIndices + shareIndices) + 1 |
---|
313 | # Connect every server with every share it can possibly store. |
---|
314 | for peerIndex in peerIndices: |
---|
315 | graph.insert(peerIndex, shareIndices) |
---|
316 | # Connect every share with the sink. |
---|
317 | for shareIndex in shareIndices: |
---|
318 | graph.insert(shareIndex, [sink_num]) |
---|
319 | # Add an empty entry for the sink. |
---|
320 | graph.append([]) |
---|
321 | return graph |
---|
322 | |
---|
323 | def share_placement(peers, readonly_peers, shares, peers_to_shares): |
---|
324 | """ |
---|
325 | Generates the allocations the upload should based on the given |
---|
326 | information. We construct a dictionary of 'share_num' -> |
---|
327 | 'server_id' and return it to the caller. Existing allocations |
---|
328 | appear as placements because attempting to place an existing |
---|
329 | allocation will renew the share. |
---|
330 | |
---|
331 | For more information on the algorithm this class implements, refer to |
---|
332 | docs/specifications/servers-of-happiness.rst |
---|
333 | """ |
---|
334 | if not peers: |
---|
335 | return dict() |
---|
336 | |
---|
337 | # First calculate share placement for the readonly servers. |
---|
338 | readonly_shares = set() |
---|
339 | readonly_map = {} |
---|
340 | for peer in sorted(peers_to_shares.keys()): |
---|
341 | if peer in readonly_peers: |
---|
342 | readonly_map.setdefault(peer, peers_to_shares[peer]) |
---|
343 | for share in peers_to_shares[peer]: |
---|
344 | readonly_shares.add(share) |
---|
345 | |
---|
346 | readonly_mappings = _calculate_mappings(readonly_peers, readonly_shares, readonly_map) |
---|
347 | used_peers, used_shares = _extract_ids(readonly_mappings) |
---|
348 | |
---|
349 | # Calculate share placement for the remaining existing allocations |
---|
350 | new_peers = set(peers) - used_peers |
---|
351 | # Squash a list of sets into one set |
---|
352 | new_shares = shares - used_shares |
---|
353 | |
---|
354 | servermap = peers_to_shares.copy() |
---|
355 | for peer in sorted(peers_to_shares.keys()): |
---|
356 | if peer in used_peers: |
---|
357 | servermap.pop(peer, None) |
---|
358 | else: |
---|
359 | servermap[peer] = set(servermap[peer]) - used_shares |
---|
360 | if servermap[peer] == set(): |
---|
361 | servermap.pop(peer, None) |
---|
362 | # allmydata.test.test_upload.EncodingParameters.test_exception_messages_during_server_selection |
---|
363 | # allmydata.test.test_upload.EncodingParameters.test_problem_layout_comment_52 |
---|
364 | # both ^^ trigger a "keyerror" here .. just ignoring is right? (fixes the tests, but ...) |
---|
365 | try: |
---|
366 | new_peers.remove(peer) |
---|
367 | except KeyError: |
---|
368 | pass |
---|
369 | |
---|
370 | existing_mappings = _calculate_mappings(new_peers, new_shares, servermap) |
---|
371 | existing_peers, existing_shares = _extract_ids(existing_mappings) |
---|
372 | |
---|
373 | # Calculate share placement for the remaining peers and shares which |
---|
374 | # won't be preserved by existing allocations. |
---|
375 | new_peers = new_peers - existing_peers - used_peers |
---|
376 | |
---|
377 | |
---|
378 | new_shares = new_shares - existing_shares - used_shares |
---|
379 | new_mappings = _calculate_mappings(new_peers, new_shares) |
---|
380 | #print("new_peers %s" % new_peers) |
---|
381 | #print("new_mappings %s" % new_mappings) |
---|
382 | mappings = dict(list(readonly_mappings.items()) + list(existing_mappings.items()) + list(new_mappings.items())) |
---|
383 | homeless_shares = set() |
---|
384 | for share in mappings: |
---|
385 | if mappings[share] is None: |
---|
386 | homeless_shares.add(share) |
---|
387 | if len(homeless_shares) != 0: |
---|
388 | # 'servermap' should contain only read/write peers |
---|
389 | _distribute_homeless_shares( |
---|
390 | mappings, homeless_shares, |
---|
391 | { |
---|
392 | k: v |
---|
393 | for k, v in list(peers_to_shares.items()) |
---|
394 | if k not in readonly_peers |
---|
395 | } |
---|
396 | ) |
---|
397 | |
---|
398 | # now, if any share is *still* mapped to None that means "don't |
---|
399 | # care which server it goes on", so we place it on a round-robin |
---|
400 | # of read-write servers |
---|
401 | |
---|
402 | def round_robin(peers): |
---|
403 | while True: |
---|
404 | for peer in peers: |
---|
405 | yield peer |
---|
406 | peer_iter = round_robin(peers - readonly_peers) |
---|
407 | |
---|
408 | return { |
---|
409 | k: v.pop() if v else next(peer_iter) |
---|
410 | for k, v in list(mappings.items()) |
---|
411 | } |
---|