1 # -*- coding: utf-8 -*-
4 from twisted.trial import unittest
5 from twisted.internet import defer
6 from allmydata import uri
7 from allmydata.util.consumer import download_to_data
8 from allmydata.immutable import upload
9 from allmydata.mutable.common import UnrecoverableFileError
10 from allmydata.storage.common import storage_index_to_dir
11 from allmydata.test.no_network import GridTestMixin
12 from allmydata.test.common import ShouldFailMixin
13 from allmydata.util.pollmixin import PollMixin
14 from allmydata.interfaces import NotEnoughSharesError
16 immutable_plaintext = "data" * 10000
17 mutable_plaintext = "muta" * 10000
19 class HungServerDownloadTest(GridTestMixin, ShouldFailMixin, PollMixin,
21 # Many of these tests take around 60 seconds on François's ARM buildslave:
22 # http://tahoe-lafs.org/buildbot/builders/FranXois%20lenny-armv5tel
23 # allmydata.test.test_hung_server.HungServerDownloadTest.test_2_good_8_broken_duplicate_share_fail
24 # once ERRORed after 197 seconds on Midnight Magic's NetBSD buildslave:
25 # http://tahoe-lafs.org/buildbot/builders/MM%20netbsd4%20i386%20warp
26 # MM's buildslave varies a lot in how long it takes to run tests.
30 def _break(self, servers):
31 for (id, ss) in servers:
32 self.g.break_server(id)
34 def _hang(self, servers, **kwargs):
35 for (id, ss) in servers:
36 self.g.hang_server(id, **kwargs)
38 def _unhang(self, servers, **kwargs):
39 for (id, ss) in servers:
40 self.g.unhang_server(id, **kwargs)
42 def _hang_shares(self, shnums, **kwargs):
43 # hang all servers who are holding the given shares
44 hung_serverids = set()
45 for (i_shnum, i_serverid, i_sharefile) in self.shares:
47 if i_serverid not in hung_serverids:
48 self.g.hang_server(i_serverid, **kwargs)
49 hung_serverids.add(i_serverid)
51 def _delete_all_shares_from(self, servers):
52 serverids = [id for (id, ss) in servers]
53 for (i_shnum, i_serverid, i_sharefile) in self.shares:
54 if i_serverid in serverids:
55 os.unlink(i_sharefile)
57 def _corrupt_all_shares_in(self, servers, corruptor_func):
58 serverids = [id for (id, ss) in servers]
59 for (i_shnum, i_serverid, i_sharefile) in self.shares:
60 if i_serverid in serverids:
61 self._corrupt_share((i_shnum, i_sharefile), corruptor_func)
63 def _copy_all_shares_from(self, from_servers, to_server):
64 serverids = [id for (id, ss) in from_servers]
65 for (i_shnum, i_serverid, i_sharefile) in self.shares:
66 if i_serverid in serverids:
67 self._copy_share((i_shnum, i_sharefile), to_server)
69 def _copy_share(self, share, to_server):
70 (sharenum, sharefile) = share
72 shares_dir = os.path.join(ss.original.storedir, "shares")
73 si = uri.from_string(self.uri).get_storage_index()
74 si_dir = os.path.join(shares_dir, storage_index_to_dir(si))
75 if not os.path.exists(si_dir):
77 new_sharefile = os.path.join(si_dir, str(sharenum))
78 shutil.copy(sharefile, new_sharefile)
79 self.shares = self.find_uri_shares(self.uri)
80 # Make sure that the storage server has the share.
81 self.failUnless((sharenum, ss.original.my_nodeid, new_sharefile)
84 def _corrupt_share(self, share, corruptor_func):
85 (sharenum, sharefile) = share
86 data = open(sharefile, "rb").read()
87 newdata = corruptor_func(data)
89 wf = open(sharefile, "wb")
93 def _set_up(self, mutable, testdir, num_clients=1, num_servers=10):
94 self.mutable = mutable
96 self.basedir = "hung_server/mutable_" + testdir
98 self.basedir = "hung_server/immutable_" + testdir
100 self.set_up_grid(num_clients=num_clients, num_servers=num_servers)
102 self.c0 = self.g.clients[0]
103 nm = self.c0.nodemaker
104 self.servers = [(id, ss) for (id, ss) in nm.storage_broker.get_all_servers()]
107 d = nm.create_mutable_file(mutable_plaintext)
108 def _uploaded_mutable(node):
109 self.uri = node.get_uri()
110 self.shares = self.find_uri_shares(self.uri)
111 d.addCallback(_uploaded_mutable)
113 data = upload.Data(immutable_plaintext, convergence="")
114 d = self.c0.upload(data)
115 def _uploaded_immutable(upload_res):
116 self.uri = upload_res.uri
117 self.shares = self.find_uri_shares(self.uri)
118 d.addCallback(_uploaded_immutable)
121 def _start_download(self):
122 n = self.c0.create_node_from_uri(self.uri)
124 d = n.download_best_version()
125 stage_4_d = None # currently we aren't doing any tests which require this for mutable files
127 d = download_to_data(n)
128 #stage_4_d = n._downloader._all_downloads.keys()[0]._stage_4_d # too ugly! FIXME
130 return (d, stage_4_d,)
132 def _wait_for_data(self, n):
134 d = n.download_best_version()
136 d = download_to_data(n)
139 def _check(self, resultingdata):
141 self.failUnlessEqual(resultingdata, mutable_plaintext)
143 self.failUnlessEqual(resultingdata, immutable_plaintext)
145 def _download_and_check(self):
146 d, stage4d = self._start_download()
147 d.addCallback(self._check)
150 def _should_fail_download(self):
152 return self.shouldFail(UnrecoverableFileError, self.basedir,
153 "no recoverable versions",
154 self._download_and_check)
156 return self.shouldFail(NotEnoughSharesError, self.basedir,
158 self._download_and_check)
161 def test_10_good_sanity_check(self):
162 d = defer.succeed(None)
163 for mutable in [False, True]:
164 d.addCallback(lambda ign: self._set_up(mutable, "test_10_good_sanity_check"))
165 d.addCallback(lambda ign: self._download_and_check())
168 def test_10_good_copied_share(self):
169 d = defer.succeed(None)
170 for mutable in [False, True]:
171 d.addCallback(lambda ign: self._set_up(mutable, "test_10_good_copied_share"))
172 d.addCallback(lambda ign: self._copy_all_shares_from(self.servers[2:3], self.servers[0]))
173 d.addCallback(lambda ign: self._download_and_check())
176 def test_3_good_7_noshares(self):
177 d = defer.succeed(None)
178 for mutable in [False, True]:
179 d.addCallback(lambda ign: self._set_up(mutable, "test_3_good_7_noshares"))
180 d.addCallback(lambda ign: self._delete_all_shares_from(self.servers[3:]))
181 d.addCallback(lambda ign: self._download_and_check())
184 def test_2_good_8_broken_fail(self):
185 d = defer.succeed(None)
186 for mutable in [False, True]:
187 d.addCallback(lambda ign: self._set_up(mutable, "test_2_good_8_broken_fail"))
188 d.addCallback(lambda ign: self._break(self.servers[2:]))
189 d.addCallback(lambda ign: self._should_fail_download())
192 def test_2_good_8_noshares_fail(self):
193 d = defer.succeed(None)
194 for mutable in [False, True]:
195 d.addCallback(lambda ign: self._set_up(mutable, "test_2_good_8_noshares_fail"))
196 d.addCallback(lambda ign: self._delete_all_shares_from(self.servers[2:]))
197 d.addCallback(lambda ign: self._should_fail_download())
200 def test_2_good_8_broken_copied_share(self):
201 d = defer.succeed(None)
202 for mutable in [False, True]:
203 d.addCallback(lambda ign: self._set_up(mutable, "test_2_good_8_broken_copied_share"))
204 d.addCallback(lambda ign: self._copy_all_shares_from(self.servers[2:3], self.servers[0]))
205 d.addCallback(lambda ign: self._break(self.servers[2:]))
206 d.addCallback(lambda ign: self._download_and_check())
209 def test_2_good_8_broken_duplicate_share_fail(self):
210 d = defer.succeed(None)
211 for mutable in [False, True]:
212 d.addCallback(lambda ign: self._set_up(mutable, "test_2_good_8_broken_duplicate_share_fail"))
213 d.addCallback(lambda ign: self._copy_all_shares_from(self.servers[1:2], self.servers[0]))
214 d.addCallback(lambda ign: self._break(self.servers[2:]))
215 d.addCallback(lambda ign: self._should_fail_download())
218 # The tests below do not currently pass for mutable files.
220 def test_3_good_7_hung_immutable(self):
221 d = defer.succeed(None)
222 d.addCallback(lambda ign: self._set_up(False, "test_3_good_7_hung"))
223 d.addCallback(lambda ign: self._hang(self.servers[3:]))
224 d.addCallback(lambda ign: self._download_and_check())
227 def test_5_overdue_immutable(self):
228 # restrict the ShareFinder to only allow 5 outstanding requests, and
229 # arrange for the first 5 servers to hang. Then trigger the OVERDUE
230 # timers (simulating 10 seconds passed), at which point the
231 # ShareFinder should send additional queries and finish the download
232 # quickly. If we didn't have OVERDUE timers, this test would fail by
235 d = self._set_up(False, "test_5_overdue_immutable")
236 def _reduce_max_outstanding_requests_and_download(ign):
237 self._hang_shares(range(5))
238 n = self.c0.create_node_from_uri(self.uri)
239 self._sf = n._cnode._node._sharefinder
240 self._sf.max_outstanding_requests = 5
241 self._sf.OVERDUE_TIMEOUT = 1000.0
242 d2 = download_to_data(n)
243 # start download, but don't wait for it to complete yet
245 done.append(res) # we will poll for this later
247 d.addCallback(_reduce_max_outstanding_requests_and_download)
248 from foolscap.eventual import fireEventually, flushEventualQueue
250 d.addCallback(lambda res: fireEventually(res))
251 d.addCallback(lambda res: flushEventualQueue())
252 d.addCallback(lambda ign: self.failIf(done))
253 def _check_waiting(ign):
254 # all the share requests should now be stuck waiting
255 self.failUnlessEqual(len(self._sf.pending_requests), 5)
256 # but none should be marked as OVERDUE until the timers expire
257 self.failUnlessEqual(len(self._sf.overdue_requests), 0)
258 d.addCallback(_check_waiting)
259 def _mark_overdue(ign):
260 # declare four requests overdue, allowing new requests to take
261 # their place, and leaving one stuck. The finder will keep
262 # sending requests until there are 5 non-overdue ones
263 # outstanding, at which point we'll have 4 OVERDUE, 1
264 # stuck-but-not-overdue, and 4 live requests. All 4 live requests
265 # will retire before the download is complete and the ShareFinder
266 # is shut off. That will leave 4 OVERDUE and 1
267 # stuck-but-not-overdue, for a total of 5 requests in in
268 # _sf.pending_requests
269 for t in self._sf.overdue_timers.values()[:4]:
271 # the timers ought to fire before the eventual-send does
272 return fireEventually()
273 d.addCallback(_mark_overdue)
276 d.addCallback(lambda ign: self.poll(_we_are_done))
277 def _check_done(ign):
278 self.failUnlessEqual(done, [immutable_plaintext])
279 self.failUnlessEqual(len(self._sf.pending_requests), 5)
280 self.failUnlessEqual(len(self._sf.overdue_requests), 4)
281 d.addCallback(_check_done)
284 def test_3_good_7_hung_mutable(self):
285 raise unittest.SkipTest("still broken")
286 d = defer.succeed(None)
287 d.addCallback(lambda ign: self._set_up(True, "test_3_good_7_hung"))
288 d.addCallback(lambda ign: self._hang(self.servers[3:]))
289 d.addCallback(lambda ign: self._download_and_check())
292 def test_2_good_8_hung_then_1_recovers_immutable(self):
293 d = defer.succeed(None)
294 d.addCallback(lambda ign: self._set_up(False, "test_2_good_8_hung_then_1_recovers"))
295 d.addCallback(lambda ign: self._hang(self.servers[2:3]))
296 d.addCallback(lambda ign: self._hang(self.servers[3:]))
297 d.addCallback(lambda ign: self._unhang(self.servers[2:3]))
298 d.addCallback(lambda ign: self._download_and_check())
301 def test_2_good_8_hung_then_1_recovers_mutable(self):
302 raise unittest.SkipTest("still broken")
303 d = defer.succeed(None)
304 d.addCallback(lambda ign: self._set_up(True, "test_2_good_8_hung_then_1_recovers"))
305 d.addCallback(lambda ign: self._hang(self.servers[2:3]))
306 d.addCallback(lambda ign: self._hang(self.servers[3:]))
307 d.addCallback(lambda ign: self._unhang(self.servers[2:3]))
308 d.addCallback(lambda ign: self._download_and_check())
311 def test_2_good_8_hung_then_1_recovers_with_2_shares_immutable(self):
312 d = defer.succeed(None)
313 d.addCallback(lambda ign: self._set_up(False, "test_2_good_8_hung_then_1_recovers_with_2_shares"))
314 d.addCallback(lambda ign: self._copy_all_shares_from(self.servers[0:1], self.servers[2]))
315 d.addCallback(lambda ign: self._hang(self.servers[2:3]))
316 d.addCallback(lambda ign: self._hang(self.servers[3:]))
317 d.addCallback(lambda ign: self._unhang(self.servers[2:3]))
318 d.addCallback(lambda ign: self._download_and_check())
321 def test_2_good_8_hung_then_1_recovers_with_2_shares_mutable(self):
322 raise unittest.SkipTest("still broken")
323 d = defer.succeed(None)
324 d.addCallback(lambda ign: self._set_up(True, "test_2_good_8_hung_then_1_recovers_with_2_shares"))
325 d.addCallback(lambda ign: self._copy_all_shares_from(self.servers[0:1], self.servers[2]))
326 d.addCallback(lambda ign: self._hang(self.servers[2:3]))
327 d.addCallback(lambda ign: self._hang(self.servers[3:]))
328 d.addCallback(lambda ign: self._unhang(self.servers[2:3]))
329 d.addCallback(lambda ign: self._download_and_check())