X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/66dd4459251971acceae41622e2e1221892a8872..ebb2559b3a09636ff687316bbe512e0e8a86b168:/sdk/python/tests/test_keep_client.py

diff --git a/sdk/python/tests/test_keep_client.py b/sdk/python/tests/test_keep_client.py
index a73ca84211..85b5bc81f0 100644
--- a/sdk/python/tests/test_keep_client.py
+++ b/sdk/python/tests/test_keep_client.py
@@ -29,14 +29,21 @@ class KeepTestCase(run_test_server.TestCaseWithServers):
                                              proxy='', local_store='')
 
     def test_KeepBasicRWTest(self):
+        self.assertEqual(0, self.keep_client.upload_counter.get())
         foo_locator = self.keep_client.put('foo')
         self.assertRegexpMatches(
             foo_locator,
             '^acbd18db4cc2f85cedef654fccc4a4d8\+3',
             'wrong md5 hash from Keep.put("foo"): ' + foo_locator)
+
+        # 6 bytes because uploaded 2 copies
+        self.assertEqual(6, self.keep_client.upload_counter.get())
+
+        self.assertEqual(0, self.keep_client.download_counter.get())
         self.assertEqual(self.keep_client.get(foo_locator),
                          'foo',
                          'wrong content from Keep.get(md5("foo"))')
+        self.assertEqual(3, self.keep_client.download_counter.get())
 
     def test_KeepBinaryRWTest(self):
         blob_str = '\xff\xfe\xf7\x00\x01\x02'
@@ -62,6 +69,7 @@ class KeepTestCase(run_test_server.TestCaseWithServers):
                          blob_str,
                          'wrong content from Keep.get(md5(<binarydata>))')
 
+    @unittest.skip("unreliable test - please fix and close #8752")
     def test_KeepSingleCopyRWTest(self):
         blob_str = '\xff\xfe\xfd\xfc\x00\x01\x02\x03'
         blob_locator = self.keep_client.put(blob_str, copies=1)
@@ -96,6 +104,17 @@ class KeepTestCase(run_test_server.TestCaseWithServers):
             # Must be a string type
             self.keep_client.put({})
 
+    def test_KeepHeadTest(self):
+        locator = self.keep_client.put('test_head')
+        self.assertRegexpMatches(
+            locator,
+            '^b9a772c7049325feb7130fff1f8333e9\+9',
+            'wrong md5 hash from Keep.put for "test_head": ' + locator)
+        self.assertEqual(True, self.keep_client.head(locator))
+        self.assertEqual(self.keep_client.get(locator),
+                         'test_head',
+                         'wrong content from Keep.get for "test_head"')
+
 class KeepPermissionTestCase(run_test_server.TestCaseWithServers):
     MAIN_SERVER = {}
     KEEP_SERVER = {'blob_signing_key': 'abcdefghijk0123456789',
@@ -221,8 +240,8 @@ class KeepProxyTestCase(run_test_server.TestCaseWithServers):
         super(KeepProxyTestCase, self).tearDown()
 
     def test_KeepProxyTest1(self):
-        # Will use ARVADOS_KEEP_PROXY environment variable that is set by
-        # setUpClass().
+        # Will use ARVADOS_KEEP_SERVICES environment variable that
+        # is set by setUpClass().
         keep_client = arvados.KeepClient(api_client=self.api_client,
                                          local_store='')
         baz_locator = keep_client.put('baz')
@@ -251,6 +270,22 @@ class KeepProxyTestCase(run_test_server.TestCaseWithServers):
                          'wrong content from Keep.get(md5("baz2"))')
         self.assertTrue(keep_client.using_proxy)
 
+    def test_KeepProxyTestMultipleURIs(self):
+        # Test using ARVADOS_KEEP_SERVICES env var overriding any
+        # existing proxy setting and setting multiple proxies
+        arvados.config.settings()['ARVADOS_KEEP_SERVICES'] = 'http://10.0.0.1 https://foo.example.org:1234/'
+        keep_client = arvados.KeepClient(api_client=self.api_client,
+                                         local_store='')
+        uris = [x['_service_root'] for x in keep_client._keep_services]
+        self.assertEqual(uris, ['http://10.0.0.1/',
+                                'https://foo.example.org:1234/'])
+
+    def test_KeepProxyTestInvalidURI(self):
+        arvados.config.settings()['ARVADOS_KEEP_SERVICES'] = 'bad.uri.org'
+        with self.assertRaises(arvados.errors.ArgumentError):
+            keep_client = arvados.KeepClient(api_client=self.api_client,
+                                             local_store='')
+
 
 class KeepClientServiceTestCase(unittest.TestCase, tutil.ApiClientMock):
     def get_service_roots(self, api_client):
@@ -287,8 +322,11 @@ class KeepClientServiceTestCase(unittest.TestCase, tutil.ApiClientMock):
                 mock.responses[0].getopt(pycurl.CONNECTTIMEOUT_MS),
                 int(arvados.KeepClient.DEFAULT_TIMEOUT[0]*1000))
             self.assertEqual(
-                mock.responses[0].getopt(pycurl.TIMEOUT_MS),
-                int(arvados.KeepClient.DEFAULT_TIMEOUT[1]*1000))
+                mock.responses[0].getopt(pycurl.LOW_SPEED_TIME),
+                int(arvados.KeepClient.DEFAULT_TIMEOUT[1]))
+            self.assertEqual(
+                mock.responses[0].getopt(pycurl.LOW_SPEED_LIMIT),
+                int(arvados.KeepClient.DEFAULT_TIMEOUT[2]))
 
     def test_put_timeout(self):
         api_client = self.mock_keep_services(count=1)
@@ -301,8 +339,28 @@ class KeepClientServiceTestCase(unittest.TestCase, tutil.ApiClientMock):
                 mock.responses[0].getopt(pycurl.CONNECTTIMEOUT_MS),
                 int(arvados.KeepClient.DEFAULT_TIMEOUT[0]*1000))
             self.assertEqual(
-                mock.responses[0].getopt(pycurl.TIMEOUT_MS),
-                int(arvados.KeepClient.DEFAULT_TIMEOUT[1]*1000))
+                mock.responses[0].getopt(pycurl.LOW_SPEED_TIME),
+                int(arvados.KeepClient.DEFAULT_TIMEOUT[1]))
+            self.assertEqual(
+                mock.responses[0].getopt(pycurl.LOW_SPEED_LIMIT),
+                int(arvados.KeepClient.DEFAULT_TIMEOUT[2]))
+
+    def test_head_timeout(self):
+        api_client = self.mock_keep_services(count=1)
+        force_timeout = socket.timeout("timed out")
+        with tutil.mock_keep_responses(force_timeout, 0) as mock:
+            keep_client = arvados.KeepClient(api_client=api_client)
+            with self.assertRaises(arvados.errors.KeepReadError):
+                keep_client.head('ffffffffffffffffffffffffffffffff')
+            self.assertEqual(
+                mock.responses[0].getopt(pycurl.CONNECTTIMEOUT_MS),
+                int(arvados.KeepClient.DEFAULT_TIMEOUT[0]*1000))
+            self.assertEqual(
+                mock.responses[0].getopt(pycurl.LOW_SPEED_TIME),
+                int(arvados.KeepClient.DEFAULT_TIMEOUT[1]))
+            self.assertEqual(
+                mock.responses[0].getopt(pycurl.LOW_SPEED_LIMIT),
+                int(arvados.KeepClient.DEFAULT_TIMEOUT[2]))
 
     def test_proxy_get_timeout(self):
         api_client = self.mock_keep_services(service_type='proxy', count=1)
@@ -315,8 +373,28 @@ class KeepClientServiceTestCase(unittest.TestCase, tutil.ApiClientMock):
                 mock.responses[0].getopt(pycurl.CONNECTTIMEOUT_MS),
                 int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[0]*1000))
             self.assertEqual(
-                mock.responses[0].getopt(pycurl.TIMEOUT_MS),
-                int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[1]*1000))
+                mock.responses[0].getopt(pycurl.LOW_SPEED_TIME),
+                int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[1]))
+            self.assertEqual(
+                mock.responses[0].getopt(pycurl.LOW_SPEED_LIMIT),
+                int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[2]))
+
+    def test_proxy_head_timeout(self):
+        api_client = self.mock_keep_services(service_type='proxy', count=1)
+        force_timeout = socket.timeout("timed out")
+        with tutil.mock_keep_responses(force_timeout, 0) as mock:
+            keep_client = arvados.KeepClient(api_client=api_client)
+            with self.assertRaises(arvados.errors.KeepReadError):
+                keep_client.head('ffffffffffffffffffffffffffffffff')
+            self.assertEqual(
+                mock.responses[0].getopt(pycurl.CONNECTTIMEOUT_MS),
+                int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[0]*1000))
+            self.assertEqual(
+                mock.responses[0].getopt(pycurl.LOW_SPEED_TIME),
+                int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[1]))
+            self.assertEqual(
+                mock.responses[0].getopt(pycurl.LOW_SPEED_LIMIT),
+                int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[2]))
 
     def test_proxy_put_timeout(self):
         api_client = self.mock_keep_services(service_type='proxy', count=1)
@@ -329,8 +407,11 @@ class KeepClientServiceTestCase(unittest.TestCase, tutil.ApiClientMock):
                 mock.responses[0].getopt(pycurl.CONNECTTIMEOUT_MS),
                 int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[0]*1000))
             self.assertEqual(
-                mock.responses[0].getopt(pycurl.TIMEOUT_MS),
-                int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[1]*1000))
+                mock.responses[0].getopt(pycurl.LOW_SPEED_TIME),
+                int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[1]))
+            self.assertEqual(
+                mock.responses[0].getopt(pycurl.LOW_SPEED_LIMIT),
+                int(arvados.KeepClient.DEFAULT_PROXY_TIMEOUT[2]))
 
     def check_no_services_error(self, verb, exc_class):
         api_client = mock.MagicMock(name='api_client')
@@ -344,6 +425,9 @@ class KeepClientServiceTestCase(unittest.TestCase, tutil.ApiClientMock):
     def test_get_error_with_no_services(self):
         self.check_no_services_error('get', arvados.errors.KeepReadError)
 
+    def test_head_error_with_no_services(self):
+        self.check_no_services_error('head', arvados.errors.KeepReadError)
+
     def test_put_error_with_no_services(self):
         self.check_no_services_error('put', arvados.errors.KeepWriteError)
 
@@ -363,12 +447,15 @@ class KeepClientServiceTestCase(unittest.TestCase, tutil.ApiClientMock):
     def test_get_error_reflects_last_retry(self):
         self.check_errors_from_last_retry('get', arvados.errors.KeepReadError)
 
+    def test_head_error_reflects_last_retry(self):
+        self.check_errors_from_last_retry('head', arvados.errors.KeepReadError)
+
     def test_put_error_reflects_last_retry(self):
         self.check_errors_from_last_retry('put', arvados.errors.KeepWriteError)
 
     def test_put_error_does_not_include_successful_puts(self):
         data = 'partial failure test'
-        data_loc = '{}+{}'.format(hashlib.md5(data).hexdigest(), len(data))
+        data_loc = tutil.str_keep_locator(data)
         api_client = self.mock_keep_services(count=3)
         with tutil.mock_keep_responses(data_loc, 200, 500, 500) as req_mock, \
                 self.assertRaises(arvados.errors.KeepWriteError) as exc_check:
@@ -378,7 +465,7 @@ class KeepClientServiceTestCase(unittest.TestCase, tutil.ApiClientMock):
 
     def test_proxy_put_with_no_writable_services(self):
         data = 'test with no writable services'
-        data_loc = '{}+{}'.format(hashlib.md5(data).hexdigest(), len(data))
+        data_loc = tutil.str_keep_locator(data)
         api_client = self.mock_keep_services(service_type='proxy', read_only=True, count=1)
         with tutil.mock_keep_responses(data_loc, 200, 500, 500) as req_mock, \
                 self.assertRaises(arvados.errors.KeepWriteError) as exc_check:
@@ -387,6 +474,36 @@ class KeepClientServiceTestCase(unittest.TestCase, tutil.ApiClientMock):
         self.assertEqual(True, ("no Keep services available" in str(exc_check.exception)))
         self.assertEqual(0, len(exc_check.exception.request_errors()))
 
+    def test_oddball_service_get(self):
+        body = 'oddball service get'
+        api_client = self.mock_keep_services(service_type='fancynewblobstore')
+        with tutil.mock_keep_responses(body, 200):
+            keep_client = arvados.KeepClient(api_client=api_client)
+            actual = keep_client.get(tutil.str_keep_locator(body))
+        self.assertEqual(body, actual)
+
+    def test_oddball_service_put(self):
+        body = 'oddball service put'
+        pdh = tutil.str_keep_locator(body)
+        api_client = self.mock_keep_services(service_type='fancynewblobstore')
+        with tutil.mock_keep_responses(pdh, 200):
+            keep_client = arvados.KeepClient(api_client=api_client)
+            actual = keep_client.put(body, copies=1)
+        self.assertEqual(pdh, actual)
+
+    def test_oddball_service_writer_count(self):
+        body = 'oddball service writer count'
+        pdh = tutil.str_keep_locator(body)
+        api_client = self.mock_keep_services(service_type='fancynewblobstore',
+                                             count=4)
+        headers = {'x-keep-replicas-stored': 3}
+        with tutil.mock_keep_responses(pdh, 200, 418, 418, 418,
+                                       **headers) as req_mock:
+            keep_client = arvados.KeepClient(api_client=api_client)
+            actual = keep_client.put(body, copies=2)
+        self.assertEqual(pdh, actual)
+        self.assertEqual(1, req_mock.call_count)
+
 
 @tutil.skip_sleep
 class KeepClientRendezvousTestCase(unittest.TestCase, tutil.ApiClientMock):
@@ -427,6 +544,10 @@ class KeepClientRendezvousTestCase(unittest.TestCase, tutil.ApiClientMock):
         self._test_probe_order_against_reference_set(
             lambda i: self.keep_client.get(self.hashes[i], num_retries=1))
 
+    def test_head_probe_order_against_reference_set(self):
+        self._test_probe_order_against_reference_set(
+            lambda i: self.keep_client.head(self.hashes[i], num_retries=1))
+
     def test_put_probe_order_against_reference_set(self):
         # copies=1 prevents the test from being sensitive to races
         # between writer threads.
@@ -452,21 +573,29 @@ class KeepClientRendezvousTestCase(unittest.TestCase, tutil.ApiClientMock):
                 got_order = [
                     re.search(r'//\[?keep0x([0-9a-f]+)', resp.getopt(pycurl.URL)).group(1)
                     for resp in mock.responses]
+                # With T threads racing to make requests, the position
+                # of a given server in the sequence of HTTP requests
+                # (got_order) cannot be more than T-1 positions
+                # earlier than that server's position in the reference
+                # probe sequence (expected_order).
+                #
+                # Loop invariant: we have accounted for +pos+ expected
+                # probes, either by seeing them in +got_order+ or by
+                # putting them in +pending+ in the hope of seeing them
+                # later. As long as +len(pending)<T+, we haven't
+                # started a request too early.
+                pending = []
                 for pos, expected in enumerate(self.expected_order[i]*3):
-                    # With C threads racing to make requests, the
-                    # position of a given server in the sequence of
-                    # HTTP requests (got_order) should be within C-1
-                    # positions of that server's position in the
-                    # reference probe sequence (expected_order).
-                    close_enough = False
-                    for diff in range(1-copies, copies):
-                        if 0 <= pos+diff < len(got_order):
-                            if expected == got_order[pos+diff]:
-                                close_enough = True
-                    self.assertEqual(
-                        True, close_enough,
-                        "With copies={}, got {}, expected {}".format(
-                            copies, repr(got_order), repr(self.expected_order[i]*3)))
+                    got = got_order[pos-len(pending)]
+                    while got in pending:
+                        del pending[pending.index(got)]
+                        got = got_order[pos-len(pending)]
+                    if got != expected:
+                        pending.append(expected)
+                        self.assertLess(
+                            len(pending), copies,
+                            "pending={}, with copies={}, got {}, expected {}".format(
+                                pending, copies, repr(got_order), repr(self.expected_order[i]*3)))
 
     def test_probe_waste_adding_one_server(self):
         hashes = [
@@ -510,7 +639,7 @@ class KeepClientRendezvousTestCase(unittest.TestCase, tutil.ApiClientMock):
     def check_64_zeros_error_order(self, verb, exc_class):
         data = '0' * 64
         if verb == 'get':
-            data = hashlib.md5(data).hexdigest() + '+1234'
+            data = tutil.str_keep_locator(data)
         # Arbitrary port number:
         aport = random.randint(1024,65535)
         api_client = self.mock_keep_services(service_port=aport, count=self.services)
@@ -532,7 +661,12 @@ class KeepClientRendezvousTestCase(unittest.TestCase, tutil.ApiClientMock):
 
 
 class KeepClientTimeout(unittest.TestCase, tutil.ApiClientMock):
-    DATA = 'x' * 2**10
+    # BANDWIDTH_LOW_LIM must be less than len(DATA) so we can transfer
+    # 1s worth of data and then trigger bandwidth errors before running
+    # out of data.
+    DATA = 'x'*2**11
+    BANDWIDTH_LOW_LIM = 1024
+    TIMEOUT_TIME = 1.0
 
     class assertTakesBetween(unittest.TestCase):
         def __init__(self, tmin, tmax):
@@ -543,8 +677,22 @@ class KeepClientTimeout(unittest.TestCase, tutil.ApiClientMock):
             self.t0 = time.time()
 
         def __exit__(self, *args, **kwargs):
-            self.assertGreater(time.time() - self.t0, self.tmin)
-            self.assertLess(time.time() - self.t0, self.tmax)
+            # Round times to milliseconds, like CURL. Otherwise, we
+            # fail when CURL reaches a 1s timeout at 0.9998s.
+            delta = round(time.time() - self.t0, 3)
+            self.assertGreaterEqual(delta, self.tmin)
+            self.assertLessEqual(delta, self.tmax)
+
+    class assertTakesGreater(unittest.TestCase):
+        def __init__(self, tmin):
+            self.tmin = tmin
+
+        def __enter__(self):
+            self.t0 = time.time()
+
+        def __exit__(self, *args, **kwargs):
+            delta = round(time.time() - self.t0, 3)
+            self.assertGreaterEqual(delta, self.tmin)
 
     def setUp(self):
         sock = socket.socket()
@@ -564,7 +712,7 @@ class KeepClientTimeout(unittest.TestCase, tutil.ApiClientMock):
     def tearDown(self):
         self.server.shutdown()
 
-    def keepClient(self, timeouts=(0.1, 1.0)):
+    def keepClient(self, timeouts=(0.1, TIMEOUT_TIME, BANDWIDTH_LOW_LIM)):
         return arvados.KeepClient(
             api_client=self.api_client,
             timeout=timeouts)
@@ -579,39 +727,92 @@ class KeepClientTimeout(unittest.TestCase, tutil.ApiClientMock):
         )
         with self.assertTakesBetween(0.1, 0.5):
             with self.assertRaises(arvados.errors.KeepWriteError):
-                self.keepClient((0.1, 1)).put(self.DATA, copies=1, num_retries=0)
+                self.keepClient().put(self.DATA, copies=1, num_retries=0)
+
+    def test_low_bandwidth_no_delays_success(self):
+        self.server.setbandwidth(2*self.BANDWIDTH_LOW_LIM)
+        kc = self.keepClient()
+        loc = kc.put(self.DATA, copies=1, num_retries=0)
+        self.assertEqual(self.DATA, kc.get(loc, num_retries=0))
+
+    def test_too_low_bandwidth_no_delays_failure(self):
+        # Check that lessening bandwidth corresponds to failing
+        kc = self.keepClient()
+        loc = kc.put(self.DATA, copies=1, num_retries=0)
+        self.server.setbandwidth(0.5*self.BANDWIDTH_LOW_LIM)
+        with self.assertTakesGreater(self.TIMEOUT_TIME):
+            with self.assertRaises(arvados.errors.KeepReadError) as e:
+                kc.get(loc, num_retries=0)
+        with self.assertTakesGreater(self.TIMEOUT_TIME):
+            with self.assertRaises(arvados.errors.KeepWriteError):
+                kc.put(self.DATA, copies=1, num_retries=0)
+
+    def test_low_bandwidth_with_server_response_delay_failure(self):
+        kc = self.keepClient()
+        loc = kc.put(self.DATA, copies=1, num_retries=0)
+        self.server.setbandwidth(self.BANDWIDTH_LOW_LIM)
+        self.server.setdelays(response=self.TIMEOUT_TIME)
+        with self.assertTakesGreater(self.TIMEOUT_TIME):
+            with self.assertRaises(arvados.errors.KeepReadError) as e:
+                kc.get(loc, num_retries=0)
+        with self.assertTakesGreater(self.TIMEOUT_TIME):
+            with self.assertRaises(arvados.errors.KeepWriteError):
+                kc.put(self.DATA, copies=1, num_retries=0)
+        with self.assertTakesGreater(self.TIMEOUT_TIME):
+            with self.assertRaises(arvados.errors.KeepReadError) as e:
+                kc.head(loc, num_retries=0)
+
+    def test_low_bandwidth_with_server_mid_delay_failure(self):
+        kc = self.keepClient()
+        loc = kc.put(self.DATA, copies=1, num_retries=0)
+        self.server.setbandwidth(self.BANDWIDTH_LOW_LIM)
+        self.server.setdelays(mid_write=self.TIMEOUT_TIME, mid_read=self.TIMEOUT_TIME)
+        with self.assertTakesGreater(self.TIMEOUT_TIME):
+            with self.assertRaises(arvados.errors.KeepReadError) as e:
+                kc.get(loc, num_retries=0)
+        with self.assertTakesGreater(self.TIMEOUT_TIME):
+            with self.assertRaises(arvados.errors.KeepWriteError):
+                kc.put(self.DATA, copies=1, num_retries=0)
 
     def test_timeout_slow_request(self):
-        self.server.setdelays(request=0.2)
-        self._test_200ms()
+        loc = self.keepClient().put(self.DATA, copies=1, num_retries=0)
+        self.server.setdelays(request=.2)
+        self._test_connect_timeout_under_200ms(loc)
+        self.server.setdelays(request=2)
+        self._test_response_timeout_under_2s(loc)
 
     def test_timeout_slow_response(self):
-        self.server.setdelays(response=0.2)
-        self._test_200ms()
+        loc = self.keepClient().put(self.DATA, copies=1, num_retries=0)
+        self.server.setdelays(response=.2)
+        self._test_connect_timeout_under_200ms(loc)
+        self.server.setdelays(response=2)
+        self._test_response_timeout_under_2s(loc)
 
     def test_timeout_slow_response_body(self):
-        self.server.setdelays(response_body=0.2)
-        self._test_200ms()
-
-    def _test_200ms(self):
-        """Connect should be t<100ms, request should be 200ms <= t < 300ms"""
+        loc = self.keepClient().put(self.DATA, copies=1, num_retries=0)
+        self.server.setdelays(response_body=.2)
+        self._test_connect_timeout_under_200ms(loc)
+        self.server.setdelays(response_body=2)
+        self._test_response_timeout_under_2s(loc)
 
+    def _test_connect_timeout_under_200ms(self, loc):
         # Allow 100ms to connect, then 1s for response. Everything
         # should work, and everything should take at least 200ms to
         # return.
-        kc = self.keepClient((.1, 1))
+        kc = self.keepClient(timeouts=(.1, 1))
         with self.assertTakesBetween(.2, .3):
-            loc = kc.put(self.DATA, copies=1, num_retries=0)
+            kc.put(self.DATA, copies=1, num_retries=0)
         with self.assertTakesBetween(.2, .3):
             self.assertEqual(self.DATA, kc.get(loc, num_retries=0))
 
-        # Allow 1s to connect, then 100ms for response. Nothing should
-        # work, and everything should take at least 100ms to return.
-        kc = self.keepClient((1, .1))
-        with self.assertTakesBetween(.1, .2):
+    def _test_response_timeout_under_2s(self, loc):
+        # Allow 10s to connect, then 1s for response. Nothing should
+        # work, and everything should take at least 1s to return.
+        kc = self.keepClient(timeouts=(10, 1))
+        with self.assertTakesBetween(1, 9):
             with self.assertRaises(arvados.errors.KeepReadError):
                 kc.get(loc, num_retries=0)
-        with self.assertTakesBetween(.1, .2):
+        with self.assertTakesBetween(1, 9):
             with self.assertRaises(arvados.errors.KeepWriteError):
                 kc.put(self.DATA, copies=1, num_retries=0)
 
@@ -642,6 +843,7 @@ class KeepClientGatewayTestCase(unittest.TestCase, tutil.ApiClientMock):
         self.assertEqual('foo', self.keepClient.get(locator))
         self.assertEqual(self.gateway_roots[0]+locator,
                          MockCurl.return_value.getopt(pycurl.URL))
+        self.assertEqual(True, self.keepClient.head(locator))
 
     @mock.patch('pycurl.Curl')
     def test_get_with_gateway_hints_in_order(self, MockCurl):
@@ -667,6 +869,30 @@ class KeepClientGatewayTestCase(unittest.TestCase, tutil.ApiClientMock):
                 mocks[i].getopt(pycurl.URL),
                 r'keep0x')
 
+    @mock.patch('pycurl.Curl')
+    def test_head_with_gateway_hints_in_order(self, MockCurl):
+        gateways = 4
+        disks = 3
+        mocks = [
+            tutil.FakeCurl.make(code=404, body='')
+            for _ in range(gateways+disks)
+        ]
+        MockCurl.side_effect = tutil.queue_with(mocks)
+        self.mock_disks_and_gateways(gateways=gateways, disks=disks)
+        locator = '+'.join(['acbd18db4cc2f85cedef654fccc4a4d8+3'] +
+                           ['K@'+gw['uuid'] for gw in self.gateways])
+        with self.assertRaises(arvados.errors.NotFoundError):
+            self.keepClient.head(locator)
+        # Gateways are tried first, in the order given.
+        for i, root in enumerate(self.gateway_roots):
+            self.assertEqual(root+locator,
+                             mocks[i].getopt(pycurl.URL))
+        # Disk services are tried next.
+        for i in range(gateways, gateways+disks):
+            self.assertRegexpMatches(
+                mocks[i].getopt(pycurl.URL),
+                r'keep0x')
+
     @mock.patch('pycurl.Curl')
     def test_get_with_remote_proxy_hint(self, MockCurl):
         MockCurl.return_value = tutil.FakeCurl.make(
@@ -677,6 +903,16 @@ class KeepClientGatewayTestCase(unittest.TestCase, tutil.ApiClientMock):
         self.assertEqual('https://keep.xyzzy.arvadosapi.com/'+locator,
                          MockCurl.return_value.getopt(pycurl.URL))
 
+    @mock.patch('pycurl.Curl')
+    def test_head_with_remote_proxy_hint(self, MockCurl):
+        MockCurl.return_value = tutil.FakeCurl.make(
+            code=200, body='foo', headers={'Content-Length': 3})
+        self.mock_disks_and_gateways()
+        locator = 'acbd18db4cc2f85cedef654fccc4a4d8+3+K@xyzzy'
+        self.assertEqual(True, self.keepClient.head(locator))
+        self.assertEqual('https://keep.xyzzy.arvadosapi.com/'+locator,
+                         MockCurl.return_value.getopt(pycurl.URL))
+
 
 class KeepClientRetryTestMixin(object):
     # Testing with a local Keep store won't exercise the retry behavior.
@@ -726,6 +962,10 @@ class KeepClientRetryTestMixin(object):
         with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 200):
             self.check_success(num_retries=3)
 
+    def test_exception_then_success(self):
+        with self.TEST_PATCHER(self.DEFAULT_EXPECT, Exception('mock err'), 200):
+            self.check_success(num_retries=3)
+
     def test_no_default_retry(self):
         with self.TEST_PATCHER(self.DEFAULT_EXPECT, 500, 200):
             self.check_exception()
@@ -788,6 +1028,43 @@ class KeepClientRetryGetTestCase(KeepClientRetryTestMixin, unittest.TestCase):
                 (self.DEFAULT_EXPECT, 200)):
             self.check_success(locator=self.HINTED_LOCATOR)
 
+@tutil.skip_sleep
+class KeepClientRetryHeadTestCase(KeepClientRetryTestMixin, unittest.TestCase):
+    DEFAULT_EXPECT = True
+    DEFAULT_EXCEPTION = arvados.errors.KeepReadError
+    HINTED_LOCATOR = KeepClientRetryTestMixin.TEST_LOCATOR + '+K@xyzzy'
+    TEST_PATCHER = staticmethod(tutil.mock_keep_responses)
+
+    def run_method(self, locator=KeepClientRetryTestMixin.TEST_LOCATOR,
+                   *args, **kwargs):
+        return self.new_client().head(locator, *args, **kwargs)
+
+    def test_specific_exception_when_not_found(self):
+        with tutil.mock_keep_responses(self.DEFAULT_EXPECT, 404, 200):
+            self.check_exception(arvados.errors.NotFoundError, num_retries=3)
+
+    def test_general_exception_with_mixed_errors(self):
+        # head should raise a NotFoundError if no server returns the block,
+        # and a high threshold of servers report that it's not found.
+        # This test rigs up 50/50 disagreement between two servers, and
+        # checks that it does not become a NotFoundError.
+        client = self.new_client()
+        with tutil.mock_keep_responses(self.DEFAULT_EXPECT, 404, 500):
+            with self.assertRaises(arvados.errors.KeepReadError) as exc_check:
+                client.head(self.HINTED_LOCATOR)
+            self.assertNotIsInstance(
+                exc_check.exception, arvados.errors.NotFoundError,
+                "mixed errors raised NotFoundError")
+
+    def test_hint_server_can_succeed_without_retries(self):
+        with tutil.mock_keep_responses(self.DEFAULT_EXPECT, 404, 200, 500):
+            self.check_success(locator=self.HINTED_LOCATOR)
+
+    def test_try_next_server_after_timeout(self):
+        with tutil.mock_keep_responses(
+                (socket.timeout("timed out"), 200),
+                (self.DEFAULT_EXPECT, 200)):
+            self.check_success(locator=self.HINTED_LOCATOR)
 
 @tutil.skip_sleep
 class KeepClientRetryPutTestCase(KeepClientRetryTestMixin, unittest.TestCase):
@@ -802,3 +1079,108 @@ class KeepClientRetryPutTestCase(KeepClientRetryTestMixin, unittest.TestCase):
     def test_do_not_send_multiple_copies_to_same_server(self):
         with tutil.mock_keep_responses(self.DEFAULT_EXPECT, 200):
             self.check_exception(copies=2, num_retries=3)
+
+
+class AvoidOverreplication(unittest.TestCase, tutil.ApiClientMock):
+
+    class FakeKeepService(object):
+        def __init__(self, delay, will_succeed=False, will_raise=None, replicas=1):
+            self.delay = delay
+            self.will_succeed = will_succeed
+            self.will_raise = will_raise
+            self._result = {}
+            self._result['headers'] = {}
+            self._result['headers']['x-keep-replicas-stored'] = str(replicas)
+            self._result['body'] = 'foobar'
+
+        def put(self, data_hash, data, timeout):
+            time.sleep(self.delay)
+            if self.will_raise is not None:
+                raise self.will_raise
+            return self.will_succeed
+
+        def last_result(self):
+            if self.will_succeed:
+                return self._result
+
+        def finished(self):
+            return False
+    
+    def setUp(self):
+        self.copies = 3
+        self.pool = arvados.KeepClient.KeepWriterThreadPool(
+            data = 'foo',
+            data_hash = 'acbd18db4cc2f85cedef654fccc4a4d8+3',
+            max_service_replicas = self.copies,
+            copies = self.copies
+        )
+
+    def test_only_write_enough_on_success(self):
+        for i in range(10):
+            ks = self.FakeKeepService(delay=i/10.0, will_succeed=True)
+            self.pool.add_task(ks, None)
+        self.pool.join()
+        self.assertEqual(self.pool.done(), self.copies)
+
+    def test_only_write_enough_on_partial_success(self):
+        for i in range(5):
+            ks = self.FakeKeepService(delay=i/10.0, will_succeed=False)
+            self.pool.add_task(ks, None)
+            ks = self.FakeKeepService(delay=i/10.0, will_succeed=True)
+            self.pool.add_task(ks, None)
+        self.pool.join()
+        self.assertEqual(self.pool.done(), self.copies)
+
+    def test_only_write_enough_when_some_crash(self):
+        for i in range(5):
+            ks = self.FakeKeepService(delay=i/10.0, will_raise=Exception())
+            self.pool.add_task(ks, None)
+            ks = self.FakeKeepService(delay=i/10.0, will_succeed=True)
+            self.pool.add_task(ks, None)
+        self.pool.join()
+        self.assertEqual(self.pool.done(), self.copies)
+
+    def test_fail_when_too_many_crash(self):
+        for i in range(self.copies+1):
+            ks = self.FakeKeepService(delay=i/10.0, will_raise=Exception())
+            self.pool.add_task(ks, None)
+        for i in range(self.copies-1):
+            ks = self.FakeKeepService(delay=i/10.0, will_succeed=True)
+            self.pool.add_task(ks, None)
+        self.pool.join()
+        self.assertEqual(self.pool.done(), self.copies-1)
+    
+
+@tutil.skip_sleep
+class RetryNeedsMultipleServices(unittest.TestCase, tutil.ApiClientMock):
+    # Test put()s that need two distinct servers to succeed, possibly
+    # requiring multiple passes through the retry loop.
+
+    def setUp(self):
+        self.api_client = self.mock_keep_services(count=2)
+        self.keep_client = arvados.KeepClient(api_client=self.api_client)
+
+    def test_success_after_exception(self):
+        with tutil.mock_keep_responses(
+                'acbd18db4cc2f85cedef654fccc4a4d8+3',
+                Exception('mock err'), 200, 200) as req_mock:
+            self.keep_client.put('foo', num_retries=1, copies=2)
+        self.assertEqual(3, req_mock.call_count)
+
+    def test_success_after_retryable_error(self):
+        with tutil.mock_keep_responses(
+                'acbd18db4cc2f85cedef654fccc4a4d8+3',
+                500, 200, 200) as req_mock:
+            self.keep_client.put('foo', num_retries=1, copies=2)
+        self.assertEqual(3, req_mock.call_count)
+
+    def test_fail_after_final_error(self):
+        # First retry loop gets a 200 (can't achieve replication by
+        # storing again on that server) and a 400 (can't retry that
+        # server at all), so we shouldn't try a third request.
+        with tutil.mock_keep_responses(
+                'acbd18db4cc2f85cedef654fccc4a4d8+3',
+                200, 400, 200) as req_mock:
+            with self.assertRaises(arvados.errors.KeepWriteError):
+                self.keep_client.put('foo', num_retries=1, copies=2)
+        self.assertEqual(2, req_mock.call_count)