3198: Implement rename() (efficient move within/between collections).
[arvados.git] / services / fuse / tests / test_mount.py
index 764a099149a452ffe4258f2f8a0b922fd381b83a..70cfbef4174a7b0277437c09676998164ba0d019 100644 (file)
@@ -12,9 +12,28 @@ import tempfile
 import threading
 import time
 import unittest
-
+import logging
+import multiprocessing
 import run_test_server
 
+logger = logging.getLogger('arvados.arv-mount')
+
+import threading, sys, traceback
+
+def dumpstacks(signal, frame):
+    id2name = dict([(th.ident, th.name) for th in threading.enumerate()])
+    code = []
+    for threadId, stack in sys._current_frames().items():
+        code.append("\n# Thread: %s(%d)" % (id2name.get(threadId,""), threadId))
+        for filename, lineno, name, line in traceback.extract_stack(stack):
+            code.append('File: "%s", line %d, in %s' % (filename, lineno, name))
+            if line:
+                code.append("  %s" % (line.strip()))
+    print "\n".join(code)
+
+import signal
+signal.signal(signal.SIGUSR1, dumpstacks)
+
 class MountTestBase(unittest.TestCase):
     def setUp(self):
         self.keeptmp = tempfile.mkdtemp()
@@ -25,13 +44,14 @@ class MountTestBase(unittest.TestCase):
         self.api = arvados.safeapi.ThreadSafeApiCache(arvados.config.settings())
 
     def make_mount(self, root_class, **root_kwargs):
-        operations = fuse.Operations(os.getuid(), os.getgid())
+        operations = fuse.Operations(os.getuid(), os.getgid(), inode_cache=2)
         operations.inodes.add_entry(root_class(
             llfuse.ROOT_INODE, operations.inodes, self.api, 0, **root_kwargs))
-        llfuse.init(operations, self.mounttmp, [])
+        llfuse.init(operations, self.mounttmp, ['debug'])
         threading.Thread(None, llfuse.main).start()
         # wait until the driver is finished initializing
         operations.initlock.wait()
+        return operations.inodes[llfuse.ROOT_INODE]
 
     def tearDown(self):
         # llfuse.close is buggy, so use fusermount instead.
@@ -51,7 +71,7 @@ class MountTestBase(unittest.TestCase):
         path = self.mounttmp
         if subdir:
             path = os.path.join(path, subdir)
-        self.assertEqual(sorted(expect_content), sorted(os.listdir(path)))
+        self.assertEqual(sorted(expect_content), sorted(llfuse.listdir(path)))
 
 
 class FuseMountTest(MountTestBase):
@@ -98,7 +118,7 @@ class FuseMountTest(MountTestBase):
         self.api.collections().create(body={"manifest_text":cw.manifest_text()}).execute()
 
     def runTest(self):
-        self.make_mount(fuse.CollectionDirectory, collection=self.testcollection)
+        self.make_mount(fuse.CollectionDirectory, collection_record=self.testcollection)
 
         self.assertDirContents(None, ['thing1.txt', 'thing2.txt',
                                       'edgecases', 'dir1', 'dir2'])
@@ -157,7 +177,7 @@ class FuseMagicTest(MountTestBase):
     def runTest(self):
         self.make_mount(fuse.MagicDirectory)
 
-        mount_ls = os.listdir(self.mounttmp)
+        mount_ls = llfuse.listdir(self.mounttmp)
         self.assertIn('README', mount_ls)
         self.assertFalse(any(arvados.util.keep_locator_pattern.match(fn) or
                              arvados.util.uuid_pattern.match(fn)
@@ -166,11 +186,11 @@ class FuseMagicTest(MountTestBase):
         self.assertDirContents(self.testcollection, ['thing1.txt'])
         self.assertDirContents(os.path.join('by_id', self.testcollection),
                                ['thing1.txt'])
-        mount_ls = os.listdir(self.mounttmp)
+        mount_ls = llfuse.listdir(self.mounttmp)
         self.assertIn('README', mount_ls)
         self.assertIn(self.testcollection, mount_ls)
         self.assertIn(self.testcollection,
-                      os.listdir(os.path.join(self.mounttmp, 'by_id')))
+                      llfuse.listdir(os.path.join(self.mounttmp, 'by_id')))
 
         files = {}
         files[os.path.join(self.mounttmp, self.testcollection, 'thing1.txt')] = 'data 1'
@@ -184,15 +204,15 @@ class FuseTagsTest(MountTestBase):
     def runTest(self):
         self.make_mount(fuse.TagsDirectory)
 
-        d1 = os.listdir(self.mounttmp)
+        d1 = llfuse.listdir(self.mounttmp)
         d1.sort()
         self.assertEqual(['foo_tag'], d1)
 
-        d2 = os.listdir(os.path.join(self.mounttmp, 'foo_tag'))
+        d2 = llfuse.listdir(os.path.join(self.mounttmp, 'foo_tag'))
         d2.sort()
         self.assertEqual(['zzzzz-4zz18-fy296fx3hot09f7'], d2)
 
-        d3 = os.listdir(os.path.join(self.mounttmp, 'foo_tag', 'zzzzz-4zz18-fy296fx3hot09f7'))
+        d3 = llfuse.listdir(os.path.join(self.mounttmp, 'foo_tag', 'zzzzz-4zz18-fy296fx3hot09f7'))
         d3.sort()
         self.assertEqual(['foo'], d3)
 
@@ -208,12 +228,12 @@ class FuseTagsUpdateTest(MountTestBase):
     def runTest(self):
         self.make_mount(fuse.TagsDirectory, poll_time=1)
 
-        self.assertIn('foo_tag', os.listdir(self.mounttmp))
+        self.assertIn('foo_tag', llfuse.listdir(self.mounttmp))
 
         bar_uuid = run_test_server.fixture('collections')['bar_file']['uuid']
         self.tag_collection(bar_uuid, 'fuse_test_tag')
         time.sleep(1)
-        self.assertIn('fuse_test_tag', os.listdir(self.mounttmp))
+        self.assertIn('fuse_test_tag', llfuse.listdir(self.mounttmp))
         self.assertDirContents('fuse_test_tag', [bar_uuid])
 
         baz_uuid = run_test_server.fixture('collections')['baz_file']['uuid']
@@ -234,24 +254,23 @@ class FuseSharedTest(MountTestBase):
         # shared_dirs is a list of the directories exposed
         # by fuse.SharedDirectory (i.e. any object visible
         # to the current user)
-        shared_dirs = os.listdir(self.mounttmp)
+        shared_dirs = llfuse.listdir(self.mounttmp)
         shared_dirs.sort()
         self.assertIn('FUSE User', shared_dirs)
 
         # fuse_user_objs is a list of the objects owned by the FUSE
         # test user (which present as files in the 'FUSE User'
         # directory)
-        fuse_user_objs = os.listdir(os.path.join(self.mounttmp, 'FUSE User'))
+        fuse_user_objs = llfuse.listdir(os.path.join(self.mounttmp, 'FUSE User'))
         fuse_user_objs.sort()
-        self.assertEqual(['Empty collection.link',                # permission link on collection
-                          'FUSE Test Project',                    # project owned by user
+        self.assertEqual(['FUSE Test Project',                    # project owned by user
                           'collection #1 owned by FUSE',          # collection owned by user
                           'collection #2 owned by FUSE',          # collection owned by user
                           'pipeline instance owned by FUSE.pipelineInstance',  # pipeline instance owned by user
                       ], fuse_user_objs)
 
         # test_proj_files is a list of the files in the FUSE Test Project.
-        test_proj_files = os.listdir(os.path.join(self.mounttmp, 'FUSE User', 'FUSE Test Project'))
+        test_proj_files = llfuse.listdir(os.path.join(self.mounttmp, 'FUSE User', 'FUSE Test Project'))
         test_proj_files.sort()
         self.assertEqual(['collection in FUSE project',
                           'pipeline instance in FUSE project.pipelineInstance',
@@ -286,10 +305,10 @@ class FuseHomeTest(MountTestBase):
         self.make_mount(fuse.ProjectDirectory,
                         project_object=self.api.users().current().execute())
 
-        d1 = os.listdir(self.mounttmp)
+        d1 = llfuse.listdir(self.mounttmp)
         self.assertIn('Unrestricted public data', d1)
 
-        d2 = os.listdir(os.path.join(self.mounttmp, 'Unrestricted public data'))
+        d2 = llfuse.listdir(os.path.join(self.mounttmp, 'Unrestricted public data'))
         public_project = run_test_server.fixture('groups')[
             'anonymously_accessible_project']
         found_in = 0
@@ -309,9 +328,322 @@ class FuseHomeTest(MountTestBase):
         self.assertNotEqual(0, found_in)
         self.assertNotEqual(0, found_not_in)
 
-        d3 = os.listdir(os.path.join(self.mounttmp, 'Unrestricted public data', 'GNU General Public License, version 3'))
+        d3 = llfuse.listdir(os.path.join(self.mounttmp, 'Unrestricted public data', 'GNU General Public License, version 3'))
         self.assertEqual(["GNU_General_Public_License,_version_3.pdf"], d3)
 
+class FuseUpdateFileTest(MountTestBase):
+    def runTest(self):
+        collection = arvados.collection.Collection(api_client=self.api)
+        with collection.open("file1.txt", "w") as f:
+            f.write("blub")
+
+        collection.save_new()
+
+        m = self.make_mount(fuse.CollectionDirectory)
+        with llfuse.lock:
+            m.new_collection(None, collection)
+
+        d1 = llfuse.listdir(self.mounttmp)
+        self.assertEqual(["file1.txt"], d1)
+        with open(os.path.join(self.mounttmp, "file1.txt")) as f:
+            self.assertEqual("blub", f.read())
+
+        with collection.open("file1.txt", "w") as f:
+            f.write("plnp")
+
+        d1 = llfuse.listdir(self.mounttmp)
+        self.assertEqual(["file1.txt"], d1)
+        with open(os.path.join(self.mounttmp, "file1.txt")) as f:
+            self.assertEqual("plnp", f.read())
+
+class FuseAddFileToCollectionTest(MountTestBase):
+    def runTest(self):
+        collection = arvados.collection.Collection(api_client=self.api)
+        with collection.open("file1.txt", "w") as f:
+            f.write("blub")
+
+        collection.save_new()
+
+        m = self.make_mount(fuse.CollectionDirectory)
+        with llfuse.lock:
+            m.new_collection(None, collection)
+
+        d1 = llfuse.listdir(self.mounttmp)
+        self.assertEqual(["file1.txt"], d1)
+
+        with collection.open("file2.txt", "w") as f:
+            f.write("plnp")
+
+        d1 = llfuse.listdir(self.mounttmp)
+        self.assertEqual(["file1.txt", "file2.txt"], sorted(d1))
+
+class FuseRemoveFileFromCollectionTest(MountTestBase):
+    def runTest(self):
+        collection = arvados.collection.Collection(api_client=self.api)
+        with collection.open("file1.txt", "w") as f:
+            f.write("blub")
+
+        with collection.open("file2.txt", "w") as f:
+            f.write("plnp")
+
+        collection.save_new()
+
+        m = self.make_mount(fuse.CollectionDirectory)
+        with llfuse.lock:
+            m.new_collection(None, collection)
+
+        d1 = llfuse.listdir(self.mounttmp)
+        self.assertEqual(["file1.txt", "file2.txt"], sorted(d1))
+
+        collection.remove("file2.txt")
+
+        d1 = llfuse.listdir(self.mounttmp)
+        self.assertEqual(["file1.txt"], d1)
+
+class FuseCreateFileTest(MountTestBase):
+    def runTest(self):
+        collection = arvados.collection.Collection(api_client=self.api)
+        collection.save_new()
+
+        collection2 = self.api.collections().get(uuid=collection.manifest_locator()).execute()
+        self.assertEqual(collection2["manifest_text"], "")
+
+        collection.save_new()
+
+        m = self.make_mount(fuse.CollectionDirectory)
+        with llfuse.lock:
+            m.new_collection(None, collection)
+        self.assertTrue(m.writable())
+
+        self.assertNotIn("file1.txt", collection)
+
+        with open(os.path.join(self.mounttmp, "file1.txt"), "w") as f:
+            pass
+
+        self.assertIn("file1.txt", collection)
+
+        d1 = llfuse.listdir(self.mounttmp)
+        self.assertEqual(["file1.txt"], d1)
+
+        collection2 = self.api.collections().get(uuid=collection.manifest_locator()).execute()
+        self.assertRegexpMatches(collection2["manifest_text"],
+            r'\. d41d8cd98f00b204e9800998ecf8427e\+0\+A[a-f0-9]{40}@[a-f0-9]{8} 0:0:file1\.txt$')
+
+def fuseWriteFileTestHelper(mounttmp):
+    with open(os.path.join(mounttmp, "file1.txt"), "r") as f:
+        return f.read() == "Hello world!"
+
+class FuseWriteFileTest(MountTestBase):
+    def runTest(self):
+        arvados.logger.setLevel(logging.DEBUG)
+
+        collection = arvados.collection.Collection(api_client=self.api)
+        collection.save_new()
+
+        m = self.make_mount(fuse.CollectionDirectory)
+        with llfuse.lock:
+            m.new_collection(None, collection)
+        self.assertTrue(m.writable())
+
+        self.assertNotIn("file1.txt", collection)
+
+        with open(os.path.join(self.mounttmp, "file1.txt"), "w") as f:
+            f.write("Hello world!")
+
+        with collection.open("file1.txt") as f:
+            self.assertEqual(f.read(), "Hello world!")
+
+        # We can't just open the collection for reading because the underlying
+        # C implementation of open() makes a fstat() syscall with the GIL still
+        # held.  When the GETATTR message comes back to llfuse (which in these
+        # tests is in the same interpreter process) it can't acquire the GIL,
+        # so it can't service the fstat() call, so it deadlocks.  The
+        # workaround is to run some of our test code in a separate process.
+        # Forturnately the multiprocessing module makes this relatively easy.
+        pool = multiprocessing.Pool(1)
+        self.assertTrue(pool.apply(fuseWriteFileTestHelper, (self.mounttmp,)))
+        pool.close()
+
+        collection2 = self.api.collections().get(uuid=collection.manifest_locator()).execute()
+        self.assertRegexpMatches(collection2["manifest_text"],
+            r'\. 86fb269d190d2c85f6e0468ceca42a20\+12\+A[a-f0-9]{40}@[a-f0-9]{8} 0:12:file1\.txt$')
+
+def fuseUpdateFileTestHelper1(mounttmp):
+    with open(os.path.join(mounttmp, "file1.txt"), "r+") as f:
+        fr = f.read()
+        if fr != "Hello world!":
+            raise Exception("Got %s expected 'Hello world!'" % fr)
+        f.seek(0)
+        f.write("Hola mundo!")
+        f.seek(0)
+        fr = f.read()
+        if fr != "Hola mundo!!":
+            raise Exception("Got %s expected 'Hola mundo!!'" % fr)
+        return True
+
+def fuseUpdateFileTestHelper2(mounttmp):
+    with open(os.path.join(mounttmp, "file1.txt"), "r") as f:
+        return f.read() == "Hola mundo!!"
+
+class FuseUpdateFileTest(MountTestBase):
+    def runTest(self):
+        arvados.logger.setLevel(logging.DEBUG)
+
+        collection = arvados.collection.Collection(api_client=self.api)
+        collection.save_new()
+
+        m = self.make_mount(fuse.CollectionDirectory)
+        with llfuse.lock:
+            m.new_collection(None, collection)
+        self.assertTrue(m.writable())
+
+        with open(os.path.join(self.mounttmp, "file1.txt"), "w") as f:
+            f.write("Hello world!")
+
+        # See note in FuseWriteFileTest
+        pool = multiprocessing.Pool(1)
+        self.assertTrue(pool.apply(fuseUpdateFileTestHelper1, (self.mounttmp,)))
+        self.assertTrue(pool.apply(fuseUpdateFileTestHelper2, (self.mounttmp,)))
+        pool.close()
+
+        collection2 = self.api.collections().get(uuid=collection.manifest_locator()).execute()
+        self.assertRegexpMatches(collection2["manifest_text"],
+            r'\. daaef200ebb921e011e3ae922dd3266b\+11\+A[a-f0-9]{40}@[a-f0-9]{8} 86fb269d190d2c85f6e0468ceca42a20\+12\+A[a-f0-9]{40}@[a-f0-9]{8} 0:11:file1\.txt 22:1:file1\.txt$')
+
+
+class FuseMkdirTest(MountTestBase):
+    def runTest(self):
+        arvados.logger.setLevel(logging.DEBUG)
+
+        collection = arvados.collection.Collection(api_client=self.api)
+        collection.save_new()
+
+        m = self.make_mount(fuse.CollectionDirectory)
+        with llfuse.lock:
+            m.new_collection(None, collection)
+        self.assertTrue(m.writable())
+
+        with self.assertRaises(IOError):
+            with open(os.path.join(self.mounttmp, "testdir", "file1.txt"), "w") as f:
+                f.write("Hello world!")
+
+        os.mkdir(os.path.join(self.mounttmp, "testdir"))
+
+        with self.assertRaises(OSError):
+            os.mkdir(os.path.join(self.mounttmp, "testdir"))
+
+        d1 = llfuse.listdir(self.mounttmp)
+        self.assertEqual(["testdir"], d1)
+
+        with open(os.path.join(self.mounttmp, "testdir", "file1.txt"), "w") as f:
+            f.write("Hello world!")
+
+        d1 = llfuse.listdir(os.path.join(self.mounttmp, "testdir"))
+        self.assertEqual(["file1.txt"], d1)
+
+        collection2 = self.api.collections().get(uuid=collection.manifest_locator()).execute()
+        self.assertRegexpMatches(collection2["manifest_text"],
+            r'\./testdir 86fb269d190d2c85f6e0468ceca42a20\+12\+A[a-f0-9]{40}@[a-f0-9]{8} 0:12:file1\.txt$')
+
+
+class FuseRmTest(MountTestBase):
+    def runTest(self):
+        arvados.logger.setLevel(logging.DEBUG)
+
+        collection = arvados.collection.Collection(api_client=self.api)
+        collection.save_new()
+
+        m = self.make_mount(fuse.CollectionDirectory)
+        with llfuse.lock:
+            m.new_collection(None, collection)
+        self.assertTrue(m.writable())
+
+        os.mkdir(os.path.join(self.mounttmp, "testdir"))
+
+        with open(os.path.join(self.mounttmp, "testdir", "file1.txt"), "w") as f:
+            f.write("Hello world!")
+
+        # Starting manifest
+        collection2 = self.api.collections().get(uuid=collection.manifest_locator()).execute()
+        self.assertRegexpMatches(collection2["manifest_text"],
+            r'\./testdir 86fb269d190d2c85f6e0468ceca42a20\+12\+A[a-f0-9]{40}@[a-f0-9]{8} 0:12:file1\.txt$')
+
+        # Can't delete because it's not empty
+        with self.assertRaises(OSError):
+            os.rmdir(os.path.join(self.mounttmp, "testdir"))
+
+        d1 = llfuse.listdir(os.path.join(self.mounttmp, "testdir"))
+        self.assertEqual(["file1.txt"], d1)
+
+        # Delete file
+        os.remove(os.path.join(self.mounttmp, "testdir", "file1.txt"))
+
+        # Make sure it's empty
+        d1 = llfuse.listdir(os.path.join(self.mounttmp, "testdir"))
+        self.assertEqual([], d1)
+
+        # Try to delete it again
+        with self.assertRaises(OSError):
+            os.remove(os.path.join(self.mounttmp, "testdir", "file1.txt"))
+
+        # Can't have empty directories :-( so manifest will be empty.
+        collection2 = self.api.collections().get(uuid=collection.manifest_locator()).execute()
+        self.assertEqual(collection2["manifest_text"], "")
+
+        # Should be able to delete now that it is empty
+        os.rmdir(os.path.join(self.mounttmp, "testdir"))
+
+        # Make sure it's empty
+        d1 = llfuse.listdir(os.path.join(self.mounttmp))
+        self.assertEqual([], d1)
+
+        # Try to delete it again
+        with self.assertRaises(OSError):
+            os.rmdir(os.path.join(self.mounttmp, "testdir"))
+
+        # manifest should be empty now.
+        collection2 = self.api.collections().get(uuid=collection.manifest_locator()).execute()
+        self.assertEqual(collection2["manifest_text"], "")
+
+
+class FuseMvTest(MountTestBase):
+    def runTest(self):
+        arvados.logger.setLevel(logging.DEBUG)
+
+        collection = arvados.collection.Collection(api_client=self.api)
+        collection.save_new()
+
+        m = self.make_mount(fuse.CollectionDirectory)
+        with llfuse.lock:
+            m.new_collection(None, collection)
+        self.assertTrue(m.writable())
+
+        os.mkdir(os.path.join(self.mounttmp, "testdir"))
+
+        with open(os.path.join(self.mounttmp, "testdir", "file1.txt"), "w") as f:
+            f.write("Hello world!")
+
+        # Starting manifest
+        collection2 = self.api.collections().get(uuid=collection.manifest_locator()).execute()
+        self.assertRegexpMatches(collection2["manifest_text"],
+            r'\./testdir 86fb269d190d2c85f6e0468ceca42a20\+12\+A[a-f0-9]{40}@[a-f0-9]{8} 0:12:file1\.txt$')
+
+        d1 = llfuse.listdir(os.path.join(self.mounttmp))
+        self.assertEqual(["testdir"], d1)
+        d1 = llfuse.listdir(os.path.join(self.mounttmp, "testdir"))
+        self.assertEqual(["file1.txt"], d1)
+
+        os.rename(os.path.join(self.mounttmp, "testdir", "file1.txt"), os.path.join(self.mounttmp, "file1.txt"))
+
+        d1 = llfuse.listdir(os.path.join(self.mounttmp))
+        self.assertEqual(["file1.txt", "testdir"], sorted(d1))
+        d1 = llfuse.listdir(os.path.join(self.mounttmp, "testdir"))
+        self.assertEqual([], d1)
+
+        collection2 = self.api.collections().get(uuid=collection.manifest_locator()).execute()
+        self.assertRegexpMatches(collection2["manifest_text"],
+            r'\. 86fb269d190d2c85f6e0468ceca42a20\+12\+A[a-f0-9]{40}@[a-f0-9]{8} 0:12:file1\.txt$')
+
 
 class FuseUnitTest(unittest.TestCase):
     def test_sanitize_filename(self):