]> git.rkrishnan.org Git - tahoe-lafs/tahoe-lafs.git/blob - src/allmydata/test/test_checker.py
d030c12a40176b225a574008e09b31732be57728
[tahoe-lafs/tahoe-lafs.git] / src / allmydata / test / test_checker.py
1
2 import simplejson
3 import os.path, shutil
4 from twisted.trial import unittest
5 from twisted.internet import defer
6 from allmydata import check_results, uri
7 from allmydata import uri as tahoe_uri
8 from allmydata.util import base32
9 from allmydata.web import check_results as web_check_results
10 from allmydata.storage_client import StorageFarmBroker, NativeStorageServer
11 from allmydata.storage.server import storage_index_to_dir
12 from allmydata.monitor import Monitor
13 from allmydata.test.no_network import GridTestMixin
14 from allmydata.immutable.upload import Data
15 from allmydata.test.common_web import WebRenderingMixin
16 from allmydata.mutable.publish import MutableData
17
18 class FakeClient:
19     def get_storage_broker(self):
20         return self.storage_broker
21
22 class WebResultsRendering(unittest.TestCase, WebRenderingMixin):
23
24     def create_fake_client(self):
25         sb = StorageFarmBroker(None, True)
26         # s.get_name() (the "short description") will be "v0-00000000".
27         # s.get_longname() will include the -long suffix.
28         # s.get_peerid() (i.e. tubid) will be "aaa.." or "777.." or "ceir.."
29         servers = [("v0-00000000-long", "\x00"*20, "peer-0"),
30                    ("v0-ffffffff-long", "\xff"*20, "peer-f"),
31                    ("v0-11111111-long", "\x11"*20, "peer-11")]
32         for (key_s, peerid, nickname) in servers:
33             tubid_b32 = base32.b2a(peerid)
34             furl = "pb://%s@nowhere/fake" % tubid_b32
35             ann = { "version": 0,
36                     "service-name": "storage",
37                     "anonymous-storage-FURL": furl,
38                     "permutation-seed-base32": "",
39                     "nickname": unicode(nickname),
40                     "app-versions": {}, # need #466 and v2 introducer
41                     "my-version": "ver",
42                     "oldest-supported": "oldest",
43                     }
44             s = NativeStorageServer(key_s, ann)
45             sb.test_add_server(peerid, s) # XXX: maybe use key_s?
46         c = FakeClient()
47         c.storage_broker = sb
48         return c
49
50     def render_json(self, page):
51         d = self.render1(page, args={"output": ["json"]})
52         return d
53
54     def test_literal(self):
55         c = self.create_fake_client()
56         lcr = web_check_results.LiteralCheckResultsRenderer(c)
57
58         d = self.render1(lcr)
59         def _check(html):
60             s = self.remove_tags(html)
61             self.failUnlessIn("Literal files are always healthy", s)
62         d.addCallback(_check)
63         d.addCallback(lambda ignored:
64                       self.render1(lcr, args={"return_to": ["FOOURL"]}))
65         def _check_return_to(html):
66             s = self.remove_tags(html)
67             self.failUnlessIn("Literal files are always healthy", s)
68             self.failUnlessIn('<a href="FOOURL">Return to file.</a>',
69                               html)
70         d.addCallback(_check_return_to)
71         d.addCallback(lambda ignored: self.render_json(lcr))
72         def _check_json(json):
73             j = simplejson.loads(json)
74             self.failUnlessEqual(j["storage-index"], "")
75             self.failUnlessEqual(j["results"]["healthy"], True)
76         d.addCallback(_check_json)
77         return d
78
79     def test_check(self):
80         c = self.create_fake_client()
81         sb = c.storage_broker
82         serverid_1 = "\x00"*20
83         serverid_f = "\xff"*20
84         server_1 = sb.get_stub_server(serverid_1)
85         server_f = sb.get_stub_server(serverid_f)
86         u = uri.CHKFileURI("\x00"*16, "\x00"*32, 3, 10, 1234)
87         data = { "count_happiness": 8,
88                  "count_shares_needed": 3,
89                  "count_shares_expected": 9,
90                  "count_shares_good": 10,
91                  "count_good_share_hosts": 11,
92                  "count_recoverable_versions": 1,
93                  "count_unrecoverable_versions": 0,
94                  "servers_responding": [],
95                  "sharemap": {"shareid1": [server_1, server_f]},
96                  "count_wrong_shares": 0,
97                  "list_corrupt_shares": [],
98                  "count_corrupt_shares": 0,
99                  "list_incompatible_shares": [],
100                  "count_incompatible_shares": 0,
101                  "report": [], "share_problems": [], "servermap": None,
102                  }
103         cr = check_results.CheckResults(u, u.get_storage_index(),
104                                         healthy=True, recoverable=True,
105                                         summary="groovy",
106                                         **data)
107         w = web_check_results.CheckResultsRenderer(c, cr)
108         html = self.render2(w)
109         s = self.remove_tags(html)
110         self.failUnlessIn("File Check Results for SI=2k6avp", s) # abbreviated
111         self.failUnlessIn("Healthy : groovy", s)
112         self.failUnlessIn("Share Counts: need 3-of-9, have 10", s)
113         self.failUnlessIn("Happiness Level: 8", s)
114         self.failUnlessIn("Hosts with good shares: 11", s)
115         self.failUnlessIn("Corrupt shares: none", s)
116         self.failUnlessIn("Wrong Shares: 0", s)
117         self.failUnlessIn("Recoverable Versions: 1", s)
118         self.failUnlessIn("Unrecoverable Versions: 0", s)
119         self.failUnlessIn("Good Shares (sorted in share order): Share ID Nickname Node ID shareid1 peer-0 00000000 peer-f ffffffff", s)
120
121         cr = check_results.CheckResults(u, u.get_storage_index(),
122                                         healthy=False, recoverable=True,
123                                         summary="ungroovy",
124                                         **data)
125         w = web_check_results.CheckResultsRenderer(c, cr)
126         html = self.render2(w)
127         s = self.remove_tags(html)
128         self.failUnlessIn("File Check Results for SI=2k6avp", s) # abbreviated
129         self.failUnlessIn("Not Healthy! : ungroovy", s)
130
131         data["count_corrupt_shares"] = 1
132         data["list_corrupt_shares"] = [(server_1, u.get_storage_index(), 2)]
133         cr = check_results.CheckResults(u, u.get_storage_index(),
134                                         healthy=False, recoverable=False,
135                                         summary="rather dead",
136                                         **data)
137         w = web_check_results.CheckResultsRenderer(c, cr)
138         html = self.render2(w)
139         s = self.remove_tags(html)
140         self.failUnlessIn("File Check Results for SI=2k6avp", s) # abbreviated
141         self.failUnlessIn("Not Recoverable! : rather dead", s)
142         self.failUnlessIn("Corrupt shares: Share ID Nickname Node ID sh#2 peer-0 00000000", s)
143
144         html = self.render2(w)
145         s = self.remove_tags(html)
146         self.failUnlessIn("File Check Results for SI=2k6avp", s) # abbreviated
147         self.failUnlessIn("Not Recoverable! : rather dead", s)
148
149         html = self.render2(w, args={"return_to": ["FOOURL"]})
150         self.failUnlessIn('<a href="FOOURL">Return to file/directory.</a>',
151                           html)
152
153         d = self.render_json(w)
154         def _check_json(jdata):
155             j = simplejson.loads(jdata)
156             self.failUnlessEqual(j["summary"], "rather dead")
157             self.failUnlessEqual(j["storage-index"],
158                                  "2k6avpjga3dho3zsjo6nnkt7n4")
159             expected = {'count-happiness': 8,
160                         'count-shares-expected': 9,
161                         'healthy': False,
162                         'count-unrecoverable-versions': 0,
163                         'count-shares-needed': 3,
164                         'sharemap': {"shareid1":
165                                      ["v0-00000000-long", "v0-ffffffff-long"]},
166                         'count-recoverable-versions': 1,
167                         'list-corrupt-shares':
168                         [["v0-00000000-long", "2k6avpjga3dho3zsjo6nnkt7n4", 2]],
169                         'count-good-share-hosts': 11,
170                         'count-wrong-shares': 0,
171                         'count-shares-good': 10,
172                         'count-corrupt-shares': 1,
173                         'servers-responding': [],
174                         'recoverable': False,
175                         }
176             self.failUnlessEqual(j["results"], expected)
177         d.addCallback(_check_json)
178         d.addCallback(lambda ignored: self.render1(w))
179         def _check(html):
180             s = self.remove_tags(html)
181             self.failUnlessIn("File Check Results for SI=2k6avp", s)
182             self.failUnlessIn("Not Recoverable! : rather dead", s)
183         d.addCallback(_check)
184         return d
185
186
187     def test_check_and_repair(self):
188         c = self.create_fake_client()
189         sb = c.storage_broker
190         serverid_1 = "\x00"*20
191         serverid_f = "\xff"*20
192         u = uri.CHKFileURI("\x00"*16, "\x00"*32, 3, 10, 1234)
193
194         data = { "count_happiness": 5,
195                  "count_shares_needed": 3,
196                  "count_shares_expected": 10,
197                  "count_shares_good": 6,
198                  "count_good_share_hosts": 7,
199                  "count_recoverable_versions": 1,
200                  "count_unrecoverable_versions": 0,
201                  "servers_responding": [],
202                  "sharemap": {"shareid1": [sb.get_stub_server(serverid_1),
203                                            sb.get_stub_server(serverid_f)]},
204                  "count_wrong_shares": 0,
205                  "list_corrupt_shares": [],
206                  "count_corrupt_shares": 0,
207                  "list_incompatible_shares": [],
208                  "count_incompatible_shares": 0,
209                  "report": [], "share_problems": [], "servermap": None,
210                  }
211         pre_cr = check_results.CheckResults(u, u.get_storage_index(),
212                                             healthy=False, recoverable=True,
213                                             summary="illing",
214                                             **data)
215
216         data = { "count_happiness": 9,
217                  "count_shares_needed": 3,
218                  "count_shares_expected": 10,
219                  "count_shares_good": 10,
220                  "count_good_share_hosts": 11,
221                  "count_recoverable_versions": 1,
222                  "count_unrecoverable_versions": 0,
223                  "servers_responding": [],
224                  "sharemap": {"shareid1": [sb.get_stub_server(serverid_1),
225                                            sb.get_stub_server(serverid_f)]},
226                  "count_wrong_shares": 0,
227                  "count_corrupt_shares": 0,
228                  "list_corrupt_shares": [],
229                  "list_incompatible_shares": [],
230                  "count_incompatible_shares": 0,
231                  "report": [], "share_problems": [], "servermap": None,
232                  }
233         post_cr = check_results.CheckResults(u, u.get_storage_index(),
234                                              healthy=True, recoverable=True,
235                                              summary="groovy",
236                                              **data)
237
238         crr = check_results.CheckAndRepairResults(u.get_storage_index())
239         crr.pre_repair_results = pre_cr
240         crr.post_repair_results = post_cr
241         crr.repair_attempted = False
242
243         w = web_check_results.CheckAndRepairResultsRenderer(c, crr)
244         html = self.render2(w)
245         s = self.remove_tags(html)
246
247         self.failUnlessIn("File Check-And-Repair Results for SI=2k6avp", s)
248         self.failUnlessIn("Healthy : groovy", s)
249         self.failUnlessIn("No repair necessary", s)
250         self.failUnlessIn("Post-Repair Checker Results:", s)
251         self.failUnlessIn("Share Counts: need 3-of-10, have 10", s)
252
253         crr.repair_attempted = True
254         crr.repair_successful = True
255         html = self.render2(w)
256         s = self.remove_tags(html)
257
258         self.failUnlessIn("File Check-And-Repair Results for SI=2k6avp", s)
259         self.failUnlessIn("Healthy : groovy", s)
260         self.failUnlessIn("Repair successful", s)
261         self.failUnlessIn("Post-Repair Checker Results:", s)
262
263         crr.repair_attempted = True
264         crr.repair_successful = False
265         post_cr = check_results.CheckResults(u, u.get_storage_index(),
266                                              healthy=False, recoverable=True,
267                                              summary="better",
268                                              **data)
269         crr.post_repair_results = post_cr
270         html = self.render2(w)
271         s = self.remove_tags(html)
272
273         self.failUnlessIn("File Check-And-Repair Results for SI=2k6avp", s)
274         self.failUnlessIn("Not Healthy! : better", s)
275         self.failUnlessIn("Repair unsuccessful", s)
276         self.failUnlessIn("Post-Repair Checker Results:", s)
277
278         crr.repair_attempted = True
279         crr.repair_successful = False
280         post_cr = check_results.CheckResults(u, u.get_storage_index(),
281                                              healthy=False, recoverable=False,
282                                              summary="worse",
283                                              **data)
284         crr.post_repair_results = post_cr
285         html = self.render2(w)
286         s = self.remove_tags(html)
287
288         self.failUnlessIn("File Check-And-Repair Results for SI=2k6avp", s)
289         self.failUnlessIn("Not Recoverable! : worse", s)
290         self.failUnlessIn("Repair unsuccessful", s)
291         self.failUnlessIn("Post-Repair Checker Results:", s)
292
293         d = self.render_json(w)
294         def _got_json(data):
295             j = simplejson.loads(data)
296             self.failUnlessEqual(j["repair-attempted"], True)
297             self.failUnlessEqual(j["storage-index"],
298                                  "2k6avpjga3dho3zsjo6nnkt7n4")
299             self.failUnlessEqual(j["pre-repair-results"]["summary"], "illing")
300             self.failUnlessEqual(j["post-repair-results"]["summary"], "worse")
301         d.addCallback(_got_json)
302
303         w2 = web_check_results.CheckAndRepairResultsRenderer(c, None)
304         d.addCallback(lambda ignored: self.render_json(w2))
305         def _got_lit_results(data):
306             j = simplejson.loads(data)
307             self.failUnlessEqual(j["repair-attempted"], False)
308             self.failUnlessEqual(j["storage-index"], "")
309         d.addCallback(_got_lit_results)
310         return d
311
312 class BalancingAct(GridTestMixin, unittest.TestCase):
313     # test for #1115 regarding the 'count-good-share-hosts' metric
314
315
316     def add_server(self, server_number, readonly=False):
317         assert self.g, "I tried to find a grid at self.g, but failed"
318         ss = self.g.make_server(server_number, readonly)
319         #log.msg("just created a server, number: %s => %s" % (server_number, ss,))
320         self.g.add_server(server_number, ss)
321
322     def add_server_with_share(self, server_number, uri, share_number=None,
323                               readonly=False):
324         self.add_server(server_number, readonly)
325         if share_number is not None:
326             self.copy_share_to_server(uri, share_number, server_number)
327
328     def copy_share_to_server(self, uri, share_number, server_number):
329         ss = self.g.servers_by_number[server_number]
330         # Copy share i from the directory associated with the first
331         # storage server to the directory associated with this one.
332         assert self.g, "I tried to find a grid at self.g, but failed"
333         assert self.shares, "I tried to find shares at self.shares, but failed"
334         old_share_location = self.shares[share_number][2]
335         new_share_location = os.path.join(ss.storedir, "shares")
336         si = tahoe_uri.from_string(self.uri).get_storage_index()
337         new_share_location = os.path.join(new_share_location,
338                                           storage_index_to_dir(si))
339         if not os.path.exists(new_share_location):
340             os.makedirs(new_share_location)
341         new_share_location = os.path.join(new_share_location,
342                                           str(share_number))
343         if old_share_location != new_share_location:
344             shutil.copy(old_share_location, new_share_location)
345         shares = self.find_uri_shares(uri)
346         # Make sure that the storage server has the share.
347         self.failUnless((share_number, ss.my_nodeid, new_share_location)
348                         in shares)
349
350     def _pretty_shares_chart(self, uri):
351         # Servers are labeled A-Z, shares are labeled 0-9
352         letters = 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'
353         assert len(self.g.servers_by_number) < len(letters), \
354             "This little printing function is only meant for < 26 servers"
355         shares_chart = {}
356         names = dict(zip([ss.my_nodeid
357                           for _,ss in self.g.servers_by_number.iteritems()],
358                          letters))
359         for shnum, serverid, _ in self.find_uri_shares(uri):
360             shares_chart.setdefault(shnum, []).append(names[serverid])
361         return shares_chart
362
363     def test_good_share_hosts(self):
364         self.basedir = "checker/BalancingAct/1115"
365         self.set_up_grid(num_servers=1)
366         c0 = self.g.clients[0]
367         c0.encoding_params['happy'] = 1
368         c0.encoding_params['n'] = 4
369         c0.encoding_params['k'] = 3
370
371         DATA = "data" * 100
372         d = c0.upload(Data(DATA, convergence=""))
373         def _stash_immutable(ur):
374             self.imm = c0.create_node_from_uri(ur.get_uri())
375             self.uri = self.imm.get_uri()
376         d.addCallback(_stash_immutable)
377         d.addCallback(lambda ign:
378             self.find_uri_shares(self.uri))
379         def _store_shares(shares):
380             self.shares = shares
381         d.addCallback(_store_shares)
382
383         def add_three(_, i):
384             # Add a new server with just share 3
385             self.add_server_with_share(i, self.uri, 3)
386             #print self._pretty_shares_chart(self.uri)
387         for i in range(1,5):
388             d.addCallback(add_three, i)
389
390         def _check_and_repair(_):
391             return self.imm.check_and_repair(Monitor())
392         def _check_counts(crr, shares_good, good_share_hosts):
393             prr = crr.get_post_repair_results()
394             #print self._pretty_shares_chart(self.uri)
395             self.failUnlessEqual(prr.get_share_counter_good(), shares_good)
396             self.failUnlessEqual(prr.get_host_counter_good_shares(),
397                                  good_share_hosts)
398
399         """
400         Initial sharemap:
401             0:[A] 1:[A] 2:[A] 3:[A,B,C,D,E]
402           4 good shares, but 5 good hosts
403         After deleting all instances of share #3 and repairing:
404             0:[A,B], 1:[A,C], 2:[A,D], 3:[E]
405           Still 4 good shares and 5 good hosts
406             """
407         d.addCallback(_check_and_repair)
408         d.addCallback(_check_counts, 4, 5)
409         d.addCallback(lambda _: self.delete_shares_numbered(self.uri, [3]))
410         d.addCallback(_check_and_repair)
411         d.addCallback(_check_counts, 4, 5)
412         d.addCallback(lambda _: [self.g.break_server(sid)
413                                  for sid in self.g.get_all_serverids()])
414         d.addCallback(_check_and_repair)
415         d.addCallback(_check_counts, 0, 0)
416         return d
417
418 class AddLease(GridTestMixin, unittest.TestCase):
419     # test for #875, in which failures in the add-lease call cause
420     # false-negatives in the checker
421
422     def test_875(self):
423         self.basedir = "checker/AddLease/875"
424         self.set_up_grid(num_servers=1)
425         c0 = self.g.clients[0]
426         c0.encoding_params['happy'] = 1
427         self.uris = {}
428         DATA = "data" * 100
429         d = c0.upload(Data(DATA, convergence=""))
430         def _stash_immutable(ur):
431             self.imm = c0.create_node_from_uri(ur.get_uri())
432         d.addCallback(_stash_immutable)
433         d.addCallback(lambda ign:
434             c0.create_mutable_file(MutableData("contents")))
435         def _stash_mutable(node):
436             self.mut = node
437         d.addCallback(_stash_mutable)
438
439         def _check_cr(cr, which):
440             self.failUnless(cr.is_healthy(), which)
441
442         # these two should work normally
443         d.addCallback(lambda ign: self.imm.check(Monitor(), add_lease=True))
444         d.addCallback(_check_cr, "immutable-normal")
445         d.addCallback(lambda ign: self.mut.check(Monitor(), add_lease=True))
446         d.addCallback(_check_cr, "mutable-normal")
447
448         really_did_break = []
449         # now break the server's remote_add_lease call
450         def _break_add_lease(ign):
451             def broken_add_lease(*args, **kwargs):
452                 really_did_break.append(1)
453                 raise KeyError("intentional failure, should be ignored")
454             assert self.g.servers_by_number[0].remote_add_lease
455             self.g.servers_by_number[0].remote_add_lease = broken_add_lease
456         d.addCallback(_break_add_lease)
457
458         # and confirm that the files still look healthy
459         d.addCallback(lambda ign: self.mut.check(Monitor(), add_lease=True))
460         d.addCallback(_check_cr, "mutable-broken")
461         d.addCallback(lambda ign: self.imm.check(Monitor(), add_lease=True))
462         d.addCallback(_check_cr, "immutable-broken")
463
464         d.addCallback(lambda ign: self.failUnless(really_did_break))
465         return d
466
467 class CounterHolder(object):
468     def __init__(self):
469         self._num_active_block_fetches = 0
470         self._max_active_block_fetches = 0
471
472 from allmydata.immutable.checker import ValidatedReadBucketProxy
473 class MockVRBP(ValidatedReadBucketProxy):
474     def __init__(self, sharenum, bucket, share_hash_tree, num_blocks, block_size, share_size, counterholder):
475         ValidatedReadBucketProxy.__init__(self, sharenum, bucket,
476                                           share_hash_tree, num_blocks,
477                                           block_size, share_size)
478         self.counterholder = counterholder
479
480     def get_block(self, blocknum):
481         self.counterholder._num_active_block_fetches += 1
482         if self.counterholder._num_active_block_fetches > self.counterholder._max_active_block_fetches:
483             self.counterholder._max_active_block_fetches = self.counterholder._num_active_block_fetches
484         d = ValidatedReadBucketProxy.get_block(self, blocknum)
485         def _mark_no_longer_active(res):
486             self.counterholder._num_active_block_fetches -= 1
487             return res
488         d.addBoth(_mark_no_longer_active)
489         return d
490
491 class TooParallel(GridTestMixin, unittest.TestCase):
492     # bug #1395: immutable verifier was aggressively parallized, checking all
493     # blocks of all shares at the same time, blowing our memory budget and
494     # crashing with MemoryErrors on >1GB files.
495
496     def test_immutable(self):
497         import allmydata.immutable.checker
498         origVRBP = allmydata.immutable.checker.ValidatedReadBucketProxy
499
500         self.basedir = "checker/TooParallel/immutable"
501
502         # If any code asks to instantiate a ValidatedReadBucketProxy,
503         # we give them a MockVRBP which is configured to use our
504         # CounterHolder.
505         counterholder = CounterHolder()
506         def make_mock_VRBP(*args, **kwargs):
507             return MockVRBP(counterholder=counterholder, *args, **kwargs)
508         allmydata.immutable.checker.ValidatedReadBucketProxy = make_mock_VRBP
509
510         d = defer.succeed(None)
511         def _start(ign):
512             self.set_up_grid(num_servers=4)
513             self.c0 = self.g.clients[0]
514             self.c0.encoding_params = { "k": 1,
515                                         "happy": 4,
516                                         "n": 4,
517                                         "max_segment_size": 5,
518                                       }
519             self.uris = {}
520             DATA = "data" * 100 # 400/5 = 80 blocks
521             return self.c0.upload(Data(DATA, convergence=""))
522         d.addCallback(_start)
523         def _do_check(ur):
524             n = self.c0.create_node_from_uri(ur.get_uri())
525             return n.check(Monitor(), verify=True)
526         d.addCallback(_do_check)
527         def _check(cr):
528             # the verifier works on all 4 shares in parallel, but only
529             # fetches one block from each share at a time, so we expect to
530             # see 4 parallel fetches
531             self.failUnlessEqual(counterholder._max_active_block_fetches, 4)
532         d.addCallback(_check)
533         def _clean_up(res):
534             allmydata.immutable.checker.ValidatedReadBucketProxy = origVRBP
535             return res
536         d.addBoth(_clean_up)
537         return d
538
539     test_immutable.timeout = 80