« back to all changes in this revision
Viewing changes to bzrlib/smart/repository.py
- browse files at revision 3211.5.1
- compare with another revision
- compare with revision 5050.3.7
- download diff
- download tarball
- view history from revision 3211.5.1
- Committer: Robert Collins
- Date: 2008-02-03 22:55:08 UTC
- mto: This revision was merged to the branch mainline in revision 3216.
- Revision ID: robertc@robertcollins.net-20080203225508-0rogbg0ggonuqfhp
Change the smart server get_parents method to take a graph search to exclude already recieved parents from. This prevents history shortcuts causing huge numbers of duplicates.
added
removed
bzrlib/smart/repository.py
31
31
SuccessfulSmartServerResponse,
32
32
)
33
33
from bzrlib import revision as _mod_revision
34
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
34
35
35
36
36
37
class SmartServerRepositoryRequest(SmartServerRequest):
58
59
# is expected)
59
60
return None
60
61
62
def recreate_search(self, repository, recipe_bytes):
63
lines = recipe_bytes.split('\n')
64
start_keys = set(lines[0].split(' '))
65
exclude_keys = set(lines[1].split(' '))
66
revision_count = int(lines[2])
67
repository.lock_read()
68
try:
69
search = repository.get_graph()._make_breadth_first_searcher(
70
start_keys)
71
while True:
72
try:
73
next_revs = search.next()
74
except StopIteration:
75
break
76
search.stop_searching_any(exclude_keys.intersection(next_revs))
77
search_result = search.get_result()
78
if search_result.get_recipe()[2] != revision_count:
79
# we got back a different amount of data than expected, this
80
# gets reported as NoSuchRevision, because less revisions
81
# indicates missing revisions, and more should never happen as
82
# the excludes list considers ghosts and ensures that ghost
83
# filling races are not a problem.
84
return (None, FailedSmartServerResponse(('NoSuchRevision',)))
85
return (search, None)
86
finally:
87
repository.unlock()
88
61
89
62
90
class SmartServerRepositoryGetParentMap(SmartServerRepositoryRequest):
91
"""Bzr 1.2+ - get parent data for revisions during a graph search."""
63
92
64
93
def do_repository_request(self, repository, *revision_ids):
65
repository.lock_read()
66
try:
67
return self._do_repository_request(repository, revision_ids)
68
finally:
69
repository.unlock()
70
71
def _do_repository_request(self, repository, revision_ids):
72
94
"""Get parent details for some revisions.
73
95
74
96
All the parents for revision_ids are returned. Additionally up to 64KB
75
97
of additional parent data found by performing a breadth first search
76
from revision_ids is returned.
98
from revision_ids is returned. The verb takes a body containing the
99
current search state, see do_body for details.
77
100
78
101
:param repository: The repository to query in.
79
102
:param revision_ids: The utf8 encoded revision_id to answer for.
103
"""
104
self._revision_ids = revision_ids
105
return None # Signal that we want a body.
106
107
def do_body(self, body_bytes):
108
"""Process the current search state and perform the parent lookup.
109
80
110
:return: A smart server response where the body contains an utf8
81
encoded flattened list of the parents of the revisions, (the same
82
format as Repository.get_revision_graph).
111
encoded flattened list of the parents of the revisions (the same
112
format as Repository.get_revision_graph) which has been gzipped.
83
113
"""
114
repository = self._repository
115
repository.lock_read()
116
try:
117
return self._do_repository_request(body_bytes)
118
finally:
119
repository.unlock()
120
121
def _do_repository_request(self, body_bytes):
122
repository = self._repository
123
revision_ids = set(self._revision_ids)
124
search, error = self.recreate_search(repository, body_bytes)
125
if error is not None:
126
return error
127
# TODO might be nice to start up the search again; but thats not
128
# written or tested yet.
129
client_seen_revs = set(search.get_result().get_keys())
130
# Always include the requested ids.
131
client_seen_revs.difference_update(revision_ids)
84
132
lines = []
85
133
repo_graph = repository.get_graph()
86
134
result = {}
96
144
# adjust for the wire
97
145
if parents == (_mod_revision.NULL_REVISION,):
98
146
parents = ()
99
# add parents to the result
100
result[revision_id] = parents
101
147
# prepare the next query
102
148
next_revs.update(parents)
103
# Approximate the serialized cost of this revision_id.
104
size_so_far += 2 + len(revision_id) + sum(map(len, parents))
105
# get all the directly asked for parents, and then flesh out to
106
# 64K or so.
107
if first_loop_done and size_so_far > 65000:
108
next_revs = set()
109
break
149
if revision_id not in client_seen_revs:
150
# Client does not have this revision, give it to it.
151
# add parents to the result
152
result[revision_id] = parents
153
# Approximate the serialized cost of this revision_id.
154
size_so_far += 2 + len(revision_id) + sum(map(len, parents))
155
# get all the directly asked for parents, and then flesh out to
156
# 64K (compressed) or so. We do one level of depth at a time to
157
# stay in sync with the client. The 185000 magic number is
158
# estimated compression ratio taken from bzr.dev itself.
159
if first_loop_done and size_so_far > 185000:
160
next_revs = set()
161
break
110
162
# don't query things we've already queried
111
163
next_revs.difference_update(queried_revs)
112
164
first_loop_done = True
113
165
114
for revision, parents in result.items():
166
# sorting trivially puts lexographically similar revision ids together.
167
# Compression FTW.
168
for revision, parents in sorted(result.items()):
115
169
lines.append(' '.join((revision, ) + tuple(parents)))
116
170
117
return SuccessfulSmartServerResponse(('ok', ), '\n'.join(lines))
171
return SuccessfulSmartServerResponse(
172
('ok', ), bytes_to_gzip('\n'.join(lines)))
118
173
119
174
120
175
class SmartServerRepositoryGetRevisionGraph(SmartServerRepositoryRequest):
342
397
"""Bzr 1.1+ streaming pull."""
343
398
344
399
def do_body(self, body_bytes):
345
lines = body_bytes.split('\n')
346
start_keys = set(lines[0].split(' '))
347
exclude_keys = set(lines[1].split(' '))
348
revision_count = int(lines[2])
349
400
repository = self._repository
350
401
repository.lock_read()
351
402
try:
352
search = repository.get_graph()._make_breadth_first_searcher(
353
start_keys)
354
while True:
355
try:
356
next_revs = search.next()
357
except StopIteration:
358
break
359
search.stop_searching_any(exclude_keys.intersection(next_revs))
360
search_result = search.get_result()
361
if search_result.get_recipe()[2] != revision_count:
362
# we got back a different amount of data than expected, this
363
# gets reported as NoSuchRevision, because less revisions
364
# indicates missing revisions, and more should never happen as
365
# the excludes list considers ghosts and ensures that ghost
366
# filling races are not a problem.
367
return FailedSmartServerResponse(('NoSuchRevision',))
368
stream = repository.get_data_stream_for_search(search_result)
403
search, error = self.recreate_search(repository, body_bytes)
404
if error is not None:
405
return error
406
stream = repository.get_data_stream_for_search(search.get_result())
369
407
except Exception:
408
# On non-error, unlocking is done by the body stream handler.
370
409
repository.unlock()
371
410
raise
372
411
return SuccessfulSmartServerResponse(('ok',),
Loggerhead is a web-based interface for Breezy
Version: 2.0.1