Show More
@@ -1,361 +1,359 b'' | |||
|
1 | 1 | # setdiscovery.py - improved discovery of common nodeset for mercurial |
|
2 | 2 | # |
|
3 | 3 | # Copyright 2010 Benoit Boissinot <bboissin@gmail.com> |
|
4 | 4 | # and Peter Arrenbrecht <peter@arrenbrecht.ch> |
|
5 | 5 | # |
|
6 | 6 | # This software may be used and distributed according to the terms of the |
|
7 | 7 | # GNU General Public License version 2 or any later version. |
|
8 | 8 | """ |
|
9 | 9 | Algorithm works in the following way. You have two repository: local and |
|
10 | 10 | remote. They both contains a DAG of changelists. |
|
11 | 11 | |
|
12 | 12 | The goal of the discovery protocol is to find one set of node *common*, |
|
13 | 13 | the set of nodes shared by local and remote. |
|
14 | 14 | |
|
15 | 15 | One of the issue with the original protocol was latency, it could |
|
16 | 16 | potentially require lots of roundtrips to discover that the local repo was a |
|
17 | 17 | subset of remote (which is a very common case, you usually have few changes |
|
18 | 18 | compared to upstream, while upstream probably had lots of development). |
|
19 | 19 | |
|
20 | 20 | The new protocol only requires one interface for the remote repo: `known()`, |
|
21 | 21 | which given a set of changelists tells you if they are present in the DAG. |
|
22 | 22 | |
|
23 | 23 | The algorithm then works as follow: |
|
24 | 24 | |
|
25 | 25 | - We will be using three sets, `common`, `missing`, `unknown`. Originally |
|
26 | 26 | all nodes are in `unknown`. |
|
27 | 27 | - Take a sample from `unknown`, call `remote.known(sample)` |
|
28 | 28 | - For each node that remote knows, move it and all its ancestors to `common` |
|
29 | 29 | - For each node that remote doesn't know, move it and all its descendants |
|
30 | 30 | to `missing` |
|
31 | 31 | - Iterate until `unknown` is empty |
|
32 | 32 | |
|
33 | 33 | There are a couple optimizations, first is instead of starting with a random |
|
34 | 34 | sample of missing, start by sending all heads, in the case where the local |
|
35 | 35 | repo is a subset, you computed the answer in one round trip. |
|
36 | 36 | |
|
37 | 37 | Then you can do something similar to the bisecting strategy used when |
|
38 | 38 | finding faulty changesets. Instead of random samples, you can try picking |
|
39 | 39 | nodes that will maximize the number of nodes that will be |
|
40 | 40 | classified with it (since all ancestors or descendants will be marked as well). |
|
41 | 41 | """ |
|
42 | 42 | |
|
43 | 43 | from __future__ import absolute_import |
|
44 | 44 | |
|
45 | 45 | import collections |
|
46 | 46 | import random |
|
47 | 47 | |
|
48 | 48 | from .i18n import _ |
|
49 | 49 | from .node import ( |
|
50 | 50 | nullid, |
|
51 | 51 | nullrev, |
|
52 | 52 | ) |
|
53 | 53 | from . import ( |
|
54 | 54 | error, |
|
55 | 55 | util, |
|
56 | 56 | ) |
|
57 | 57 | |
|
58 | 58 | def _updatesample(revs, heads, sample, parentfn, quicksamplesize=0): |
|
59 | 59 | """update an existing sample to match the expected size |
|
60 | 60 | |
|
61 | 61 | The sample is updated with revs exponentially distant from each head of the |
|
62 | 62 | <revs> set. (H~1, H~2, H~4, H~8, etc). |
|
63 | 63 | |
|
64 | 64 | If a target size is specified, the sampling will stop once this size is |
|
65 | 65 | reached. Otherwise sampling will happen until roots of the <revs> set are |
|
66 | 66 | reached. |
|
67 | 67 | |
|
68 | 68 | :revs: set of revs we want to discover (if None, assume the whole dag) |
|
69 | 69 | :heads: set of DAG head revs |
|
70 | 70 | :sample: a sample to update |
|
71 | 71 | :parentfn: a callable to resolve parents for a revision |
|
72 | 72 | :quicksamplesize: optional target size of the sample""" |
|
73 | 73 | dist = {} |
|
74 | 74 | visit = collections.deque(heads) |
|
75 | 75 | seen = set() |
|
76 | 76 | factor = 1 |
|
77 | 77 | while visit: |
|
78 | 78 | curr = visit.popleft() |
|
79 | 79 | if curr in seen: |
|
80 | 80 | continue |
|
81 | 81 | d = dist.setdefault(curr, 1) |
|
82 | 82 | if d > factor: |
|
83 | 83 | factor *= 2 |
|
84 | 84 | if d == factor: |
|
85 | 85 | sample.add(curr) |
|
86 | 86 | if quicksamplesize and (len(sample) >= quicksamplesize): |
|
87 | 87 | return |
|
88 | 88 | seen.add(curr) |
|
89 | 89 | |
|
90 | 90 | for p in parentfn(curr): |
|
91 | 91 | if p != nullrev and (not revs or p in revs): |
|
92 | 92 | dist.setdefault(p, d + 1) |
|
93 | 93 | visit.append(p) |
|
94 | 94 | |
|
95 | 95 | def _takequicksample(repo, headrevs, revs, size): |
|
96 | 96 | """takes a quick sample of size <size> |
|
97 | 97 | |
|
98 | 98 | It is meant for initial sampling and focuses on querying heads and close |
|
99 | 99 | ancestors of heads. |
|
100 | 100 | |
|
101 | 101 | :dag: a dag object |
|
102 | 102 | :headrevs: set of head revisions in local DAG to consider |
|
103 | 103 | :revs: set of revs to discover |
|
104 | 104 | :size: the maximum size of the sample""" |
|
105 | 105 | if len(revs) <= size: |
|
106 | 106 | return list(revs) |
|
107 | 107 | sample = set(repo.revs('heads(%ld)', revs)) |
|
108 | 108 | |
|
109 | 109 | if len(sample) >= size: |
|
110 | 110 | return _limitsample(sample, size) |
|
111 | 111 | |
|
112 | 112 | _updatesample(None, headrevs, sample, repo.changelog.parentrevs, |
|
113 | 113 | quicksamplesize=size) |
|
114 | 114 | return sample |
|
115 | 115 | |
|
116 | 116 | def _takefullsample(repo, headrevs, revs, size): |
|
117 | 117 | if len(revs) <= size: |
|
118 | 118 | return list(revs) |
|
119 | 119 | sample = set(repo.revs('heads(%ld)', revs)) |
|
120 | 120 | |
|
121 | 121 | # update from heads |
|
122 | 122 | revsheads = set(repo.revs('heads(%ld)', revs)) |
|
123 | 123 | _updatesample(revs, revsheads, sample, repo.changelog.parentrevs) |
|
124 | 124 | |
|
125 | 125 | # update from roots |
|
126 | 126 | revsroots = set(repo.revs('roots(%ld)', revs)) |
|
127 | 127 | |
|
128 | 128 | # _updatesample() essentially does interaction over revisions to look up |
|
129 | 129 | # their children. This lookup is expensive and doing it in a loop is |
|
130 | 130 | # quadratic. We precompute the children for all relevant revisions and |
|
131 | 131 | # make the lookup in _updatesample() a simple dict lookup. |
|
132 | 132 | # |
|
133 | 133 | # Because this function can be called multiple times during discovery, we |
|
134 | 134 | # may still perform redundant work and there is room to optimize this by |
|
135 | 135 | # keeping a persistent cache of children across invocations. |
|
136 | 136 | children = {} |
|
137 | 137 | |
|
138 | 138 | parentrevs = repo.changelog.parentrevs |
|
139 | 139 | for rev in repo.changelog.revs(start=min(revsroots)): |
|
140 | 140 | # Always ensure revision has an entry so we don't need to worry about |
|
141 | 141 | # missing keys. |
|
142 | 142 | children.setdefault(rev, []) |
|
143 | 143 | |
|
144 | 144 | for prev in parentrevs(rev): |
|
145 | 145 | if prev == nullrev: |
|
146 | 146 | continue |
|
147 | 147 | |
|
148 | 148 | children.setdefault(prev, []).append(rev) |
|
149 | 149 | |
|
150 | 150 | _updatesample(revs, revsroots, sample, children.__getitem__) |
|
151 | 151 | assert sample |
|
152 | 152 | sample = _limitsample(sample, size) |
|
153 | 153 | if len(sample) < size: |
|
154 | 154 | more = size - len(sample) |
|
155 | 155 | sample.update(random.sample(list(revs - sample), more)) |
|
156 | 156 | return sample |
|
157 | 157 | |
|
158 | 158 | def _limitsample(sample, desiredlen): |
|
159 | 159 | """return a random subset of sample of at most desiredlen item""" |
|
160 | 160 | if len(sample) > desiredlen: |
|
161 | 161 | sample = set(random.sample(sample, desiredlen)) |
|
162 | 162 | return sample |
|
163 | 163 | |
|
164 | 164 | class partialdiscovery(object): |
|
165 | 165 | """an object representing ongoing discovery |
|
166 | 166 | |
|
167 | 167 | Feed with data from the remote repository, this object keep track of the |
|
168 | 168 | current set of changeset in various states: |
|
169 | 169 | |
|
170 | 170 | - common: revs also known remotely |
|
171 | 171 | - undecided: revs we don't have information on yet |
|
172 | 172 | - missing: revs missing remotely |
|
173 | 173 | (all tracked revisions are known locally) |
|
174 | 174 | """ |
|
175 | 175 | |
|
176 | 176 | def __init__(self, repo, targetheads): |
|
177 | 177 | self._repo = repo |
|
178 | 178 | self._targetheads = targetheads |
|
179 | 179 | self._common = repo.changelog.incrementalmissingrevs() |
|
180 | 180 | self._undecided = None |
|
181 | 181 | self.missing = set() |
|
182 | 182 | |
|
183 | 183 | def addcommons(self, commons): |
|
184 | 184 | """registrer nodes known as common""" |
|
185 | 185 | self._common.addbases(commons) |
|
186 | 186 | self._common.removeancestorsfrom(self.undecided) |
|
187 | 187 | |
|
188 | 188 | def addmissings(self, missings): |
|
189 | 189 | """registrer some nodes as missing""" |
|
190 | 190 | if self.missing: |
|
191 | 191 | new = self._repo.revs('descendants(%ld) - descendants(%ld)', |
|
192 | 192 | missings, self.missing) |
|
193 | 193 | self.missing.update(new) |
|
194 | 194 | else: |
|
195 | 195 | self.missing.update(self._repo.revs('descendants(%ld)', missings)) |
|
196 | 196 | |
|
197 | 197 | self.undecided.difference_update(self.missing) |
|
198 | 198 | |
|
199 | 199 | def addinfo(self, sample): |
|
200 | 200 | """consume an iterable of (rev, known) tuples""" |
|
201 | 201 | common = set() |
|
202 | 202 | missing = set() |
|
203 | 203 | for rev, known in sample: |
|
204 | 204 | if known: |
|
205 | 205 | common.add(rev) |
|
206 | 206 | else: |
|
207 | 207 | missing.add(rev) |
|
208 | 208 | if common: |
|
209 | 209 | self.addcommons(common) |
|
210 | 210 | if missing: |
|
211 | 211 | self.addmissings(missing) |
|
212 | 212 | |
|
213 | 213 | def hasinfo(self): |
|
214 | 214 | """return True is we have any clue about the remote state""" |
|
215 | 215 | return self._common.hasbases() |
|
216 | 216 | |
|
217 | 217 | def iscomplete(self): |
|
218 | 218 | """True if all the necessary data have been gathered""" |
|
219 | 219 | return self._undecided is not None and not self._undecided |
|
220 | 220 | |
|
221 | 221 | @property |
|
222 | 222 | def undecided(self): |
|
223 | 223 | if self._undecided is not None: |
|
224 | 224 | return self._undecided |
|
225 | 225 | self._undecided = set(self._common.missingancestors(self._targetheads)) |
|
226 | 226 | return self._undecided |
|
227 | 227 | |
|
228 | 228 | def commonheads(self): |
|
229 | 229 | """the heads of the known common set""" |
|
230 | 230 | # heads(common) == heads(common.bases) since common represents |
|
231 | 231 | # common.bases and all its ancestors |
|
232 | # The presence of nullrev will confuse heads(). So filter it out. | |
|
233 | return set(self._repo.revs('heads(%ld)', | |
|
234 | self._common.bases - {nullrev})) | |
|
232 | return self._common.basesheads() | |
|
235 | 233 | |
|
236 | 234 | def findcommonheads(ui, local, remote, |
|
237 | 235 | initialsamplesize=100, |
|
238 | 236 | fullsamplesize=200, |
|
239 | 237 | abortwhenunrelated=True, |
|
240 | 238 | ancestorsof=None): |
|
241 | 239 | '''Return a tuple (common, anyincoming, remoteheads) used to identify |
|
242 | 240 | missing nodes from or in remote. |
|
243 | 241 | ''' |
|
244 | 242 | start = util.timer() |
|
245 | 243 | |
|
246 | 244 | roundtrips = 0 |
|
247 | 245 | cl = local.changelog |
|
248 | 246 | clnode = cl.node |
|
249 | 247 | clrev = cl.rev |
|
250 | 248 | |
|
251 | 249 | if ancestorsof is not None: |
|
252 | 250 | ownheads = [clrev(n) for n in ancestorsof] |
|
253 | 251 | else: |
|
254 | 252 | ownheads = [rev for rev in cl.headrevs() if rev != nullrev] |
|
255 | 253 | |
|
256 | 254 | # early exit if we know all the specified remote heads already |
|
257 | 255 | ui.debug("query 1; heads\n") |
|
258 | 256 | roundtrips += 1 |
|
259 | 257 | sample = _limitsample(ownheads, initialsamplesize) |
|
260 | 258 | # indices between sample and externalized version must match |
|
261 | 259 | sample = list(sample) |
|
262 | 260 | |
|
263 | 261 | with remote.commandexecutor() as e: |
|
264 | 262 | fheads = e.callcommand('heads', {}) |
|
265 | 263 | fknown = e.callcommand('known', { |
|
266 | 264 | 'nodes': [clnode(r) for r in sample], |
|
267 | 265 | }) |
|
268 | 266 | |
|
269 | 267 | srvheadhashes, yesno = fheads.result(), fknown.result() |
|
270 | 268 | |
|
271 | 269 | if cl.tip() == nullid: |
|
272 | 270 | if srvheadhashes != [nullid]: |
|
273 | 271 | return [nullid], True, srvheadhashes |
|
274 | 272 | return [nullid], False, [] |
|
275 | 273 | |
|
276 | 274 | # start actual discovery (we note this before the next "if" for |
|
277 | 275 | # compatibility reasons) |
|
278 | 276 | ui.status(_("searching for changes\n")) |
|
279 | 277 | |
|
280 | 278 | srvheads = [] |
|
281 | 279 | for node in srvheadhashes: |
|
282 | 280 | if node == nullid: |
|
283 | 281 | continue |
|
284 | 282 | |
|
285 | 283 | try: |
|
286 | 284 | srvheads.append(clrev(node)) |
|
287 | 285 | # Catches unknown and filtered nodes. |
|
288 | 286 | except error.LookupError: |
|
289 | 287 | continue |
|
290 | 288 | |
|
291 | 289 | if len(srvheads) == len(srvheadhashes): |
|
292 | 290 | ui.debug("all remote heads known locally\n") |
|
293 | 291 | return srvheadhashes, False, srvheadhashes |
|
294 | 292 | |
|
295 | 293 | if len(sample) == len(ownheads) and all(yesno): |
|
296 | 294 | ui.note(_("all local heads known remotely\n")) |
|
297 | 295 | ownheadhashes = [clnode(r) for r in ownheads] |
|
298 | 296 | return ownheadhashes, True, srvheadhashes |
|
299 | 297 | |
|
300 | 298 | # full blown discovery |
|
301 | 299 | |
|
302 | 300 | disco = partialdiscovery(local, ownheads) |
|
303 | 301 | # treat remote heads (and maybe own heads) as a first implicit sample |
|
304 | 302 | # response |
|
305 | 303 | disco.addcommons(srvheads) |
|
306 | 304 | disco.addinfo(zip(sample, yesno)) |
|
307 | 305 | |
|
308 | 306 | full = False |
|
309 | 307 | progress = ui.makeprogress(_('searching'), unit=_('queries')) |
|
310 | 308 | while not disco.iscomplete(): |
|
311 | 309 | |
|
312 | 310 | if full or disco.hasinfo(): |
|
313 | 311 | if full: |
|
314 | 312 | ui.note(_("sampling from both directions\n")) |
|
315 | 313 | else: |
|
316 | 314 | ui.debug("taking initial sample\n") |
|
317 | 315 | samplefunc = _takefullsample |
|
318 | 316 | targetsize = fullsamplesize |
|
319 | 317 | else: |
|
320 | 318 | # use even cheaper initial sample |
|
321 | 319 | ui.debug("taking quick initial sample\n") |
|
322 | 320 | samplefunc = _takequicksample |
|
323 | 321 | targetsize = initialsamplesize |
|
324 | 322 | sample = samplefunc(local, ownheads, disco.undecided, targetsize) |
|
325 | 323 | |
|
326 | 324 | roundtrips += 1 |
|
327 | 325 | progress.update(roundtrips) |
|
328 | 326 | ui.debug("query %i; still undecided: %i, sample size is: %i\n" |
|
329 | 327 | % (roundtrips, len(disco.undecided), len(sample))) |
|
330 | 328 | # indices between sample and externalized version must match |
|
331 | 329 | sample = list(sample) |
|
332 | 330 | |
|
333 | 331 | with remote.commandexecutor() as e: |
|
334 | 332 | yesno = e.callcommand('known', { |
|
335 | 333 | 'nodes': [clnode(r) for r in sample], |
|
336 | 334 | }).result() |
|
337 | 335 | |
|
338 | 336 | full = True |
|
339 | 337 | |
|
340 | 338 | disco.addinfo(zip(sample, yesno)) |
|
341 | 339 | |
|
342 | 340 | result = disco.commonheads() |
|
343 | 341 | elapsed = util.timer() - start |
|
344 | 342 | progress.complete() |
|
345 | 343 | ui.debug("%d total queries in %.4fs\n" % (roundtrips, elapsed)) |
|
346 | 344 | msg = ('found %d common and %d unknown server heads,' |
|
347 | 345 | ' %d roundtrips in %.4fs\n') |
|
348 | 346 | missing = set(result) - set(srvheads) |
|
349 | 347 | ui.log('discovery', msg, len(result), len(missing), roundtrips, |
|
350 | 348 | elapsed) |
|
351 | 349 | |
|
352 | 350 | if not result and srvheadhashes != [nullid]: |
|
353 | 351 | if abortwhenunrelated: |
|
354 | 352 | raise error.Abort(_("repository is unrelated")) |
|
355 | 353 | else: |
|
356 | 354 | ui.warn(_("warning: repository is unrelated\n")) |
|
357 | 355 | return ({nullid}, True, srvheadhashes,) |
|
358 | 356 | |
|
359 | 357 | anyincoming = (srvheadhashes != [nullid]) |
|
360 | 358 | result = {clnode(r) for r in result} |
|
361 | 359 | return result, anyincoming, srvheadhashes |
General Comments 0
You need to be logged in to leave comments.
Login now