Merge branch 'master' into 8857-cwl-job-reuse
[arvados.git] / sdk / python / tests / test_keep_client.py
1 import hashlib
2 import mock
3 import os
4 import pycurl
5 import random
6 import re
7 import socket
8 import threading
9 import time
10 import unittest
11 import urlparse
12
13 import arvados
14 import arvados.retry
15 import arvados_testutil as tutil
16 import keepstub
17 import run_test_server
18
19 class KeepTestCase(run_test_server.TestCaseWithServers):
20     MAIN_SERVER = {}
21     KEEP_SERVER = {}
22
23     @classmethod
24     def setUpClass(cls):
25         super(KeepTestCase, cls).setUpClass()
26         run_test_server.authorize_with("admin")
27         cls.api_client = arvados.api('v1')
28         cls.keep_client = arvados.KeepClient(api_client=cls.api_client,
29                                              proxy='', local_store='')
30
31     def test_KeepBasicRWTest(self):
32         self.assertEqual(0, self.keep_client.upload_counter.get())
33         foo_locator = self.keep_client.put('foo')
34         self.assertRegexpMatches(
35             foo_locator,
36             '^acbd18db4cc2f85cedef654fccc4a4d8\+3',
37             'wrong md5 hash from Keep.put("foo"): ' + foo_locator)
38
39         # 6 bytes because uploaded 2 copies
40         self.assertEqual(6, self.keep_client.upload_counter.get())
41
42         self.assertEqual(0, self.keep_client.download_counter.get())
43         self.assertEqual(self.keep_client.get(foo_locator),
44                          'foo',
45                          'wrong content from Keep.get(md5("foo"))')
46         self.assertEqual(3, self.keep_client.download_counter.get())
47
48     def test_KeepBinaryRWTest(self):
49         blob_str = '\xff\xfe\xf7\x00\x01\x02'
50         blob_locator = self.keep_client.put(blob_str)
51         self.assertRegexpMatches(
52             blob_locator,
53             '^7fc7c53b45e53926ba52821140fef396\+6',
54             ('wrong locator from Keep.put(<binarydata>):' + blob_locator))
55         self.assertEqual(self.keep_client.get(blob_locator),
56                          blob_str,
57                          'wrong content from Keep.get(md5(<binarydata>))')
58
59     def test_KeepLongBinaryRWTest(self):
60         blob_str = '\xff\xfe\xfd\xfc\x00\x01\x02\x03'
61         for i in range(0,23):
62             blob_str = blob_str + blob_str
63         blob_locator = self.keep_client.put(blob_str)
64         self.assertRegexpMatches(
65             blob_locator,
66             '^84d90fc0d8175dd5dcfab04b999bc956\+67108864',
67             ('wrong locator from Keep.put(<binarydata>): ' + blob_locator))
68         self.assertEqual(self.keep_client.get(blob_locator),
69                          blob_str,
70                          'wrong content from Keep.get(md5(<binarydata>))')
71
72     @unittest.skip("unreliable test - please fix and close #8752")
73     def test_KeepSingleCopyRWTest(self):
74         blob_str = '\xff\xfe\xfd\xfc\x00\x01\x02\x03'
75         blob_locator = self.keep_client.put(blob_str, copies=1)
76         self.assertRegexpMatches(
77             blob_locator,
78             '^c902006bc98a3eb4a3663b65ab4a6fab\+8',
79             ('wrong locator from Keep.put(<binarydata>): ' + blob_locator))
80         self.assertEqual(self.keep_client.get(blob_locator),
81                          blob_str,
82                          'wrong content from Keep.get(md5(<binarydata>))')
83
84     def test_KeepEmptyCollectionTest(self):
85         blob_locator = self.keep_client.put('', copies=1)
86         self.assertRegexpMatches(
87             blob_locator,
88             '^d41d8cd98f00b204e9800998ecf8427e\+0',
89             ('wrong locator from Keep.put(""): ' + blob_locator))
90
91     def test_unicode_must_be_ascii(self):
92         # If unicode type, must only consist of valid ASCII
93         foo_locator = self.keep_client.put(u'foo')
94         self.assertRegexpMatches(
95             foo_locator,
96             '^acbd18db4cc2f85cedef654fccc4a4d8\+3',
97             'wrong md5 hash from Keep.put("foo"): ' + foo_locator)
98
99         with self.assertRaises(UnicodeEncodeError):
100             # Error if it is not ASCII
101             self.keep_client.put(u'\xe2')
102
103         with self.assertRaises(arvados.errors.ArgumentError):
104             # Must be a string type
105             self.keep_client.put({})
106
107 class KeepPermissionTestCase(run_test_server.TestCaseWithServers):
108     MAIN_SERVER = {}
109     KEEP_SERVER = {'blob_signing_key': 'abcdefghijk0123456789',
110                    'enforce_permissions': True}
111
112     def test_KeepBasicRWTest(self):
113         run_test_server.authorize_with('active')
114         keep_client = arvados.KeepClient()
115         foo_locator = keep_client.put('foo')
116         self.assertRegexpMatches(
117             foo_locator,
118             r'^acbd18db4cc2f85cedef654fccc4a4d8\+3\+A[a-f0-9]+@[a-f0-9]+$',
119             'invalid locator from Keep.put("foo"): ' + foo_locator)
120         self.assertEqual(keep_client.get(foo_locator),
121                          'foo',
122                          'wrong content from Keep.get(md5("foo"))')
123
124         # GET with an unsigned locator => NotFound
125         bar_locator = keep_client.put('bar')
126         unsigned_bar_locator = "37b51d194a7513e45b56f6524f2d51f2+3"
127         self.assertRegexpMatches(
128             bar_locator,
129             r'^37b51d194a7513e45b56f6524f2d51f2\+3\+A[a-f0-9]+@[a-f0-9]+$',
130             'invalid locator from Keep.put("bar"): ' + bar_locator)
131         self.assertRaises(arvados.errors.NotFoundError,
132                           keep_client.get,
133                           unsigned_bar_locator)
134
135         # GET from a different user => NotFound
136         run_test_server.authorize_with('spectator')
137         self.assertRaises(arvados.errors.NotFoundError,
138                           arvados.Keep.get,
139                           bar_locator)
140
141         # Unauthenticated GET for a signed locator => NotFound
142         # Unauthenticated GET for an unsigned locator => NotFound
143         keep_client.api_token = ''
144         self.assertRaises(arvados.errors.NotFoundError,
145                           keep_client.get,
146                           bar_locator)
147         self.assertRaises(arvados.errors.NotFoundError,
148                           keep_client.get,
149                           unsigned_bar_locator)
150
151
152 # KeepOptionalPermission: starts Keep with --permission-key-file
153 # but not --enforce-permissions (i.e. generate signatures on PUT
154 # requests, but do not require them for GET requests)
155 #
156 # All of these requests should succeed when permissions are optional:
157 # * authenticated request, signed locator
158 # * authenticated request, unsigned locator
159 # * unauthenticated request, signed locator
160 # * unauthenticated request, unsigned locator
161 class KeepOptionalPermission(run_test_server.TestCaseWithServers):
162     MAIN_SERVER = {}
163     KEEP_SERVER = {'blob_signing_key': 'abcdefghijk0123456789',
164                    'enforce_permissions': False}
165
166     @classmethod
167     def setUpClass(cls):
168         super(KeepOptionalPermission, cls).setUpClass()
169         run_test_server.authorize_with("admin")
170         cls.api_client = arvados.api('v1')
171
172     def setUp(self):
173         super(KeepOptionalPermission, self).setUp()
174         self.keep_client = arvados.KeepClient(api_client=self.api_client,
175                                               proxy='', local_store='')
176
177     def _put_foo_and_check(self):
178         signed_locator = self.keep_client.put('foo')
179         self.assertRegexpMatches(
180             signed_locator,
181             r'^acbd18db4cc2f85cedef654fccc4a4d8\+3\+A[a-f0-9]+@[a-f0-9]+$',
182             'invalid locator from Keep.put("foo"): ' + signed_locator)
183         return signed_locator
184
185     def test_KeepAuthenticatedSignedTest(self):
186         signed_locator = self._put_foo_and_check()
187         self.assertEqual(self.keep_client.get(signed_locator),
188                          'foo',
189                          'wrong content from Keep.get(md5("foo"))')
190
191     def test_KeepAuthenticatedUnsignedTest(self):
192         signed_locator = self._put_foo_and_check()
193         self.assertEqual(self.keep_client.get("acbd18db4cc2f85cedef654fccc4a4d8"),
194                          'foo',
195                          'wrong content from Keep.get(md5("foo"))')
196
197     def test_KeepUnauthenticatedSignedTest(self):
198         # Check that signed GET requests work even when permissions
199         # enforcement is off.
200         signed_locator = self._put_foo_and_check()
201         self.keep_client.api_token = ''
202         self.assertEqual(self.keep_client.get(signed_locator),
203                          'foo',
204                          'wrong content from Keep.get(md5("foo"))')
205
206     def test_KeepUnauthenticatedUnsignedTest(self):
207         # Since --enforce-permissions is not in effect, GET requests
208         # need not be authenticated.
209         signed_locator = self._put_foo_and_check()
210         self.keep_client.api_token = ''
211         self.assertEqual(self.keep_client.get("acbd18db4cc2f85cedef654fccc4a4d8"),
212                          'foo',
213                          'wrong content from Keep.get(md5("foo"))')
214
215
216 class KeepProxyTestCase(run_test_server.TestCaseWithServers):
217     MAIN_SERVER = {}
218     KEEP_SERVER = {}
219     KEEP_PROXY_SERVER = {}
220
221     @classmethod
222     def setUpClass(cls):
223         super(KeepProxyTestCase, cls).setUpClass()
224         run_test_server.authorize_with('active')
225         cls.api_client = arvados.api('v1')
226
227     def tearDown(self):
228         arvados.config.settings().pop('ARVADOS_EXTERNAL_CLIENT', None)
229         super(KeepProxyTestCase, self).tearDown()
230
231     def test_KeepProxyTest1(self):
232         # Will use ARVADOS_KEEP_PROXY environment variable that is set by
233         # setUpClass().
234         keep_client = arvados.KeepClient(api_client=self.api_client,
235                                          local_store='')
236         baz_locator = keep_client.put('baz')
237         self.assertRegexpMatches(
238             baz_locator,
239             '^73feffa4b7f6bb68e44cf984c85f6e88\+3',
240             'wrong md5 hash from Keep.put("baz"): ' + baz_locator)
241         self.assertEqual(keep_client.get(baz_locator),
242                          'baz',
243                          'wrong content from Keep.get(md5("baz"))')
244         self.assertTrue(keep_client.using_proxy)
245
246     def test_KeepProxyTest2(self):
247         # Don't instantiate the proxy directly, but set the X-External-Client
248         # header.  The API server should direct us to the proxy.
249         arvados.config.settings()['ARVADOS_EXTERNAL_CLIENT'] = 'true'
250         keep_client = arvados.KeepClient(api_client=self.api_client,
251                                          proxy='', local_store='')
252         baz_locator = keep_client.put('baz2')
253         self.assertRegexpMatches(
254             baz_locator,
255             '^91f372a266fe2bf2823cb8ec7fda31ce\+4',
256             'wrong md5 hash from Keep.put("baz2"): ' + baz_locator)
257         self.assertEqual(keep_client.get(baz_locator),
258                          'baz2',
259                          'wrong content from Keep.get(md5("baz2"))')
260         self.assertTrue(keep_client.using_proxy)
261
262
263 class KeepClientServiceTestCase(unittest.TestCase, tutil.ApiClientMock):
264     def get_service_roots(self, api_client):
265         keep_client = arvados.KeepClient(api_client=api_client)
266         services = keep_client.weighted_service_roots(arvados.KeepLocator('0'*32))
267         return [urlparse.urlparse(url) for url in sorted(services)]
268
269     def test_ssl_flag_respected_in_roots(self):
270         for ssl_flag in [False, True]:
271             services = self.get_service_roots(self.mock_keep_services(
272                 service_ssl_flag=ssl_flag))
273             self.assertEqual(
274                 ('https' if ssl_flag else 'http'), services[0].scheme)
275
276     def test_correct_ports_with_ipv6_addresses(self):
277         service = self.get_service_roots(self.mock_keep_services(
278             service_type='proxy', service_host='100::1', service_port=10, count=1))[0]
279         self.assertEqual('100::1', service.hostname)
280         self.assertEqual(10, service.port)
281
282     # test_*_timeout verify that KeepClient instructs pycurl to use
283     # the appropriate connection and read timeouts. They don't care
284     # whether pycurl actually exhibits the expected timeout behavior
285     # -- those tests are in the KeepClientTimeout test class.
286
287     def test_get_timeout(self):
288         api_client = self.mock_keep_services(count=1)
289         force_timeout = socket.timeout("timed out")
290         with tutil.mock_keep_responses(force_timeout, 0) as mock:
291             keep_client = arvados.KeepClient(api_client=api_client)
292             with self.assertRaises(arvados.errors.KeepReadError):
293                 keep_client.get('ffffffffffffffffffffffffffffffff')
294             self.assertEqual(
295                 mock.responses[0].getopt(pycurl.CONNECTTIMEOUT_MS),
296                 int(arvados.KeepClient.DEFAULT_TIMEOUT[0]*1000))
297             self.assertEqual(
298                 mock.responses[0].getopt(pycurl.LOW_SPEED_TIME),
299                 int(arvados.KeepClient.DEFAULT_TIMEOUT[1]))
300             self.assertEqual(
301                 mock.responses[0].getopt(pycurl.LOW_SPEED_LIMIT),
302                 int(arvados.KeepClient.DEFAULT_TIMEOUT[2]))
303
304     def test_put_timeout(self):
305         api_client = self.mock_keep_services(count=1)
306         force_timeout = socket.timeout("timed out")
307         with tutil.mock_keep_responses(force_timeout, 0) as mock:
308             keep_client = arvados.KeepClient(api_client=api_client)
309             with self.assertRaises(arvados.errors.KeepWriteError):
310                 keep_client.put('foo')
311             self.assertEqual(
312                 mock.responses[0].getopt(pycurl.CONNECTTIMEOUT_MS),
313                 int(arvados.KeepClient.DEFAULT_TIMEOUT[0]*1000))
314             self.assertEqual(
315                 mock.responses[0].getopt(pycurl.LOW_SPEED_TIME),
316                 int(arvados.KeepClient.DEFAULT_TIMEOUT[1]))
317             self.assertEqual(
318                 mock.responses[0].getopt(pycurl.LOW_SPEED_LIMIT),
319                 int(arvados.KeepClient.DEFAULT_TIMEOUT[2]))
320
321     def test_proxy_get_timeout(self):
322         api_client = self.mock_keep_services(service_type='proxy', count=1)
323         force_timeout = socket.timeout("timed out")
324         with tutil.mock_keep_responses(force_timeout, 0) as mock:
325             keep_client = arvados.KeepClient(api_client=api_client)
326             with self.assertRaises(arvados.errors.KeepReadError):
327                 keep_client.get('ffffffffffffffffffffffffffffffff')
328             self.assertEqual(
329                 mock.responses[0].getopt(pycurl.CONNECTTIMEOUT_MS),
330                 int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[0]*1000))
331             self.assertEqual(
332                 mock.responses[0].getopt(pycurl.LOW_SPEED_TIME),
333                 int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[1]))
334             self.assertEqual(
335                 mock.responses[0].getopt(pycurl.LOW_SPEED_LIMIT),
336                 int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[2]))
337
338     def test_proxy_put_timeout(self):
339         api_client = self.mock_keep_services(service_type='proxy', count=1)
340         force_timeout = socket.timeout("timed out")
341         with tutil.mock_keep_responses(force_timeout, 0) as mock:
342             keep_client = arvados.KeepClient(api_client=api_client)
343             with self.assertRaises(arvados.errors.KeepWriteError):
344                 keep_client.put('foo')
345             self.assertEqual(
346                 mock.responses[0].getopt(pycurl.CONNECTTIMEOUT_MS),
347                 int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[0]*1000))
348             self.assertEqual(
349                 mock.responses[0].getopt(pycurl.LOW_SPEED_TIME),
350                 int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[1]))
351             self.assertEqual(
352                 mock.responses[0].getopt(pycurl.LOW_SPEED_LIMIT),
353                 int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[2]))
354
355     def check_no_services_error(self, verb, exc_class):
356         api_client = mock.MagicMock(name='api_client')
357         api_client.keep_services().accessible().execute.side_effect = (
358             arvados.errors.ApiError)
359         keep_client = arvados.KeepClient(api_client=api_client)
360         with self.assertRaises(exc_class) as err_check:
361             getattr(keep_client, verb)('d41d8cd98f00b204e9800998ecf8427e+0')
362         self.assertEqual(0, len(err_check.exception.request_errors()))
363
364     def test_get_error_with_no_services(self):
365         self.check_no_services_error('get', arvados.errors.KeepReadError)
366
367     def test_put_error_with_no_services(self):
368         self.check_no_services_error('put', arvados.errors.KeepWriteError)
369
370     def check_errors_from_last_retry(self, verb, exc_class):
371         api_client = self.mock_keep_services(count=2)
372         req_mock = tutil.mock_keep_responses(
373             "retry error reporting test", 500, 500, 403, 403)
374         with req_mock, tutil.skip_sleep, \
375                 self.assertRaises(exc_class) as err_check:
376             keep_client = arvados.KeepClient(api_client=api_client)
377             getattr(keep_client, verb)('d41d8cd98f00b204e9800998ecf8427e+0',
378                                        num_retries=3)
379         self.assertEqual([403, 403], [
380                 getattr(error, 'status_code', None)
381                 for error in err_check.exception.request_errors().itervalues()])
382
383     def test_get_error_reflects_last_retry(self):
384         self.check_errors_from_last_retry('get', arvados.errors.KeepReadError)
385
386     def test_put_error_reflects_last_retry(self):
387         self.check_errors_from_last_retry('put', arvados.errors.KeepWriteError)
388
389     def test_put_error_does_not_include_successful_puts(self):
390         data = 'partial failure test'
391         data_loc = tutil.str_keep_locator(data)
392         api_client = self.mock_keep_services(count=3)
393         with tutil.mock_keep_responses(data_loc, 200, 500, 500) as req_mock, \
394                 self.assertRaises(arvados.errors.KeepWriteError) as exc_check:
395             keep_client = arvados.KeepClient(api_client=api_client)
396             keep_client.put(data)
397         self.assertEqual(2, len(exc_check.exception.request_errors()))
398
399     def test_proxy_put_with_no_writable_services(self):
400         data = 'test with no writable services'
401         data_loc = tutil.str_keep_locator(data)
402         api_client = self.mock_keep_services(service_type='proxy', read_only=True, count=1)
403         with tutil.mock_keep_responses(data_loc, 200, 500, 500) as req_mock, \
404                 self.assertRaises(arvados.errors.KeepWriteError) as exc_check:
405           keep_client = arvados.KeepClient(api_client=api_client)
406           keep_client.put(data)
407         self.assertEqual(True, ("no Keep services available" in str(exc_check.exception)))
408         self.assertEqual(0, len(exc_check.exception.request_errors()))
409
410     def test_oddball_service_get(self):
411         body = 'oddball service get'
412         api_client = self.mock_keep_services(service_type='fancynewblobstore')
413         with tutil.mock_keep_responses(body, 200):
414             keep_client = arvados.KeepClient(api_client=api_client)
415             actual = keep_client.get(tutil.str_keep_locator(body))
416         self.assertEqual(body, actual)
417
418     def test_oddball_service_put(self):
419         body = 'oddball service put'
420         pdh = tutil.str_keep_locator(body)
421         api_client = self.mock_keep_services(service_type='fancynewblobstore')
422         with tutil.mock_keep_responses(pdh, 200):
423             keep_client = arvados.KeepClient(api_client=api_client)
424             actual = keep_client.put(body, copies=1)
425         self.assertEqual(pdh, actual)
426
427     def test_oddball_service_writer_count(self):
428         body = 'oddball service writer count'
429         pdh = tutil.str_keep_locator(body)
430         api_client = self.mock_keep_services(service_type='fancynewblobstore',
431                                              count=4)
432         headers = {'x-keep-replicas-stored': 3}
433         with tutil.mock_keep_responses(pdh, 200, 418, 418, 418,
434                                        **headers) as req_mock:
435             keep_client = arvados.KeepClient(api_client=api_client)
436             actual = keep_client.put(body, copies=2)
437         self.assertEqual(pdh, actual)
438         self.assertEqual(1, req_mock.call_count)
439
440
441 @tutil.skip_sleep
442 class KeepClientRendezvousTestCase(unittest.TestCase, tutil.ApiClientMock):
443
444     def setUp(self):
445         # expected_order[i] is the probe order for
446         # hash=md5(sprintf("%064x",i)) where there are 16 services
447         # with uuid sprintf("anything-%015x",j) with j in 0..15. E.g.,
448         # the first probe for the block consisting of 64 "0"
449         # characters is the service whose uuid is
450         # "zzzzz-bi6l4-000000000000003", so expected_order[0][0]=='3'.
451         self.services = 16
452         self.expected_order = [
453             list('3eab2d5fc9681074'),
454             list('097dba52e648f1c3'),
455             list('c5b4e023f8a7d691'),
456             list('9d81c02e76a3bf54'),
457             ]
458         self.blocks = [
459             "{:064x}".format(x)
460             for x in range(len(self.expected_order))]
461         self.hashes = [
462             hashlib.md5(self.blocks[x]).hexdigest()
463             for x in range(len(self.expected_order))]
464         self.api_client = self.mock_keep_services(count=self.services)
465         self.keep_client = arvados.KeepClient(api_client=self.api_client)
466
467     def test_weighted_service_roots_against_reference_set(self):
468         # Confirm weighted_service_roots() returns the correct order
469         for i, hash in enumerate(self.hashes):
470             roots = self.keep_client.weighted_service_roots(arvados.KeepLocator(hash))
471             got_order = [
472                 re.search(r'//\[?keep0x([0-9a-f]+)', root).group(1)
473                 for root in roots]
474             self.assertEqual(self.expected_order[i], got_order)
475
476     def test_get_probe_order_against_reference_set(self):
477         self._test_probe_order_against_reference_set(
478             lambda i: self.keep_client.get(self.hashes[i], num_retries=1))
479
480     def test_put_probe_order_against_reference_set(self):
481         # copies=1 prevents the test from being sensitive to races
482         # between writer threads.
483         self._test_probe_order_against_reference_set(
484             lambda i: self.keep_client.put(self.blocks[i], num_retries=1, copies=1))
485
486     def _test_probe_order_against_reference_set(self, op):
487         for i in range(len(self.blocks)):
488             with tutil.mock_keep_responses('', *[500 for _ in range(self.services*2)]) as mock, \
489                  self.assertRaises(arvados.errors.KeepRequestError):
490                 op(i)
491             got_order = [
492                 re.search(r'//\[?keep0x([0-9a-f]+)', resp.getopt(pycurl.URL)).group(1)
493                 for resp in mock.responses]
494             self.assertEqual(self.expected_order[i]*2, got_order)
495
496     def test_put_probe_order_multiple_copies(self):
497         for copies in range(2, 4):
498             for i in range(len(self.blocks)):
499                 with tutil.mock_keep_responses('', *[500 for _ in range(self.services*3)]) as mock, \
500                      self.assertRaises(arvados.errors.KeepWriteError):
501                     self.keep_client.put(self.blocks[i], num_retries=2, copies=copies)
502                 got_order = [
503                     re.search(r'//\[?keep0x([0-9a-f]+)', resp.getopt(pycurl.URL)).group(1)
504                     for resp in mock.responses]
505                 # With T threads racing to make requests, the position
506                 # of a given server in the sequence of HTTP requests
507                 # (got_order) cannot be more than T-1 positions
508                 # earlier than that server's position in the reference
509                 # probe sequence (expected_order).
510                 #
511                 # Loop invariant: we have accounted for +pos+ expected
512                 # probes, either by seeing them in +got_order+ or by
513                 # putting them in +pending+ in the hope of seeing them
514                 # later. As long as +len(pending)<T+, we haven't
515                 # started a request too early.
516                 pending = []
517                 for pos, expected in enumerate(self.expected_order[i]*3):
518                     got = got_order[pos-len(pending)]
519                     while got in pending:
520                         del pending[pending.index(got)]
521                         got = got_order[pos-len(pending)]
522                     if got != expected:
523                         pending.append(expected)
524                         self.assertLess(
525                             len(pending), copies,
526                             "pending={}, with copies={}, got {}, expected {}".format(
527                                 pending, copies, repr(got_order), repr(self.expected_order[i]*3)))
528
529     def test_probe_waste_adding_one_server(self):
530         hashes = [
531             hashlib.md5("{:064x}".format(x)).hexdigest() for x in range(100)]
532         initial_services = 12
533         self.api_client = self.mock_keep_services(count=initial_services)
534         self.keep_client = arvados.KeepClient(api_client=self.api_client)
535         probes_before = [
536             self.keep_client.weighted_service_roots(arvados.KeepLocator(hash)) for hash in hashes]
537         for added_services in range(1, 12):
538             api_client = self.mock_keep_services(count=initial_services+added_services)
539             keep_client = arvados.KeepClient(api_client=api_client)
540             total_penalty = 0
541             for hash_index in range(len(hashes)):
542                 probe_after = keep_client.weighted_service_roots(
543                     arvados.KeepLocator(hashes[hash_index]))
544                 penalty = probe_after.index(probes_before[hash_index][0])
545                 self.assertLessEqual(penalty, added_services)
546                 total_penalty += penalty
547             # Average penalty per block should not exceed
548             # N(added)/N(orig) by more than 20%, and should get closer
549             # to the ideal as we add data points.
550             expect_penalty = (
551                 added_services *
552                 len(hashes) / initial_services)
553             max_penalty = (
554                 expect_penalty *
555                 (120 - added_services)/100)
556             min_penalty = (
557                 expect_penalty * 8/10)
558             self.assertTrue(
559                 min_penalty <= total_penalty <= max_penalty,
560                 "With {}+{} services, {} blocks, penalty {} but expected {}..{}".format(
561                     initial_services,
562                     added_services,
563                     len(hashes),
564                     total_penalty,
565                     min_penalty,
566                     max_penalty))
567
568     def check_64_zeros_error_order(self, verb, exc_class):
569         data = '0' * 64
570         if verb == 'get':
571             data = tutil.str_keep_locator(data)
572         # Arbitrary port number:
573         aport = random.randint(1024,65535)
574         api_client = self.mock_keep_services(service_port=aport, count=self.services)
575         keep_client = arvados.KeepClient(api_client=api_client)
576         with mock.patch('pycurl.Curl') as curl_mock, \
577              self.assertRaises(exc_class) as err_check:
578             curl_mock.return_value.side_effect = socket.timeout
579             getattr(keep_client, verb)(data)
580         urls = [urlparse.urlparse(url)
581                 for url in err_check.exception.request_errors()]
582         self.assertEqual([('keep0x' + c, aport) for c in '3eab2d5fc9681074'],
583                          [(url.hostname, url.port) for url in urls])
584
585     def test_get_error_shows_probe_order(self):
586         self.check_64_zeros_error_order('get', arvados.errors.KeepReadError)
587
588     def test_put_error_shows_probe_order(self):
589         self.check_64_zeros_error_order('put', arvados.errors.KeepWriteError)
590
591
592 class KeepClientTimeout(unittest.TestCase, tutil.ApiClientMock):
593     # BANDWIDTH_LOW_LIM must be less than len(DATA) so we can transfer
594     # 1s worth of data and then trigger bandwidth errors before running
595     # out of data.
596     DATA = 'x'*2**11
597     BANDWIDTH_LOW_LIM = 1024
598     TIMEOUT_TIME = 1.0
599
600     class assertTakesBetween(unittest.TestCase):
601         def __init__(self, tmin, tmax):
602             self.tmin = tmin
603             self.tmax = tmax
604
605         def __enter__(self):
606             self.t0 = time.time()
607
608         def __exit__(self, *args, **kwargs):
609             # Round times to milliseconds, like CURL. Otherwise, we
610             # fail when CURL reaches a 1s timeout at 0.9998s.
611             delta = round(time.time() - self.t0, 3)
612             self.assertGreaterEqual(delta, self.tmin)
613             self.assertLessEqual(delta, self.tmax)
614
615     class assertTakesGreater(unittest.TestCase):
616         def __init__(self, tmin):
617             self.tmin = tmin
618
619         def __enter__(self):
620             self.t0 = time.time()
621
622         def __exit__(self, *args, **kwargs):
623             delta = round(time.time() - self.t0, 3)
624             self.assertGreaterEqual(delta, self.tmin)
625
626     def setUp(self):
627         sock = socket.socket()
628         sock.bind(('0.0.0.0', 0))
629         self.port = sock.getsockname()[1]
630         sock.close()
631         self.server = keepstub.Server(('0.0.0.0', self.port), keepstub.Handler)
632         self.thread = threading.Thread(target=self.server.serve_forever)
633         self.thread.daemon = True # Exit thread if main proc exits
634         self.thread.start()
635         self.api_client = self.mock_keep_services(
636             count=1,
637             service_host='localhost',
638             service_port=self.port,
639         )
640
641     def tearDown(self):
642         self.server.shutdown()
643
644     def keepClient(self, timeouts=(0.1, TIMEOUT_TIME, BANDWIDTH_LOW_LIM)):
645         return arvados.KeepClient(
646             api_client=self.api_client,
647             timeout=timeouts)
648
649     def test_timeout_slow_connect(self):
650         # Can't simulate TCP delays with our own socket. Leave our
651         # stub server running uselessly, and try to connect to an
652         # unroutable IP address instead.
653         self.api_client = self.mock_keep_services(
654             count=1,
655             service_host='240.0.0.0',
656         )
657         with self.assertTakesBetween(0.1, 0.5):
658             with self.assertRaises(arvados.errors.KeepWriteError):
659                 self.keepClient().put(self.DATA, copies=1, num_retries=0)
660
661     def test_low_bandwidth_no_delays_success(self):
662         self.server.setbandwidth(2*self.BANDWIDTH_LOW_LIM)
663         kc = self.keepClient()
664         loc = kc.put(self.DATA, copies=1, num_retries=0)
665         self.assertEqual(self.DATA, kc.get(loc, num_retries=0))
666
667     def test_too_low_bandwidth_no_delays_failure(self):
668         # Check that lessening bandwidth corresponds to failing
669         kc = self.keepClient()
670         loc = kc.put(self.DATA, copies=1, num_retries=0)
671         self.server.setbandwidth(0.5*self.BANDWIDTH_LOW_LIM)
672         with self.assertTakesGreater(self.TIMEOUT_TIME):
673             with self.assertRaises(arvados.errors.KeepReadError) as e:
674                 kc.get(loc, num_retries=0)
675         with self.assertTakesGreater(self.TIMEOUT_TIME):
676             with self.assertRaises(arvados.errors.KeepWriteError):
677                 kc.put(self.DATA, copies=1, num_retries=0)
678
679     def test_low_bandwidth_with_server_response_delay_failure(self):
680         kc = self.keepClient()
681         loc = kc.put(self.DATA, copies=1, num_retries=0)
682         self.server.setbandwidth(self.BANDWIDTH_LOW_LIM)
683         self.server.setdelays(response=self.TIMEOUT_TIME)
684         with self.assertTakesGreater(self.TIMEOUT_TIME):
685             with self.assertRaises(arvados.errors.KeepReadError) as e:
686                 kc.get(loc, num_retries=0)
687         with self.assertTakesGreater(self.TIMEOUT_TIME):
688             with self.assertRaises(arvados.errors.KeepWriteError):
689                 kc.put(self.DATA, copies=1, num_retries=0)
690
691     def test_low_bandwidth_with_server_mid_delay_failure(self):
692         kc = self.keepClient()
693         loc = kc.put(self.DATA, copies=1, num_retries=0)
694         self.server.setbandwidth(self.BANDWIDTH_LOW_LIM)
695         self.server.setdelays(mid_write=self.TIMEOUT_TIME, mid_read=self.TIMEOUT_TIME)
696         with self.assertTakesGreater(self.TIMEOUT_TIME):
697             with self.assertRaises(arvados.errors.KeepReadError) as e:
698                 kc.get(loc, num_retries=0)
699         with self.assertTakesGreater(self.TIMEOUT_TIME):
700             with self.assertRaises(arvados.errors.KeepWriteError):
701                 kc.put(self.DATA, copies=1, num_retries=0)
702
703     def test_timeout_slow_request(self):
704         loc = self.keepClient().put(self.DATA, copies=1, num_retries=0)
705         self.server.setdelays(request=.2)
706         self._test_connect_timeout_under_200ms(loc)
707         self.server.setdelays(request=2)
708         self._test_response_timeout_under_2s(loc)
709
710     def test_timeout_slow_response(self):
711         loc = self.keepClient().put(self.DATA, copies=1, num_retries=0)
712         self.server.setdelays(response=.2)
713         self._test_connect_timeout_under_200ms(loc)
714         self.server.setdelays(response=2)
715         self._test_response_timeout_under_2s(loc)
716
717     def test_timeout_slow_response_body(self):
718         loc = self.keepClient().put(self.DATA, copies=1, num_retries=0)
719         self.server.setdelays(response_body=.2)
720         self._test_connect_timeout_under_200ms(loc)
721         self.server.setdelays(response_body=2)
722         self._test_response_timeout_under_2s(loc)
723
724     def _test_connect_timeout_under_200ms(self, loc):
725         # Allow 100ms to connect, then 1s for response. Everything
726         # should work, and everything should take at least 200ms to
727         # return.
728         kc = self.keepClient(timeouts=(.1, 1))
729         with self.assertTakesBetween(.2, .3):
730             kc.put(self.DATA, copies=1, num_retries=0)
731         with self.assertTakesBetween(.2, .3):
732             self.assertEqual(self.DATA, kc.get(loc, num_retries=0))
733
734     def _test_response_timeout_under_2s(self, loc):
735         # Allow 10s to connect, then 1s for response. Nothing should
736         # work, and everything should take at least 1s to return.
737         kc = self.keepClient(timeouts=(10, 1))
738         with self.assertTakesBetween(1, 1.9):
739             with self.assertRaises(arvados.errors.KeepReadError):
740                 kc.get(loc, num_retries=0)
741         with self.assertTakesBetween(1, 1.9):
742             with self.assertRaises(arvados.errors.KeepWriteError):
743                 kc.put(self.DATA, copies=1, num_retries=0)
744
745
746 class KeepClientGatewayTestCase(unittest.TestCase, tutil.ApiClientMock):
747     def mock_disks_and_gateways(self, disks=3, gateways=1):
748         self.gateways = [{
749                 'uuid': 'zzzzz-bi6l4-gateway{:08d}'.format(i),
750                 'owner_uuid': 'zzzzz-tpzed-000000000000000',
751                 'service_host': 'gatewayhost{}'.format(i),
752                 'service_port': 12345,
753                 'service_ssl_flag': True,
754                 'service_type': 'gateway:test',
755         } for i in range(gateways)]
756         self.gateway_roots = [
757             "https://{service_host}:{service_port}/".format(**gw)
758             for gw in self.gateways]
759         self.api_client = self.mock_keep_services(
760             count=disks, additional_services=self.gateways)
761         self.keepClient = arvados.KeepClient(api_client=self.api_client)
762
763     @mock.patch('pycurl.Curl')
764     def test_get_with_gateway_hint_first(self, MockCurl):
765         MockCurl.return_value = tutil.FakeCurl.make(
766             code=200, body='foo', headers={'Content-Length': 3})
767         self.mock_disks_and_gateways()
768         locator = 'acbd18db4cc2f85cedef654fccc4a4d8+3+K@' + self.gateways[0]['uuid']
769         self.assertEqual('foo', self.keepClient.get(locator))
770         self.assertEqual(self.gateway_roots[0]+locator,
771                          MockCurl.return_value.getopt(pycurl.URL))
772
773     @mock.patch('pycurl.Curl')
774     def test_get_with_gateway_hints_in_order(self, MockCurl):
775         gateways = 4
776         disks = 3
777         mocks = [
778             tutil.FakeCurl.make(code=404, body='')
779             for _ in range(gateways+disks)
780         ]
781         MockCurl.side_effect = tutil.queue_with(mocks)
782         self.mock_disks_and_gateways(gateways=gateways, disks=disks)
783         locator = '+'.join(['acbd18db4cc2f85cedef654fccc4a4d8+3'] +
784                            ['K@'+gw['uuid'] for gw in self.gateways])
785         with self.assertRaises(arvados.errors.NotFoundError):
786             self.keepClient.get(locator)
787         # Gateways are tried first, in the order given.
788         for i, root in enumerate(self.gateway_roots):
789             self.assertEqual(root+locator,
790                              mocks[i].getopt(pycurl.URL))
791         # Disk services are tried next.
792         for i in range(gateways, gateways+disks):
793             self.assertRegexpMatches(
794                 mocks[i].getopt(pycurl.URL),
795                 r'keep0x')
796
797     @mock.patch('pycurl.Curl')
798     def test_get_with_remote_proxy_hint(self, MockCurl):
799         MockCurl.return_value = tutil.FakeCurl.make(
800             code=200, body='foo', headers={'Content-Length': 3})
801         self.mock_disks_and_gateways()
802         locator = 'acbd18db4cc2f85cedef654fccc4a4d8+3+K@xyzzy'
803         self.assertEqual('foo', self.keepClient.get(locator))
804         self.assertEqual('https://keep.xyzzy.arvadosapi.com/'+locator,
805                          MockCurl.return_value.getopt(pycurl.URL))
806
807
808 class KeepClientRetryTestMixin(object):
809     # Testing with a local Keep store won't exercise the retry behavior.
810     # Instead, our strategy is:
811     # * Create a client with one proxy specified (pointed at a black
812     #   hole), so there's no need to instantiate an API client, and
813     #   all HTTP requests come from one place.
814     # * Mock httplib's request method to provide simulated responses.
815     # This lets us test the retry logic extensively without relying on any
816     # supporting servers, and prevents side effects in case something hiccups.
817     # To use this mixin, define DEFAULT_EXPECT, DEFAULT_EXCEPTION, and
818     # run_method().
819     #
820     # Test classes must define TEST_PATCHER to a method that mocks
821     # out appropriate methods in the client.
822
823     PROXY_ADDR = 'http://[%s]:65535/' % (tutil.TEST_HOST,)
824     TEST_DATA = 'testdata'
825     TEST_LOCATOR = 'ef654c40ab4f1747fc699915d4f70902+8'
826
827     def setUp(self):
828         self.client_kwargs = {'proxy': self.PROXY_ADDR, 'local_store': ''}
829
830     def new_client(self, **caller_kwargs):
831         kwargs = self.client_kwargs.copy()
832         kwargs.update(caller_kwargs)
833         return arvados.KeepClient(**kwargs)
834
835     def run_method(self, *args, **kwargs):
836         raise NotImplementedError("test subclasses must define run_method")
837
838     def check_success(self, expected=None, *args, **kwargs):
839         if expected is None:
840             expected = self.DEFAULT_EXPECT
841         self.assertEqual(expected, self.run_method(*args, **kwargs))
842
843     def check_exception(self, error_class=None, *args, **kwargs):
844         if error_class is None:
845             error_class = self.DEFAULT_EXCEPTION
846         self.assertRaises(error_class, self.run_method, *args, **kwargs)
847
848     def test_immediate_success(self):
849         with self.TEST_PATCHER(self.DEFAULT_EXPECT, 200):
850             self.check_success()
851
852     def test_retry_then_success(self):
853         with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 200):
854             self.check_success(num_retries=3)
855
856     def test_exception_then_success(self):
857         with self.TEST_PATCHER(self.DEFAULT_EXPECT, Exception('mock err'), 200):
858             self.check_success(num_retries=3)
859
860     def test_no_default_retry(self):
861         with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 200):
862             self.check_exception()
863
864     def test_no_retry_after_permanent_error(self):
865         with self.TEST_PATCHER(self.DEFAULT_EXPECT, 403, 200):
866             self.check_exception(num_retries=3)
867
868     def test_error_after_retries_exhausted(self):
869         with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 500, 200):
870             self.check_exception(num_retries=1)
871
872     def test_num_retries_instance_fallback(self):
873         self.client_kwargs['num_retries'] = 3
874         with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 200):
875             self.check_success()
876
877
878 @tutil.skip_sleep
879 class KeepClientRetryGetTestCase(KeepClientRetryTestMixin, unittest.TestCase):
880     DEFAULT_EXPECT = KeepClientRetryTestMixin.TEST_DATA
881     DEFAULT_EXCEPTION = arvados.errors.KeepReadError
882     HINTED_LOCATOR = KeepClientRetryTestMixin.TEST_LOCATOR + '+K@xyzzy'
883     TEST_PATCHER = staticmethod(tutil.mock_keep_responses)
884
885     def run_method(self, locator=KeepClientRetryTestMixin.TEST_LOCATOR,
886                    *args, **kwargs):
887         return self.new_client().get(locator, *args, **kwargs)
888
889     def test_specific_exception_when_not_found(self):
890         with tutil.mock_keep_responses(self.DEFAULT_EXPECT, 404, 200):
891             self.check_exception(arvados.errors.NotFoundError, num_retries=3)
892
893     def test_general_exception_with_mixed_errors(self):
894         # get should raise a NotFoundError if no server returns the block,
895         # and a high threshold of servers report that it's not found.
896         # This test rigs up 50/50 disagreement between two servers, and
897         # checks that it does not become a NotFoundError.
898         client = self.new_client()
899         with tutil.mock_keep_responses(self.DEFAULT_EXPECT, 404, 500):
900             with self.assertRaises(arvados.errors.KeepReadError) as exc_check:
901                 client.get(self.HINTED_LOCATOR)
902             self.assertNotIsInstance(
903                 exc_check.exception, arvados.errors.NotFoundError,
904                 "mixed errors raised NotFoundError")
905
906     def test_hint_server_can_succeed_without_retries(self):
907         with tutil.mock_keep_responses(self.DEFAULT_EXPECT, 404, 200, 500):
908             self.check_success(locator=self.HINTED_LOCATOR)
909
910     def test_try_next_server_after_timeout(self):
911         with tutil.mock_keep_responses(
912                 (socket.timeout("timed out"), 200),
913                 (self.DEFAULT_EXPECT, 200)):
914             self.check_success(locator=self.HINTED_LOCATOR)
915
916     def test_retry_data_with_wrong_checksum(self):
917         with tutil.mock_keep_responses(
918                 ('baddata', 200),
919                 (self.DEFAULT_EXPECT, 200)):
920             self.check_success(locator=self.HINTED_LOCATOR)
921
922
923 @tutil.skip_sleep
924 class KeepClientRetryPutTestCase(KeepClientRetryTestMixin, unittest.TestCase):
925     DEFAULT_EXPECT = KeepClientRetryTestMixin.TEST_LOCATOR
926     DEFAULT_EXCEPTION = arvados.errors.KeepWriteError
927     TEST_PATCHER = staticmethod(tutil.mock_keep_responses)
928
929     def run_method(self, data=KeepClientRetryTestMixin.TEST_DATA,
930                    copies=1, *args, **kwargs):
931         return self.new_client().put(data, copies, *args, **kwargs)
932
933     def test_do_not_send_multiple_copies_to_same_server(self):
934         with tutil.mock_keep_responses(self.DEFAULT_EXPECT, 200):
935             self.check_exception(copies=2, num_retries=3)
936
937
938 @tutil.skip_sleep
939 class RetryNeedsMultipleServices(unittest.TestCase, tutil.ApiClientMock):
940     # Test put()s that need two distinct servers to succeed, possibly
941     # requiring multiple passes through the retry loop.
942
943     def setUp(self):
944         self.api_client = self.mock_keep_services(count=2)
945         self.keep_client = arvados.KeepClient(api_client=self.api_client)
946
947     def test_success_after_exception(self):
948         with tutil.mock_keep_responses(
949                 'acbd18db4cc2f85cedef654fccc4a4d8+3',
950                 Exception('mock err'), 200, 200) as req_mock:
951             self.keep_client.put('foo', num_retries=1, copies=2)
952         self.assertTrue(3, req_mock.call_count)
953
954     def test_success_after_retryable_error(self):
955         with tutil.mock_keep_responses(
956                 'acbd18db4cc2f85cedef654fccc4a4d8+3',
957                 500, 200, 200) as req_mock:
958             self.keep_client.put('foo', num_retries=1, copies=2)
959         self.assertTrue(3, req_mock.call_count)
960
961     def test_fail_after_final_error(self):
962         # First retry loop gets a 200 (can't achieve replication by
963         # storing again on that server) and a 400 (can't retry that
964         # server at all), so we shouldn't try a third request.
965         with tutil.mock_keep_responses(
966                 'acbd18db4cc2f85cedef654fccc4a4d8+3',
967                 200, 400, 200) as req_mock:
968             with self.assertRaises(arvados.errors.KeepWriteError):
969                 self.keep_client.put('foo', num_retries=1, copies=2)
970         self.assertTrue(2, req_mock.call_count)