Show More
@@ -1,255 +1,252 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 node import nullid, nullrev |
|
44 | 44 | from i18n import _ |
|
45 | 45 | import random |
|
46 | 46 | import util, dagutil |
|
47 | 47 | |
|
48 |
def _updatesample(dag, nodes, sample, |
|
|
48 | def _updatesample(dag, nodes, sample, quicksamplesize=0): | |
|
49 | 49 | """update an existing sample to match the expected size |
|
50 | 50 | |
|
51 | 51 | The sample is updated with nodes exponentially distant from each head of the |
|
52 | 52 | <nodes> set. (H~1, H~2, H~4, H~8, etc). |
|
53 | 53 | |
|
54 | 54 | If a target size is specified, the sampling will stop once this size is |
|
55 | 55 | reached. Otherwise sampling will happen until roots of the <nodes> set are |
|
56 | 56 | reached. |
|
57 | 57 | |
|
58 | 58 | :dag: a dag object from dagutil |
|
59 | 59 | :nodes: set of nodes we want to discover (if None, assume the whole dag) |
|
60 | 60 | :sample: a sample to update |
|
61 | :always: set of notable nodes that will be part of the sample anyway | |
|
62 | 61 | :quicksamplesize: optional target size of the sample""" |
|
63 | 62 | # if nodes is empty we scan the entire graph |
|
64 | 63 | if nodes: |
|
65 | 64 | heads = dag.headsetofconnecteds(nodes) |
|
66 | 65 | else: |
|
67 | 66 | heads = dag.heads() |
|
68 | 67 | dist = {} |
|
69 | 68 | visit = util.deque(heads) |
|
70 | 69 | seen = set() |
|
71 | 70 | factor = 1 |
|
72 | 71 | while visit: |
|
73 | 72 | curr = visit.popleft() |
|
74 | 73 | if curr in seen: |
|
75 | 74 | continue |
|
76 | 75 | d = dist.setdefault(curr, 1) |
|
77 | 76 | if d > factor: |
|
78 | 77 | factor *= 2 |
|
79 | 78 | if d == factor: |
|
80 | if curr not in always: # need this check for the early exit below | |
|
81 | 79 |
|
|
82 | 80 |
|
|
83 | 81 |
|
|
84 | 82 | seen.add(curr) |
|
85 | 83 | for p in dag.parents(curr): |
|
86 | 84 | if not nodes or p in nodes: |
|
87 | 85 | dist.setdefault(p, d + 1) |
|
88 | 86 | visit.append(p) |
|
89 | 87 | |
|
90 | 88 | def _setupsample(dag, nodes, size): |
|
91 | 89 | always = dag.headsetofconnecteds(nodes) |
|
92 | 90 | desiredlen = size - len(always) |
|
93 | 91 | if desiredlen <= 0: |
|
94 | 92 | # This could be bad if there are very many heads, all unknown to the |
|
95 | 93 | # server. We're counting on long request support here. |
|
96 | 94 | return always, None, desiredlen |
|
97 | 95 | return always, set(), desiredlen |
|
98 | 96 | |
|
99 | 97 | def _takequicksample(dag, nodes, size): |
|
100 | 98 | always, sample, desiredlen = _setupsample(dag, nodes, size) |
|
101 | 99 | if sample is None: |
|
102 | 100 | return always |
|
103 | _updatesample(dag, None, sample, always, quicksamplesize=desiredlen) | |
|
104 | sample.update(always) | |
|
101 | sample = always | |
|
102 | _updatesample(dag, None, sample, quicksamplesize=size) | |
|
105 | 103 | return sample |
|
106 | 104 | |
|
107 | 105 | def _takefullsample(dag, nodes, size): |
|
108 |
sample |
|
|
106 | sample = dag.headsetofconnecteds(nodes) | |
|
109 | 107 | # update from heads |
|
110 |
_updatesample(dag, nodes, sample |
|
|
108 | _updatesample(dag, nodes, sample) | |
|
111 | 109 | # update from roots |
|
112 |
_updatesample(dag.inverse(), nodes, sample |
|
|
110 | _updatesample(dag.inverse(), nodes, sample) | |
|
113 | 111 | assert sample |
|
114 | sample.update(always) | |
|
115 | 112 | sample = _limitsample(sample, size) |
|
116 | 113 | if len(sample) < size: |
|
117 | 114 | more = size - len(sample) |
|
118 | 115 | sample.update(random.sample(list(nodes - sample), more)) |
|
119 | 116 | return sample |
|
120 | 117 | |
|
121 | 118 | def _limitsample(sample, desiredlen): |
|
122 | 119 | """return a random subset of sample of at most desiredlen item""" |
|
123 | 120 | if len(sample) > desiredlen: |
|
124 | 121 | sample = set(random.sample(sample, desiredlen)) |
|
125 | 122 | return sample |
|
126 | 123 | |
|
127 | 124 | def findcommonheads(ui, local, remote, |
|
128 | 125 | initialsamplesize=100, |
|
129 | 126 | fullsamplesize=200, |
|
130 | 127 | abortwhenunrelated=True): |
|
131 | 128 | '''Return a tuple (common, anyincoming, remoteheads) used to identify |
|
132 | 129 | missing nodes from or in remote. |
|
133 | 130 | ''' |
|
134 | 131 | roundtrips = 0 |
|
135 | 132 | cl = local.changelog |
|
136 | 133 | dag = dagutil.revlogdag(cl) |
|
137 | 134 | |
|
138 | 135 | # early exit if we know all the specified remote heads already |
|
139 | 136 | ui.debug("query 1; heads\n") |
|
140 | 137 | roundtrips += 1 |
|
141 | 138 | ownheads = dag.heads() |
|
142 | 139 | sample = _limitsample(ownheads, initialsamplesize) |
|
143 | 140 | # indices between sample and externalized version must match |
|
144 | 141 | sample = list(sample) |
|
145 | 142 | if remote.local(): |
|
146 | 143 | # stopgap until we have a proper localpeer that supports batch() |
|
147 | 144 | srvheadhashes = remote.heads() |
|
148 | 145 | yesno = remote.known(dag.externalizeall(sample)) |
|
149 | 146 | elif remote.capable('batch'): |
|
150 | 147 | batch = remote.batch() |
|
151 | 148 | srvheadhashesref = batch.heads() |
|
152 | 149 | yesnoref = batch.known(dag.externalizeall(sample)) |
|
153 | 150 | batch.submit() |
|
154 | 151 | srvheadhashes = srvheadhashesref.value |
|
155 | 152 | yesno = yesnoref.value |
|
156 | 153 | else: |
|
157 | 154 | # compatibility with pre-batch, but post-known remotes during 1.9 |
|
158 | 155 | # development |
|
159 | 156 | srvheadhashes = remote.heads() |
|
160 | 157 | sample = [] |
|
161 | 158 | |
|
162 | 159 | if cl.tip() == nullid: |
|
163 | 160 | if srvheadhashes != [nullid]: |
|
164 | 161 | return [nullid], True, srvheadhashes |
|
165 | 162 | return [nullid], False, [] |
|
166 | 163 | |
|
167 | 164 | # start actual discovery (we note this before the next "if" for |
|
168 | 165 | # compatibility reasons) |
|
169 | 166 | ui.status(_("searching for changes\n")) |
|
170 | 167 | |
|
171 | 168 | srvheads = dag.internalizeall(srvheadhashes, filterunknown=True) |
|
172 | 169 | if len(srvheads) == len(srvheadhashes): |
|
173 | 170 | ui.debug("all remote heads known locally\n") |
|
174 | 171 | return (srvheadhashes, False, srvheadhashes,) |
|
175 | 172 | |
|
176 | 173 | if sample and len(ownheads) <= initialsamplesize and util.all(yesno): |
|
177 | 174 | ui.note(_("all local heads known remotely\n")) |
|
178 | 175 | ownheadhashes = dag.externalizeall(ownheads) |
|
179 | 176 | return (ownheadhashes, True, srvheadhashes,) |
|
180 | 177 | |
|
181 | 178 | # full blown discovery |
|
182 | 179 | |
|
183 | 180 | # own nodes I know we both know |
|
184 | 181 | # treat remote heads (and maybe own heads) as a first implicit sample |
|
185 | 182 | # response |
|
186 | 183 | common = cl.incrementalmissingrevs(srvheads) |
|
187 | 184 | commoninsample = set(n for i, n in enumerate(sample) if yesno[i]) |
|
188 | 185 | common.addbases(commoninsample) |
|
189 | 186 | # own nodes where I don't know if remote knows them |
|
190 | 187 | undecided = set(common.missingancestors(ownheads)) |
|
191 | 188 | # own nodes I know remote lacks |
|
192 | 189 | missing = set() |
|
193 | 190 | |
|
194 | 191 | full = False |
|
195 | 192 | while undecided: |
|
196 | 193 | |
|
197 | 194 | if sample: |
|
198 | 195 | missinginsample = [n for i, n in enumerate(sample) if not yesno[i]] |
|
199 | 196 | missing.update(dag.descendantset(missinginsample, missing)) |
|
200 | 197 | |
|
201 | 198 | undecided.difference_update(missing) |
|
202 | 199 | |
|
203 | 200 | if not undecided: |
|
204 | 201 | break |
|
205 | 202 | |
|
206 | 203 | if full or common.hasbases(): |
|
207 | 204 | if full: |
|
208 | 205 | ui.note(_("sampling from both directions\n")) |
|
209 | 206 | else: |
|
210 | 207 | ui.debug("taking initial sample\n") |
|
211 | 208 | samplefunc = _takefullsample |
|
212 | 209 | targetsize = fullsamplesize |
|
213 | 210 | else: |
|
214 | 211 | # use even cheaper initial sample |
|
215 | 212 | ui.debug("taking quick initial sample\n") |
|
216 | 213 | samplefunc = _takequicksample |
|
217 | 214 | targetsize = initialsamplesize |
|
218 | 215 | if len(undecided) < targetsize: |
|
219 | 216 | sample = list(undecided) |
|
220 | 217 | else: |
|
221 | 218 | sample = samplefunc(dag, undecided, targetsize) |
|
222 | 219 | sample = _limitsample(sample, targetsize) |
|
223 | 220 | |
|
224 | 221 | roundtrips += 1 |
|
225 | 222 | ui.progress(_('searching'), roundtrips, unit=_('queries')) |
|
226 | 223 | ui.debug("query %i; still undecided: %i, sample size is: %i\n" |
|
227 | 224 | % (roundtrips, len(undecided), len(sample))) |
|
228 | 225 | # indices between sample and externalized version must match |
|
229 | 226 | sample = list(sample) |
|
230 | 227 | yesno = remote.known(dag.externalizeall(sample)) |
|
231 | 228 | full = True |
|
232 | 229 | |
|
233 | 230 | if sample: |
|
234 | 231 | commoninsample = set(n for i, n in enumerate(sample) if yesno[i]) |
|
235 | 232 | common.addbases(commoninsample) |
|
236 | 233 | common.removeancestorsfrom(undecided) |
|
237 | 234 | |
|
238 | 235 | # heads(common) == heads(common.bases) since common represents common.bases |
|
239 | 236 | # and all its ancestors |
|
240 | 237 | result = dag.headsetofconnecteds(common.bases) |
|
241 | 238 | # common.bases can include nullrev, but our contract requires us to not |
|
242 | 239 | # return any heads in that case, so discard that |
|
243 | 240 | result.discard(nullrev) |
|
244 | 241 | ui.progress(_('searching'), None) |
|
245 | 242 | ui.debug("%d total queries\n" % roundtrips) |
|
246 | 243 | |
|
247 | 244 | if not result and srvheadhashes != [nullid]: |
|
248 | 245 | if abortwhenunrelated: |
|
249 | 246 | raise util.Abort(_("repository is unrelated")) |
|
250 | 247 | else: |
|
251 | 248 | ui.warn(_("warning: repository is unrelated\n")) |
|
252 | 249 | return (set([nullid]), True, srvheadhashes,) |
|
253 | 250 | |
|
254 | 251 | anyincoming = (srvheadhashes != [nullid]) |
|
255 | 252 | return dag.externalizeall(result), anyincoming, srvheadhashes |
General Comments 0
You need to be logged in to leave comments.
Login now