]> git.rkrishnan.org Git - tahoe-lafs/tahoe-lafs.git/blob - src/allmydata/test/test_checker.py
CheckResults: use fat init, add type-checking assertions
[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         serverid_1 = "\x00"*20
82         serverid_f = "\xff"*20
83         u = uri.CHKFileURI("\x00"*16, "\x00"*32, 3, 10, 1234)
84         data = { "count_shares_needed": 3,
85                  "count_shares_expected": 9,
86                  "count_shares_good": 10,
87                  "count_good_share_hosts": 11,
88                  "count_recoverable_versions": 1,
89                  "count_unrecoverable_versions": 0,
90                  "servers_responding": [],
91                  "sharemap": {"shareid1": [serverid_1, serverid_f]},
92                  "count_wrong_shares": 0,
93                  "list_corrupt_shares": [],
94                  "count_corrupt_shares": 0,
95                  "list_incompatible_shares": [],
96                  "count_incompatible_shares": 0,
97                  "report": [], "share_problems": [], "servermap": None,
98                  }
99         cr = check_results.CheckResults(u, u.get_storage_index(),
100                                         healthy=True, recoverable=True,
101                                         needs_rebalancing=False,
102                                         summary="groovy",
103                                         **data)
104         w = web_check_results.CheckResultsRenderer(c, cr)
105         html = self.render2(w)
106         s = self.remove_tags(html)
107         self.failUnlessIn("File Check Results for SI=2k6avp", s) # abbreviated
108         self.failUnlessIn("Healthy : groovy", s)
109         self.failUnlessIn("Share Counts: need 3-of-9, have 10", s)
110         self.failUnlessIn("Hosts with good shares: 11", s)
111         self.failUnlessIn("Corrupt shares: none", s)
112         self.failUnlessIn("Wrong Shares: 0", s)
113         self.failUnlessIn("Recoverable Versions: 1", s)
114         self.failUnlessIn("Unrecoverable Versions: 0", s)
115
116         cr = check_results.CheckResults(u, u.get_storage_index(),
117                                         healthy=False, recoverable=True,
118                                         needs_rebalancing=False,
119                                         summary="ungroovy",
120                                         **data)
121         w = web_check_results.CheckResultsRenderer(c, cr)
122         html = self.render2(w)
123         s = self.remove_tags(html)
124         self.failUnlessIn("File Check Results for SI=2k6avp", s) # abbreviated
125         self.failUnlessIn("Not Healthy! : ungroovy", s)
126
127         data["count_corrupt_shares"] = 1
128         data["list_corrupt_shares"] = [(serverid_1, u.get_storage_index(), 2)]
129         cr = check_results.CheckResults(u, u.get_storage_index(),
130                                         healthy=False, recoverable=False,
131                                         needs_rebalancing=False,
132                                         summary="rather dead",
133                                         **data)
134         w = web_check_results.CheckResultsRenderer(c, cr)
135         html = self.render2(w)
136         s = self.remove_tags(html)
137         self.failUnlessIn("File Check Results for SI=2k6avp", s) # abbreviated
138         self.failUnlessIn("Not Recoverable! : rather dead", s)
139         self.failUnlessIn("Corrupt shares: Share ID Nickname Node ID sh#2 peer-0 aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa", s)
140
141         html = self.render2(w)
142         s = self.remove_tags(html)
143         self.failUnlessIn("File Check Results for SI=2k6avp", s) # abbreviated
144         self.failUnlessIn("Not Recoverable! : rather dead", s)
145
146         html = self.render2(w, args={"return_to": ["FOOURL"]})
147         self.failUnlessIn('<a href="FOOURL">Return to file/directory.</a>',
148                           html)
149
150         d = self.render_json(w)
151         def _check_json(jdata):
152             j = simplejson.loads(jdata)
153             self.failUnlessEqual(j["summary"], "rather dead")
154             self.failUnlessEqual(j["storage-index"],
155                                  "2k6avpjga3dho3zsjo6nnkt7n4")
156             expected = {'needs-rebalancing': False,
157                         'count-shares-expected': 9,
158                         'healthy': False,
159                         'count-unrecoverable-versions': 0,
160                         'count-shares-needed': 3,
161                         'sharemap': {"shareid1":
162                                      ["aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa",
163                                       "77777777777777777777777777777777"]},
164                         'count-recoverable-versions': 1,
165                         'list-corrupt-shares':
166                         [["aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa",
167                           "2k6avpjga3dho3zsjo6nnkt7n4", 2]],
168                         'count-good-share-hosts': 11,
169                         'count-wrong-shares': 0,
170                         'count-shares-good': 10,
171                         'count-corrupt-shares': 1,
172                         'servers-responding': [],
173                         'recoverable': False,
174                         }
175             self.failUnlessEqual(j["results"], expected)
176         d.addCallback(_check_json)
177         d.addCallback(lambda ignored: self.render1(w))
178         def _check(html):
179             s = self.remove_tags(html)
180             self.failUnlessIn("File Check Results for SI=2k6avp", s)
181             self.failUnlessIn("Not Recoverable! : rather dead", s)
182         d.addCallback(_check)
183         return d
184
185
186     def test_check_and_repair(self):
187         c = self.create_fake_client()
188         serverid_1 = "\x00"*20
189         serverid_f = "\xff"*20
190         u = uri.CHKFileURI("\x00"*16, "\x00"*32, 3, 10, 1234)
191
192         data = { "count_shares_needed": 3,
193                  "count_shares_expected": 10,
194                  "count_shares_good": 6,
195                  "count_good_share_hosts": 7,
196                  "count_recoverable_versions": 1,
197                  "count_unrecoverable_versions": 0,
198                  "servers_responding": [],
199                  "sharemap": {"shareid1": [serverid_1, serverid_f]},
200                  "count_wrong_shares": 0,
201                  "list_corrupt_shares": [],
202                  "count_corrupt_shares": 0,
203                  "list_incompatible_shares": [],
204                  "count_incompatible_shares": 0,
205                  "report": [], "share_problems": [], "servermap": None,
206                  }
207         pre_cr = check_results.CheckResults(u, u.get_storage_index(),
208                                             healthy=False, recoverable=True,
209                                             needs_rebalancing=False,
210                                             summary="illing",
211                                             **data)
212
213         data = { "count_shares_needed": 3,
214                  "count_shares_expected": 10,
215                  "count_shares_good": 10,
216                  "count_good_share_hosts": 11,
217                  "count_recoverable_versions": 1,
218                  "count_unrecoverable_versions": 0,
219                  "servers_responding": [],
220                  "sharemap": {"shareid1": [serverid_1, serverid_f]},
221                  "count_wrong_shares": 0,
222                  "count_corrupt_shares": 0,
223                  "list_corrupt_shares": [],
224                  "list_incompatible_shares": [],
225                  "count_incompatible_shares": 0,
226                  "report": [], "share_problems": [], "servermap": None,
227                  }
228         post_cr = check_results.CheckResults(u, u.get_storage_index(),
229                                              healthy=True, recoverable=True,
230                                              needs_rebalancing=False,
231                                              summary="groovy",
232                                              **data)
233
234         crr = check_results.CheckAndRepairResults(u.get_storage_index())
235         crr.pre_repair_results = pre_cr
236         crr.post_repair_results = post_cr
237         crr.repair_attempted = False
238
239         w = web_check_results.CheckAndRepairResultsRenderer(c, crr)
240         html = self.render2(w)
241         s = self.remove_tags(html)
242
243         self.failUnlessIn("File Check-And-Repair Results for SI=2k6avp", s)
244         self.failUnlessIn("Healthy : groovy", s)
245         self.failUnlessIn("No repair necessary", s)
246         self.failUnlessIn("Post-Repair Checker Results:", s)
247         self.failUnlessIn("Share Counts: need 3-of-10, have 10", s)
248
249         crr.repair_attempted = True
250         crr.repair_successful = True
251         html = self.render2(w)
252         s = self.remove_tags(html)
253
254         self.failUnlessIn("File Check-And-Repair Results for SI=2k6avp", s)
255         self.failUnlessIn("Healthy : groovy", s)
256         self.failUnlessIn("Repair successful", s)
257         self.failUnlessIn("Post-Repair Checker Results:", s)
258
259         crr.repair_attempted = True
260         crr.repair_successful = False
261         post_cr = check_results.CheckResults(u, u.get_storage_index(),
262                                              healthy=False, recoverable=True,
263                                              needs_rebalancing=False,
264                                              summary="better",
265                                              **data)
266         crr.post_repair_results = post_cr
267         html = self.render2(w)
268         s = self.remove_tags(html)
269
270         self.failUnlessIn("File Check-And-Repair Results for SI=2k6avp", s)
271         self.failUnlessIn("Not Healthy! : better", s)
272         self.failUnlessIn("Repair unsuccessful", s)
273         self.failUnlessIn("Post-Repair Checker Results:", s)
274
275         crr.repair_attempted = True
276         crr.repair_successful = False
277         post_cr = check_results.CheckResults(u, u.get_storage_index(),
278                                              healthy=False, recoverable=False,
279                                              needs_rebalancing=False,
280                                              summary="worse",
281                                              **data)
282         crr.post_repair_results = post_cr
283         html = self.render2(w)
284         s = self.remove_tags(html)
285
286         self.failUnlessIn("File Check-And-Repair Results for SI=2k6avp", s)
287         self.failUnlessIn("Not Recoverable! : worse", s)
288         self.failUnlessIn("Repair unsuccessful", s)
289         self.failUnlessIn("Post-Repair Checker Results:", s)
290
291         d = self.render_json(w)
292         def _got_json(data):
293             j = simplejson.loads(data)
294             self.failUnlessEqual(j["repair-attempted"], True)
295             self.failUnlessEqual(j["storage-index"],
296                                  "2k6avpjga3dho3zsjo6nnkt7n4")
297             self.failUnlessEqual(j["pre-repair-results"]["summary"], "illing")
298             self.failUnlessEqual(j["post-repair-results"]["summary"], "worse")
299         d.addCallback(_got_json)
300
301         w2 = web_check_results.CheckAndRepairResultsRenderer(c, None)
302         d.addCallback(lambda ignored: self.render_json(w2))
303         def _got_lit_results(data):
304             j = simplejson.loads(data)
305             self.failUnlessEqual(j["repair-attempted"], False)
306             self.failUnlessEqual(j["storage-index"], "")
307         d.addCallback(_got_lit_results)
308         return d
309
310 class BalancingAct(GridTestMixin, unittest.TestCase):
311     # test for #1115 regarding the 'count-good-share-hosts' metric
312
313
314     def add_server(self, server_number, readonly=False):
315         assert self.g, "I tried to find a grid at self.g, but failed"
316         ss = self.g.make_server(server_number, readonly)
317         #log.msg("just created a server, number: %s => %s" % (server_number, ss,))
318         self.g.add_server(server_number, ss)
319
320     def add_server_with_share(self, server_number, uri, share_number=None,
321                               readonly=False):
322         self.add_server(server_number, readonly)
323         if share_number is not None:
324             self.copy_share_to_server(uri, share_number, server_number)
325
326     def copy_share_to_server(self, uri, share_number, server_number):
327         ss = self.g.servers_by_number[server_number]
328         # Copy share i from the directory associated with the first
329         # storage server to the directory associated with this one.
330         assert self.g, "I tried to find a grid at self.g, but failed"
331         assert self.shares, "I tried to find shares at self.shares, but failed"
332         old_share_location = self.shares[share_number][2]
333         new_share_location = os.path.join(ss.storedir, "shares")
334         si = tahoe_uri.from_string(self.uri).get_storage_index()
335         new_share_location = os.path.join(new_share_location,
336                                           storage_index_to_dir(si))
337         if not os.path.exists(new_share_location):
338             os.makedirs(new_share_location)
339         new_share_location = os.path.join(new_share_location,
340                                           str(share_number))
341         if old_share_location != new_share_location:
342             shutil.copy(old_share_location, new_share_location)
343         shares = self.find_uri_shares(uri)
344         # Make sure that the storage server has the share.
345         self.failUnless((share_number, ss.my_nodeid, new_share_location)
346                         in shares)
347
348     def _pretty_shares_chart(self, uri):
349         # Servers are labeled A-Z, shares are labeled 0-9
350         letters = 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'
351         assert len(self.g.servers_by_number) < len(letters), \
352             "This little printing function is only meant for < 26 servers"
353         shares_chart = {}
354         names = dict(zip([ss.my_nodeid
355                           for _,ss in self.g.servers_by_number.iteritems()],
356                          letters))
357         for shnum, serverid, _ in self.find_uri_shares(uri):
358             shares_chart.setdefault(shnum, []).append(names[serverid])
359         return shares_chart
360
361     def test_good_share_hosts(self):
362         self.basedir = "checker/BalancingAct/1115"
363         self.set_up_grid(num_servers=1)
364         c0 = self.g.clients[0]
365         c0.DEFAULT_ENCODING_PARAMETERS['happy'] = 1
366         c0.DEFAULT_ENCODING_PARAMETERS['n'] = 4
367         c0.DEFAULT_ENCODING_PARAMETERS['k'] = 3
368
369         DATA = "data" * 100
370         d = c0.upload(Data(DATA, convergence=""))
371         def _stash_immutable(ur):
372             self.imm = c0.create_node_from_uri(ur.get_uri())
373             self.uri = self.imm.get_uri()
374         d.addCallback(_stash_immutable)
375         d.addCallback(lambda ign:
376             self.find_uri_shares(self.uri))
377         def _store_shares(shares):
378             self.shares = shares
379         d.addCallback(_store_shares)
380
381         def add_three(_, i):
382             # Add a new server with just share 3
383             self.add_server_with_share(i, self.uri, 3)
384             #print self._pretty_shares_chart(self.uri)
385         for i in range(1,5):
386             d.addCallback(add_three, i)
387
388         def _check_and_repair(_):
389             return self.imm.check_and_repair(Monitor())
390         def _check_counts(crr, shares_good, good_share_hosts):
391             prr = crr.get_post_repair_results()
392             #print self._pretty_shares_chart(self.uri)
393             self.failUnlessEqual(prr.get_share_counter_good(), shares_good)
394             self.failUnlessEqual(prr.get_host_counter_good_shares(),
395                                  good_share_hosts)
396
397         """
398         Initial sharemap:
399             0:[A] 1:[A] 2:[A] 3:[A,B,C,D,E]
400           4 good shares, but 5 good hosts
401         After deleting all instances of share #3 and repairing:
402             0:[A,B], 1:[A,C], 2:[A,D], 3:[E]
403           Still 4 good shares and 5 good hosts
404             """
405         d.addCallback(_check_and_repair)
406         d.addCallback(_check_counts, 4, 5)
407         d.addCallback(lambda _: self.delete_shares_numbered(self.uri, [3]))
408         d.addCallback(_check_and_repair)
409         d.addCallback(_check_counts, 4, 5)
410         d.addCallback(lambda _: [self.g.break_server(sid)
411                                  for sid in self.g.get_all_serverids()])
412         d.addCallback(_check_and_repair)
413         d.addCallback(_check_counts, 0, 0)
414         return d
415
416 class AddLease(GridTestMixin, unittest.TestCase):
417     # test for #875, in which failures in the add-lease call cause
418     # false-negatives in the checker
419
420     def test_875(self):
421         self.basedir = "checker/AddLease/875"
422         self.set_up_grid(num_servers=1)
423         c0 = self.g.clients[0]
424         c0.DEFAULT_ENCODING_PARAMETERS['happy'] = 1
425         self.uris = {}
426         DATA = "data" * 100
427         d = c0.upload(Data(DATA, convergence=""))
428         def _stash_immutable(ur):
429             self.imm = c0.create_node_from_uri(ur.get_uri())
430         d.addCallback(_stash_immutable)
431         d.addCallback(lambda ign:
432             c0.create_mutable_file(MutableData("contents")))
433         def _stash_mutable(node):
434             self.mut = node
435         d.addCallback(_stash_mutable)
436
437         def _check_cr(cr, which):
438             self.failUnless(cr.is_healthy(), which)
439
440         # these two should work normally
441         d.addCallback(lambda ign: self.imm.check(Monitor(), add_lease=True))
442         d.addCallback(_check_cr, "immutable-normal")
443         d.addCallback(lambda ign: self.mut.check(Monitor(), add_lease=True))
444         d.addCallback(_check_cr, "mutable-normal")
445
446         really_did_break = []
447         # now break the server's remote_add_lease call
448         def _break_add_lease(ign):
449             def broken_add_lease(*args, **kwargs):
450                 really_did_break.append(1)
451                 raise KeyError("intentional failure, should be ignored")
452             assert self.g.servers_by_number[0].remote_add_lease
453             self.g.servers_by_number[0].remote_add_lease = broken_add_lease
454         d.addCallback(_break_add_lease)
455
456         # and confirm that the files still look healthy
457         d.addCallback(lambda ign: self.mut.check(Monitor(), add_lease=True))
458         d.addCallback(_check_cr, "mutable-broken")
459         d.addCallback(lambda ign: self.imm.check(Monitor(), add_lease=True))
460         d.addCallback(_check_cr, "immutable-broken")
461
462         d.addCallback(lambda ign: self.failUnless(really_did_break))
463         return d
464
465 class CounterHolder(object):
466     def __init__(self):
467         self._num_active_block_fetches = 0
468         self._max_active_block_fetches = 0
469
470 from allmydata.immutable.checker import ValidatedReadBucketProxy
471 class MockVRBP(ValidatedReadBucketProxy):
472     def __init__(self, sharenum, bucket, share_hash_tree, num_blocks, block_size, share_size, counterholder):
473         ValidatedReadBucketProxy.__init__(self, sharenum, bucket,
474                                           share_hash_tree, num_blocks,
475                                           block_size, share_size)
476         self.counterholder = counterholder
477
478     def get_block(self, blocknum):
479         self.counterholder._num_active_block_fetches += 1
480         if self.counterholder._num_active_block_fetches > self.counterholder._max_active_block_fetches:
481             self.counterholder._max_active_block_fetches = self.counterholder._num_active_block_fetches
482         d = ValidatedReadBucketProxy.get_block(self, blocknum)
483         def _mark_no_longer_active(res):
484             self.counterholder._num_active_block_fetches -= 1
485             return res
486         d.addBoth(_mark_no_longer_active)
487         return d
488
489 class TooParallel(GridTestMixin, unittest.TestCase):
490     # bug #1395: immutable verifier was aggressively parallized, checking all
491     # blocks of all shares at the same time, blowing our memory budget and
492     # crashing with MemoryErrors on >1GB files.
493
494     def test_immutable(self):
495         import allmydata.immutable.checker
496         origVRBP = allmydata.immutable.checker.ValidatedReadBucketProxy
497
498         self.basedir = "checker/TooParallel/immutable"
499
500         # If any code asks to instantiate a ValidatedReadBucketProxy,
501         # we give them a MockVRBP which is configured to use our
502         # CounterHolder.
503         counterholder = CounterHolder()
504         def make_mock_VRBP(*args, **kwargs):
505             return MockVRBP(counterholder=counterholder, *args, **kwargs)
506         allmydata.immutable.checker.ValidatedReadBucketProxy = make_mock_VRBP
507
508         d = defer.succeed(None)
509         def _start(ign):
510             self.set_up_grid(num_servers=4)
511             self.c0 = self.g.clients[0]
512             self.c0.DEFAULT_ENCODING_PARAMETERS = { "k": 1,
513                                                "happy": 4,
514                                                "n": 4,
515                                                "max_segment_size": 5,
516                                                }
517             self.uris = {}
518             DATA = "data" * 100 # 400/5 = 80 blocks
519             return self.c0.upload(Data(DATA, convergence=""))
520         d.addCallback(_start)
521         def _do_check(ur):
522             n = self.c0.create_node_from_uri(ur.get_uri())
523             return n.check(Monitor(), verify=True)
524         d.addCallback(_do_check)
525         def _check(cr):
526             # the verifier works on all 4 shares in parallel, but only
527             # fetches one block from each share at a time, so we expect to
528             # see 4 parallel fetches
529             self.failUnlessEqual(counterholder._max_active_block_fetches, 4)
530         d.addCallback(_check)
531         def _clean_up(res):
532             allmydata.immutable.checker.ValidatedReadBucketProxy = origVRBP
533             return res
534         d.addBoth(_clean_up)
535         return d
536
537     test_immutable.timeout = 80