test_system.py: clean up control flow, reduce use of stall()
authorBrian Warner <warner@lothar.com>
Tue, 12 Jun 2012 01:22:35 +0000 (18:22 -0700)
committerBrian Warner <warner@lothar.com>
Tue, 12 Jun 2012 01:22:35 +0000 (18:22 -0700)
The _upload_resumable() test interrupts a Helper upload partway
through (by shutting down the Helper), then restarts the Helper and
resumes the upload. The control flow is kind of tricky: to do anything
"partway through" requires adding a hook to the Uploadable. The previous
flow depended upon a (fragile) call to self.stall(), which waits a fixed
number of seconds.

This removes one of those stall() calls (the remainder is in
test/common.py and I'll try removing it in a subsequent revision). It
also removes some now-redundant wait_for_connections() calls, since
bounce_client() doesn't fire its Deferred until the client has finished
coming back up (and uses wait_for_connections() internally to do so).

src/allmydata/test/test_system.py

index bf9683272ad9502c99283880744198a53d8fe20d..9eb8811a2a20aff2c9899773e9f7a3c8a103be0e 100644 (file)
@@ -24,7 +24,7 @@ from allmydata.monitor import Monitor
 from allmydata.mutable.common import NotWriteableError
 from allmydata.mutable import layout as mutable_layout
 from allmydata.mutable.publish import MutableData
-from foolscap.api import DeadReferenceError
+from foolscap.api import DeadReferenceError, fireEventually
 from twisted.python.failure import Failure
 from twisted.web.client import getPage
 from twisted.web.error import Error
@@ -244,17 +244,22 @@ class SystemTest(SystemTestMixin, RunBinTahoeMixin, unittest.TestCase):
         if convergence is not None:
             d.addCallback(_upload_duplicate_with_helper)
 
+        d.addCallback(fireEventually)
+
         def _upload_resumable(res):
             DATA = "Data that needs help to upload and gets interrupted" * 1000
             u1 = CountingDataUploadable(DATA, convergence=convergence)
             u2 = CountingDataUploadable(DATA, convergence=convergence)
 
             # we interrupt the connection after about 5kB by shutting down
-            # the helper, then restartingit.
+            # the helper, then restarting it.
             u1.interrupt_after = 5000
             u1.interrupt_after_d = defer.Deferred()
-            u1.interrupt_after_d.addCallback(lambda res:
-                                             self.bounce_client(0))
+            bounced_d = defer.Deferred()
+            def _do_bounce(res):
+                d = self.bounce_client(0)
+                d.addBoth(bounced_d.callback)
+            u1.interrupt_after_d.addCallback(_do_bounce)
 
             # sneak into the helper and reduce its chunk size, so that our
             # debug_interrupt will sever the connection on about the fifth
@@ -265,34 +270,27 @@ class SystemTest(SystemTestMixin, RunBinTahoeMixin, unittest.TestCase):
             # this same test run, but I'm not currently worried about it.
             offloaded.CHKCiphertextFetcher.CHUNK_SIZE = 1000
 
-            d = self.extra_node.upload(u1)
-
-            def _should_not_finish(res):
-                self.fail("interrupted upload should have failed, not finished"
-                          " with result %s" % (res,))
-            def _interrupted(f):
-                f.trap(DeadReferenceError)
-
-                # make sure we actually interrupted it before finishing the
-                # file
-                self.failUnless(u1.bytes_read < len(DATA),
-                                "read %d out of %d total" % (u1.bytes_read,
-                                                             len(DATA)))
-
-                log.msg("waiting for reconnect", level=log.NOISY,
-                        facility="tahoe.test.test_system")
-                # now, we need to give the nodes a chance to notice that this
-                # connection has gone away. When this happens, the storage
-                # servers will be told to abort their uploads, removing the
-                # partial shares. Unfortunately this involves TCP messages
-                # going through the loopback interface, and we can't easily
-                # predict how long that will take. If it were all local, we
-                # could use fireEventually() to stall. Since we don't have
-                # the right introduction hooks, the best we can do is use a
-                # fixed delay. TODO: this is fragile.
-                u1.interrupt_after_d.addCallback(self.stall, 2.0)
-                return u1.interrupt_after_d
-            d.addCallbacks(_should_not_finish, _interrupted)
+            upload_d = self.extra_node.upload(u1)
+            # The upload will start, and bounce_client() will be called after
+            # about 5kB. bounced_d will fire after bounce_client() finishes
+            # shutting down and restarting the node.
+            d = bounced_d
+            def _bounced(ign):
+                # By this point, the upload should have failed because of the
+                # interruption. upload_d will fire in a moment
+                def _should_not_finish(res):
+                    self.fail("interrupted upload should have failed, not"
+                              " finished with result %s" % (res,))
+                def _interrupted(f):
+                    f.trap(DeadReferenceError)
+                    # make sure we actually interrupted it before finishing
+                    # the file
+                    self.failUnless(u1.bytes_read < len(DATA),
+                                    "read %d out of %d total" %
+                                    (u1.bytes_read, len(DATA)))
+                upload_d.addCallbacks(_should_not_finish, _interrupted)
+                return upload_d
+            d.addCallback(_bounced)
 
             def _disconnected(res):
                 # check to make sure the storage servers aren't still hanging
@@ -311,8 +309,6 @@ class SystemTest(SystemTestMixin, RunBinTahoeMixin, unittest.TestCase):
             d.addCallback(lambda res:
                           log.msg("wait_for_connections", level=log.NOISY,
                                   facility="tahoe.test.test_system"))
-            d.addCallback(lambda res: self.wait_for_connections())
-
 
             d.addCallback(lambda res:
                           log.msg("uploading again", level=log.NOISY,