def test_uri_parsing(self):
self.assertEqual(is_uri('notauri!'), False)
self.assertEqual(is_uri('they://did/the/monster/mash'), True)
self.assertEqual(is_s3_uri('s3://a/uri'), True)
self.assertEqual(is_s3_uri('s3n://a/uri'), True)
self.assertEqual(is_s3_uri('hdfs://a/uri'), False)
self.assertEqual(parse_s3_uri('s3://bucket/loc'), ('bucket', 'loc'))
def test_uri_parsing(self):
self.assertEqual(is_uri('notauri!'), False)
self.assertEqual(is_uri('they://did/the/monster/mash'), True)
self.assertEqual(is_s3_uri('s3://a/uri'), True)
self.assertEqual(is_s3_uri('s3n://a/uri'), True)
self.assertEqual(is_s3_uri('hdfs://a/uri'), False)
self.assertEqual(parse_s3_uri('s3://bucket/loc'), ('bucket', 'loc'))
def join(self, dirname, filename):
"""Join *filename* onto *dirname* (which may be a URI)"""
if is_uri(filename):
return filename
elif is_uri(dirname):
return posixpath.join(dirname, filename)
else:
return os.path.join(dirname, filename)
def join(self, dirname, filename):
"""Join *filename* onto *dirname* (which may be a URI)"""
if is_uri(filename):
return filename
elif is_uri(dirname):
return posixpath.join(dirname, filename)
else:
return os.path.join(dirname, filename)
def _create_input_manifest_if_needed(self):
"""Create a file with a list of URIs of input files."""
if self._input_manifest_path or not self._uses_input_manifest():
return
uris = []
log.info('finding input files to add to manifest...')
for path in self._get_input_paths():
log.debug(' in %s' % path)
if is_uri(path):
# URIs might be globs
for uri in self.fs.ls(path):
uris.append(uri)
else:
# local paths are expected to be single files
# (shell would resolve globs)
if self._upload_mgr:
uris.append(self._upload_mgr.uri(path))
else:
# just make sure job can find files from it's working dir
uris.append(os.path.abspath(path))
log.info('found %d input files' % len(uris))
path = os.path.join(self._get_local_tmp_dir(), 'input-manifest.txt')
self._write_script(uris, path, 'input manifest')
self._input_manifest_path = path
if self._upload_mgr:
self._upload_mgr.add(self._input_manifest_path)
def test_default(self):
runner = SparkMRJobRunner()
self.assertFalse(is_uri(runner._spark_tmp_dir))
self.assertIsNone(runner._upload_mgr)
self.assertEqual(runner._spark_tmp_dir[-6:], '-spark')
def ls(self, path_glob):
if not is_uri(path_glob):
for path in super(HadoopJobRunner, self).ls(path_glob):
yield path
return
components = urlparse(path_glob)
hdfs_prefix = '%s://%s' % (components.scheme, components.netloc)
stdout = self._invoke_hadoop(
['fs', '-lsr', path_glob],
return_stdout=True,
ok_stderr=[HADOOP_LSR_NO_SUCH_FILE])
for line in StringIO(stdout):
fields = line.rstrip('\r\n').split()
# expect lines like:
# -rw-r--r-- 3 dave users 3276 2010-01-13 14:00 /foo/bar
if len(fields) < 8:
raise Exception('unexpected ls line from hadoop: %r' % line)
# ignore directories
if fields[0].startswith('d'):
continue
# not sure if you can have spaces in filenames; just to be safe
path = ' '.join(fields[7:])
yield hdfs_prefix + path
def test_spark_master_mesos(self):
runner = SparkMRJobRunner(spark_master='mesos://host:12345')
self.assertTrue(is_uri(runner._spark_tmp_dir))
self.assertEqual(runner._spark_tmp_dir[:8], 'hdfs:///')
self.assertIsNotNone(runner._upload_mgr)
def _cat_file(self, filename):
if is_uri(filename):
# stream from HDFS
cat_args = self._opts['hadoop_bin'] + ['fs', '-cat', filename]
log.debug('> %s' % cmd_line(cat_args))
cat_proc = Popen(cat_args, stdout=PIPE, stderr=PIPE)
def stream():
for line in cat_proc.stdout:
yield line
# there shouldn't be any stderr
for line in cat_proc.stderr:
log.error('STDERR: ' + line)
returncode = cat_proc.wait()
if returncode != 0:
raise CalledProcessError(returncode, cat_args)
return read_file(filename, stream())
else:
# read from local filesystem
return super(HadoopJobRunner, self)._cat_file(filename)
def test_spark_master_yarn(self):
runner = SparkMRJobRunner(spark_master='yarn')
self.assertTrue(is_uri(runner._spark_tmp_dir))
self.assertEqual(runner._spark_tmp_dir[:8], 'hdfs:///')
self.assertIsNotNone(runner._upload_mgr)
def _setup_input(self):
"""Copy local input files (if any) to a special directory on HDFS.
Set self._hdfs_input_files
"""
# winnow out HDFS files from local ones
self._hdfs_input_files = []
local_input_files = []
for path in self._input_paths:
if is_uri(path):
# Don't even bother running the job if the input isn't there.
if not self.ls(path):
raise AssertionError(
'Input path %s does not exist!' % (path,))
self._hdfs_input_files.append(path)
else:
local_input_files.append(path)
# copy local files into an input directory, with names like
# 00000-actual_name.ext
if local_input_files:
hdfs_input_dir = posixpath.join(self._hdfs_tmp_dir, 'input')
log.info('Uploading input to %s' % hdfs_input_dir)
self._mkdir_on_hdfs(hdfs_input_dir)
for i, path in enumerate(local_input_files):
if path == '-':
path = self._dump_stdin_to_local_file()
target = '%s/%05i-%s' % (
hdfs_input_dir, i, os.path.basename(path))
self._upload_to_hdfs(path, target)
self._hdfs_input_files.append(hdfs_input_dir)
def test_default(self):
runner = SparkMRJobRunner()
self.assertFalse(is_uri(runner._spark_tmp_dir))
self.assertIsNone(runner._upload_mgr)
self.assertEqual(runner._spark_tmp_dir[-6:], '-spark')
def _cat_file(self, filename):
if is_uri(filename):
# stream from HDFS
cat_args = self._opts['hadoop_bin'] + ['fs', '-cat', filename]
log.debug('> %s' % cmd_line(cat_args))
cat_proc = Popen(cat_args, stdout=PIPE, stderr=PIPE)
def stream():
for line in cat_proc.stdout:
yield line
# there shouldn't be any stderr
for line in cat_proc.stderr:
log.error('STDERR: ' + line)
returncode = cat_proc.wait()
if returncode != 0:
raise CalledProcessError(returncode, cat_args)
return read_file(filename, stream())
else:
# read from local filesystem
return super(HadoopJobRunner, self)._cat_file(filename)
def test_spark_master_yarn(self):
runner = SparkMRJobRunner(spark_master='yarn')
self.assertTrue(is_uri(runner._spark_tmp_dir))
self.assertEqual(runner._spark_tmp_dir[:8], 'hdfs:///')
self.assertIsNotNone(runner._upload_mgr)
def ls(self, path_glob):
if not is_uri(path_glob):
for path in super(HadoopJobRunner, self).ls(path_glob):
yield path
return
components = urlparse(path_glob)
hdfs_prefix = '%s://%s' % (components.scheme, components.netloc)
stdout = self._invoke_hadoop(['fs', '-lsr', path_glob],
return_stdout=True,
ok_stderr=[HADOOP_LSR_NO_SUCH_FILE])
for line in StringIO(stdout):
fields = line.rstrip('\r\n').split()
# expect lines like:
# -rw-r--r-- 3 dave users 3276 2010-01-13 14:00 /foo/bar
if len(fields) < 8:
raise Exception('unexpected ls line from hadoop: %r' % line)
# ignore directories
if fields[0].startswith('d'):
continue
# not sure if you can have spaces in filenames; just to be safe
path = ' '.join(fields[7:])
yield hdfs_prefix + path
def _create_input_manifest_if_needed(self):
"""Create a file with a list of URIs of input files."""
if self._input_manifest_path or not self._uses_input_manifest():
return
uris = []
log.info('finding input files to add to manifest...')
for path in self._get_input_paths():
log.debug(' in %s' % path)
if is_uri(path):
# URIs might be globs
for uri in self.fs.ls(path):
uris.append(uri)
else:
# local paths are expected to be single files
# (shell would resolve globs)
if self._upload_mgr:
uris.append(self._upload_mgr.uri(path))
else:
# just make sure job can find files from it's working dir
uris.append(os.path.abspath(path))
log.info('found %d input files' % len(uris))
path = os.path.join(self._get_local_tmp_dir(), 'input-manifest.txt')
self._write_script(uris, path, 'input manifest')
self._input_manifest_path = path
if self._upload_mgr:
self._upload_mgr.add(self._input_manifest_path)
def test_spark_master_mesos(self):
runner = SparkMRJobRunner(spark_master='mesos://host:12345')
self.assertTrue(is_uri(runner._spark_tmp_dir))
self.assertEqual(runner._spark_tmp_dir[:8], 'hdfs:///')
self.assertIsNotNone(runner._upload_mgr)
def _fully_qualify_subnetwork_uri(uri, project_id, region):
if '/' not in uri: # just a name
uri = 'projects/%s/%s/subnetworks/%s' % (project_id, region, uri)
if not is_uri(uri):
uri = 'https://www.googleapis.com/compute/v1/' + uri
return uri
def fully_qualify_hdfs_path(path):
"""If path isn't an ``hdfs://`` URL, turn it into one."""
if is_uri(path):
return path
elif path.startswith('/'):
return 'hdfs://' + path
else:
return 'hdfs:///user/%s/%s' % (getpass.getuser(), path)
def _from_file_uri(path_or_uri):
if is_uri(path_or_uri):
if path_or_uri.startswith('file:///'):
return path_or_uri[7:] # keep last /
else:
raise ValueError('Not a file:/// URI')
else:
return path_or_uri
def fully_qualify_hdfs_path(path):
"""If path isn't an ``hdfs://`` URL, turn it into one."""
if is_uri(path):
return path
elif path.startswith('/'):
return 'hdfs://' + path
else:
return 'hdfs:///user/%s/%s' % (getpass.getuser(), path)
def test_copy_files_with_rename_to_remote_wd_mirror(self):
self.add_mock_s3_data(
{'walrus': {
'fish': b'salmon',
'fowl': b'goose'
}})
foe_path = self.makefile('foe', b'giant')
run_spark_submit = self.start(
patch('mrjob.bin.MRJobBinRunner._run_spark_submit',
return_value=0))
job = MRSparkOSWalk([
'-r', 'spark', '--spark-master', 'mesos://host:9999',
'--spark-tmp-dir', 's3://walrus/tmp', '--file',
's3://walrus/fish#ghoti', '--file', 's3://walrus/fowl', '--file',
foe_path
])
job.sandbox()
with job.make_runner() as runner:
runner.run()
# check working dir mirror
wd_mirror = runner._wd_mirror()
fs = runner.fs
self.assertIsNotNone(wd_mirror)
self.assertTrue(is_uri(wd_mirror))
self.assertTrue(fs.exists(wd_mirror))
# uploaded for rename
self.assertTrue(fs.exists(fs.join(wd_mirror, 'ghoti')))
# wrong name
self.assertFalse(fs.exists(fs.join(wd_mirror, 'fish')))
# no need to upload, already visible
self.assertFalse(fs.exists(fs.join(wd_mirror, 'fowl')))
# need to upload from local to remote
self.assertTrue(fs.exists(fs.join(wd_mirror, 'foe')))
run_spark_submit.assert_called_once_with(ANY,
ANY,
record_callback=ANY)
spark_submit_args = run_spark_submit.call_args[0][0]
self.assertIn('--files', spark_submit_args)
files_arg = spark_submit_args[spark_submit_args.index('--files') +
1]
self.assertEqual(
files_arg, ','.join([
fs.join(wd_mirror, 'foe'),
's3://walrus/fowl',
fs.join(wd_mirror, 'ghoti'),
fs.join(wd_mirror, 'mr_spark_os_walk.py'),
]))
def ls(self, path_glob):
components = urlparse(path_glob)
hdfs_prefix = '%s://%s' % (components.scheme, components.netloc)
version = self.get_hadoop_version()
# use ls -R on Hadoop 2 (see #1152)
if uses_yarn(version):
args = ['fs', '-ls', '-R', path_glob]
else:
args = ['fs', '-lsr', path_glob]
try:
stdout = self.invoke_hadoop(args,
return_stdout=True,
ok_stderr=[_HADOOP_LS_NO_SUCH_FILE])
except CalledProcessError:
raise IOError("Could not ls %s" % path_glob)
for line in BytesIO(stdout):
line = line.rstrip(b'\r\n')
# ignore total item count
if line.startswith(b'Found '):
continue
fields = line.split(b' ')
# Throw out directories
if fields[0].startswith(b'd'):
continue
# Try to figure out which part of the line is the path
# Expected lines:
#
# HDFS:
# -rw-r--r-- 3 dave users 3276 2010-01-13 14:00 /foo/bar
#
# S3:
# -rwxrwxrwx 1 3276 010-01-13 14:00 /foo/bar
path_index = None
for index, field in enumerate(fields):
# look for time field, and pick one after that
# (can't use field[2] because that's an int in Python 3)
if len(field) == 5 and field[2:3] == b':':
path_index = (index + 1)
if not path_index:
raise IOError("Could not locate path in string %r" % line)
path = to_string(line.split(b' ', path_index)[-1])
# handle fully qualified URIs from newer versions of Hadoop ls
# (see Pull Request #577)
if is_uri(path):
yield path
else:
yield hdfs_prefix + path
def uri(self, path):
"""Get the URI for the given path. If *path* is a URI, just return it.
"""
if is_uri(path):
return path
if path in self._path_to_name:
return posixpath.join(self.prefix, self._path_to_name[path])
else:
raise ValueError('%r is not a URI or a known local file' % (path,))
def uri(self, path):
"""Get the URI for the given path. If *path* is a URI, just return it.
"""
if (not os.path.exists(path)) and is_uri(path):
return path
if path in self._path_to_name:
return posixpath.join(self.prefix, self._path_to_name[path])
else:
raise ValueError('%r is not a URI or a known local file' % (path,))
def ls(self, path_glob):
components = urlparse(path_glob)
hdfs_prefix = '%s://%s' % (components.scheme, components.netloc)
version = self.get_hadoop_version()
# use ls -R on Hadoop 2 (see #1152)
if uses_yarn(version):
args = ['fs', '-ls', '-R', path_glob]
else:
args = ['fs', '-lsr', path_glob]
try:
stdout = self.invoke_hadoop(args, return_stdout=True,
ok_stderr=[_HADOOP_LS_NO_SUCH_FILE])
except CalledProcessError:
raise IOError("Could not ls %s" % path_glob)
for line in BytesIO(stdout):
line = line.rstrip(b'\r\n')
# ignore total item count
if line.startswith(b'Found '):
continue
fields = line.split(b' ')
# Throw out directories
if fields[0].startswith(b'd'):
continue
# Try to figure out which part of the line is the path
# Expected lines:
#
# HDFS:
# -rw-r--r-- 3 dave users 3276 2010-01-13 14:00 /foo/bar
#
# S3:
# -rwxrwxrwx 1 3276 010-01-13 14:00 /foo/bar
path_index = None
for index, field in enumerate(fields):
# look for time field, and pick one after that
# (can't use field[2] because that's an int in Python 3)
if len(field) == 5 and field[2:3] == b':':
path_index = (index + 1)
if not path_index:
raise IOError("Could not locate path in string %r" % line)
path = to_unicode(line.split(b' ', path_index)[-1])
# handle fully qualified URIs from newer versions of Hadoop ls
# (see Pull Request #577)
if is_uri(path):
yield path
else:
yield hdfs_prefix + path
def path_exists(self, path_glob):
"""Does the given path exist?
If dest is a directory (ends with a "/"), we check if there are
any files starting with that path.
"""
if not is_uri(path_glob):
return super(HadoopJobRunner, self).path_exists(path_glob)
return bool(self._invoke_hadoop(['fs', '-test', '-e', path_glob],
ok_returncodes=(0, 1)))
def _endpoint_url(host_or_uri):
"""If *host_or_uri* is non-empty and isn't a URI, prepend ``'https://'``.
Otherwise, pass through as-is.
"""
if not host_or_uri:
return host_or_uri
elif is_uri(host_or_uri):
return host_or_uri
else:
return 'https://' + host_or_uri
def _endpoint_url(host_or_uri):
"""If *host_or_uri* is non-empty and isn't a URI, prepend ``'https://'``.
Otherwise, pass through as-is.
"""
if not host_or_uri:
return host_or_uri
elif is_uri(host_or_uri):
return host_or_uri
else:
return 'https://' + host_or_uri
def path_exists(self, path_glob):
"""Does the given path exist?
If dest is a directory (ends with a "/"), we check if there are
any files starting with that path.
"""
if not is_uri(path_glob):
return super(HadoopJobRunner, self).path_exists(path_glob)
return bool(
self._invoke_hadoop(['fs', '-test', '-e', path_glob],
ok_returncodes=(0, 1)))
def test_copy_files_with_rename_to_remote_wd_mirror(self):
self.add_mock_s3_data({'walrus': {'fish': b'salmon',
'fowl': b'goose'}})
foe_path = self.makefile('foe', b'giant')
run_spark_submit = self.start(patch(
'mrjob.bin.MRJobBinRunner._run_spark_submit',
return_value=0))
job = MRSparkOSWalk(['-r', 'spark',
'--spark-master', 'mesos://host:9999',
'--spark-tmp-dir', 's3://walrus/tmp',
'--file', 's3://walrus/fish#ghoti',
'--file', 's3://walrus/fowl',
'--file', foe_path])
job.sandbox()
with job.make_runner() as runner:
runner.run()
# check working dir mirror
wd_mirror = runner._wd_mirror()
fs = runner.fs
self.assertIsNotNone(wd_mirror)
self.assertTrue(is_uri(wd_mirror))
self.assertTrue(fs.exists(wd_mirror))
# uploaded for rename
self.assertTrue(fs.exists(fs.join(wd_mirror, 'ghoti')))
# wrong name
self.assertFalse(fs.exists(fs.join(wd_mirror, 'fish')))
# no need to upload, already visible
self.assertFalse(fs.exists(fs.join(wd_mirror, 'fowl')))
# need to upload from local to remote
self.assertTrue(fs.exists(fs.join(wd_mirror, 'foe')))
run_spark_submit.assert_called_once_with(
ANY, ANY, record_callback=ANY)
spark_submit_args = run_spark_submit.call_args[0][0]
self.assertIn('--files', spark_submit_args)
files_arg = spark_submit_args[
spark_submit_args.index('--files') + 1]
self.assertEqual(
files_arg, ','.join([
fs.join(wd_mirror, 'foe'),
's3://walrus/fowl',
fs.join(wd_mirror, 'ghoti'),
fs.join(wd_mirror, 'mr_spark_os_walk.py'),
]))
def join(self, path, *paths):
"""Join *paths* onto *path* (which may be a URI)"""
all_paths = (path,) + paths
# if there's a URI, we only care about it and what follows
for i in range(len(all_paths), 0, -1):
if is_uri(all_paths[i - 1]):
scheme, netloc, uri_path = urlparse(all_paths[i - 1])[:3]
return '%s://%s%s' % (
scheme, netloc, posixpath.join(
uri_path or '/', *all_paths[i:]))
else:
return os.path.join(*all_paths)
def add(self, path):
"""Add a path. If *path* hasn't been added before, assign it a name.
If *path* is a URI don't add it; just return the URI.
:return: the URI assigned to the path"""
if is_uri(path):
return path
if path not in self._path_to_name:
name = name_uniquely(path, names_taken=self._names_taken)
self._names_taken.add(name)
self._path_to_name[path] = name
return self.uri(path)
def du(self, path_glob):
"""Get the size of a file, or None if it's not a file or doesn't
exist."""
if not is_uri(path_glob):
return super(HadoopJobRunner, self).dus(path_glob)
stdout = self._invoke_hadoop(['fs', '-du', path_glob],
return_stdout=True)
try:
return int(stdout.split()[1])
except (ValueError, TypeError, IndexError):
raise Exception('Unexpected output from hadoop fs -du: %r' %
stdout)
def du(self, path_glob):
"""Get the size of a file, or None if it's not a file or doesn't
exist."""
if not is_uri(path_glob):
return super(HadoopJobRunner, self).dus(path_glob)
stdout = self._invoke_hadoop(['fs', '-du', path_glob],
return_stdout=True)
try:
return int(stdout.split()[1])
except (ValueError, TypeError, IndexError):
raise Exception(
'Unexpected output from hadoop fs -du: %r' % stdout)
def add(self, path):
"""Add a path. If *path* hasn't been added before, assign it a name.
If *path* is a URI don't add it; just return the URI.
:return: the URI assigned to the path"""
if (not os.path.exists(path)) and is_uri(path):
return path
if path not in self._path_to_name:
name = name_uniquely(path, names_taken=self._names_taken)
self._names_taken.add(name)
self._path_to_name[path] = name
return self.uri(path)
def fully_qualify_hadoop_path(path):
"""If we're on MapR, we should get an alternative to hdfs://. CDH4 will fail"""
process = Popen(HADOOP_FETCH_URI_SCHEME, shell=True, stdout=PIPE, stderr=STDOUT)
uri_scheme = process.communicate()[0]
if process.returncode != 0:
uri_scheme='hdfs://'
else:
uri_scheme = HADOOP_FETCH_URI_CLEANUP.sub('//', uri_scheme)
if is_uri(path):
return path
elif path.startswith('/'):
return uri_scheme + path
else:
return '%s/user/%s/%s' % (uri_scheme, getpass.getuser(), path)
def setUp(self):
super(CompositeFilesystemTestCase, self).setUp()
self.log = self.start(patch('mrjob.fs.composite.log'))
self.hadoop_fs = Mock(spec=Filesystem)
self.hadoop_fs.get_hadoop_version = Mock()
self.hadoop_fs.can_handle_path.side_effect = is_uri
self.local_fs = Mock(spec=Filesystem)
self.local_fs.can_handle_path.side_effect = lambda p: not is_uri(p)
self.s3_fs = Mock(spec=Filesystem)
self.s3_fs.create_bucket = Mock()
self.s3_fs.can_handle_path.side_effect = is_s3_uri
def test_is_uri(self):
self.assertEqual(is_uri('notauri!'), False)
self.assertEqual(is_uri('they://did/the/monster/mash'), True)
self.assertEqual(is_uri('C:\some\windows\path'), False)
# test #1455
self.assertEqual(is_uri('2016-10-11T06:29:17'), False)
# sorry, we only care about file URIs
self.assertEqual(is_uri('mailto:[email protected]'), False)
# urlparse has to accept it
self.assertEqual(is_uri('://'), False)
def rm(self, path_glob):
if not is_uri(path_glob):
super(HadoopJobRunner, self).rm(path_glob)
if self.path_exists(path_glob):
# hadoop fs -rmr will print something like:
# Moved to trash: hdfs://hdnamenode:54310/user/dave/asdf
# to STDOUT, which we don't care about.
#
# if we ask to delete a path that doesn't exist, it prints
# to STDERR something like:
# rmr: <path>
# which we can safely ignore
self._invoke_hadoop(
['fs', '-rmr', path_glob],
return_stdout=True, ok_stderr=[HADOOP_RMR_NO_SUCH_FILE])
def rm(self, path_glob):
if not is_uri(path_glob):
super(HadoopFilesystem, self).rm(path_glob)
version = self.get_hadoop_version()
if uses_yarn(version):
args = ['fs', '-rm', '-R', '-f', '-skipTrash', path_glob]
else:
args = ['fs', '-rmr', '-skipTrash', path_glob]
try:
self.invoke_hadoop(args,
return_stdout=True,
ok_stderr=[_HADOOP_RM_NO_SUCH_FILE])
except CalledProcessError:
raise IOError("Could not rm %s" % path_glob)
def _dir_archive_path(self, dir_path):
"""Assign a path for the archive of *dir_path* but don't
actually create anything."""
if dir_path not in self._dir_to_archive_path:
# we can check local paths now
if not (is_uri(dir_path) or os.path.isdir(dir_path)):
raise OSError('%s is not a directory!' % dir_path)
name = name_uniquely(
dir_path, names_taken=self._dir_archive_names_taken)
self._dir_archive_names_taken.add(name)
self._dir_to_archive_path[dir_path] = os.path.join(
self._get_local_tmp_dir(), 'archives', name + '.tar.gz')
return self._dir_to_archive_path[dir_path]
def rm(self, path_glob):
if not is_uri(path_glob):
super(HadoopJobRunner, self).rm(path_glob)
if self.path_exists(path_glob):
# hadoop fs -rmr will print something like:
# Moved to trash: hdfs://hdnamenode:54310/user/dave/asdf
# to STDOUT, which we don't care about.
#
# if we ask to delete a path that doesn't exist, it prints
# to STDERR something like:
# rmr: <path>
# which we can safely ignore
self._invoke_hadoop(['fs', '-rmr', path_glob],
return_stdout=True,
ok_stderr=[HADOOP_RMR_NO_SUCH_FILE])
def _check_input_exists(self):
"""Make sure all input exists before continuing with our job.
"""
if not self._opts['check_input_paths']:
return
for path in self._input_paths:
if path == '-':
continue # STDIN always exists
if is_uri(path) and not is_gcs_uri(path):
continue # can't check non-GCS URIs, hope for the best
if not self.fs.exists(path):
raise AssertionError(
'Input path %s does not exist!' % (path,))
def rm(self, path_glob):
if not is_uri(path_glob):
super(HadoopFilesystem, self).rm(path_glob)
version = self.get_hadoop_version()
if uses_yarn(version):
args = ['fs', '-rm', '-R', '-f', '-skipTrash', path_glob]
else:
args = ['fs', '-rmr', '-skipTrash', path_glob]
try:
self.invoke_hadoop(
args,
return_stdout=True, ok_stderr=[_HADOOP_RM_NO_SUCH_FILE])
except CalledProcessError:
raise IOError("Could not rm %s" % path_glob)
def _check_input_exists(self):
"""Make sure all input exists before continuing with our job.
"""
if not self._opts['check_input_paths']:
return
for path in self._input_paths:
if path == '-':
continue # STDIN always exists
if is_uri(path) and not is_gcs_uri(path):
continue # can't check non-GCS URIs, hope for the best
if not self.fs.exists(path):
raise AssertionError('Input path %s does not exist!' %
(path, ))
def __init__(self, **kwargs):
super(SparkMRJobRunner, self).__init__(**kwargs)
self._spark_tmp_dir = self._pick_spark_tmp_dir()
# where local files are uploaded into Spark
if is_uri(self._spark_tmp_dir):
spark_files_dir = posixpath.join(self._spark_tmp_dir, 'files', '')
self._upload_mgr = UploadDirManager(spark_files_dir)
# where to put job output (if not set explicitly)
if not self._output_dir:
self._output_dir = posixpath.join(self._spark_tmp_dir, 'output')
# keep track of where the spark-submit binary is
self._spark_submit_bin = self._opts['spark_submit_bin']
def _dir_archive_path(self, dir_path):
"""Assign a path for the archive of *dir_path* but don't
actually create anything."""
if dir_path not in self._dir_to_archive_path:
# we can check local paths now
if not (is_uri(dir_path) or os.path.isdir(dir_path)):
raise OSError('%s is not a directory!' % dir_path)
name = name_uniquely(dir_path,
names_taken=self._dir_archive_names_taken)
self._dir_archive_names_taken.add(name)
self._dir_to_archive_path[dir_path] = os.path.join(
self._get_local_tmp_dir(), 'archives', name + '.tar.gz')
return self._dir_to_archive_path[dir_path]
def _create_dir_archive(self, dir_path):
"""Helper for :py:meth:`archive_dir`"""
if not self.fs.exists(dir_path):
raise OSError('%s does not exist')
tar_gz_path = self._dir_archive_path(dir_path)
if tar_gz_path in self._dir_archives_created:
return # already created
if not os.path.isdir(os.path.dirname(tar_gz_path)):
os.makedirs(os.path.dirname(tar_gz_path))
# for remote files
tmp_download_path = os.path.join(self._get_local_tmp_dir(),
'tmp-download')
log.info('Archiving %s -> %s' % (dir_path, tar_gz_path))
with tarfile.open(tar_gz_path, mode='w:gz') as tar_gz:
for path in self.fs.ls(dir_path):
# fs.ls() only lists files
if path == dir_path:
raise OSError('%s is a file, not a directory!' % dir_path)
# TODO: do we need this?
if os.path.realpath(path) == os.path.realpath(tar_gz_path):
raise OSError('attempted to archive %s into itself!' %
tar_gz_path)
if is_uri(path):
path_in_tar_gz = path[len(dir_path):].lstrip('/')
log.info(' downloading %s -> %s' %
(path, tmp_download_path))
with open(tmp_download_path, 'wb') as f:
for chunk in self.fs.cat(path):
f.write(chunk)
local_path = tmp_download_path
else:
path_in_tar_gz = path[len(dir_path):].lstrip(os.sep)
local_path = path
log.debug(' adding %s to %s' % (path, tar_gz_path))
tar_gz.add(local_path, path_in_tar_gz, recursive=False)
self._dir_archives_created.add(tar_gz_path)
def _create_dir_archive(self, dir_path):
"""Helper for :py:meth:`archive_dir`"""
if not self.fs.exists(dir_path):
raise OSError('%s does not exist')
tar_gz_path = self._dir_archive_path(dir_path)
if tar_gz_path in self._dir_archives_created:
return # already created
if not os.path.isdir(os.path.dirname(tar_gz_path)):
os.makedirs(os.path.dirname(tar_gz_path))
# for remote files
tmp_download_path = os.path.join(
self._get_local_tmp_dir(), 'tmp-download')
log.info('Archiving %s -> %s' % (dir_path, tar_gz_path))
with tarfile.open(tar_gz_path, mode='w:gz') as tar_gz:
for path in self.fs.ls(dir_path):
# fs.ls() only lists files
if path == dir_path:
raise OSError('%s is a file, not a directory!' % dir_path)
# TODO: do we need this?
if os.path.realpath(path) == os.path.realpath(tar_gz_path):
raise OSError(
'attempted to archive %s into itself!' % tar_gz_path)
if is_uri(path):
path_in_tar_gz = path[len(dir_path):].lstrip('/')
log.info(' downloading %s -> %s' % (
path, tmp_download_path))
with open(tmp_download_path, 'wb') as f:
for chunk in self.fs.cat(path):
f.write(chunk)
local_path = tmp_download_path
else:
path_in_tar_gz = path[len(dir_path):].lstrip(os.sep)
local_path = path
log.debug(' adding %s to %s' % (path, tar_gz_path))
tar_gz.add(local_path, path_in_tar_gz, recursive=False)
self._dir_archives_created.add(tar_gz_path)
def add(self, path):
"""Add a path. If *path* hasn't been added before, assign it a name.
If *path* is a URI don't add it; just return the URI.
:return: the URI assigned to the path"""
if is_uri(path):
return path
if path not in self._path_to_name:
# use unhide so that input files won't be hidden from Hadoop,
# see #1200
name = name_uniquely(
path, names_taken=self._names_taken, unhide=True)
self._names_taken.add(name)
self._path_to_name[path] = name
return self.uri(path)
def _upload_non_input_files(self):
"""Copy files to HDFS, and set the 'hdfs_uri' field for each file.
"""
self._pick_hdfs_uris_for_files()
hdfs_files_dir = posixpath.join(self._hdfs_tmp_dir, 'files', '')
self._mkdir_on_hdfs(hdfs_files_dir)
log.info('Copying non-input files into %s' % hdfs_files_dir)
for file_dict in self._files:
path = file_dict['path']
# don't bother with files already in HDFS
if is_uri(path):
continue
self._upload_to_hdfs(path, file_dict['hdfs_uri'])
def add(self, path):
"""Add a path. If *path* hasn't been added before, assign it a name.
If *path* is a URI don't add it; just return the URI.
:return: the URI assigned to the path"""
if is_uri(path):
return path
if path not in self._path_to_name:
# use unhide so that input files won't be hidden from Hadoop,
# see #1200
name = name_uniquely(
path, names_taken=self._names_taken, unhide=True)
self._names_taken.add(name)
self._path_to_name[path] = name
return self.uri(path)
def rm(self, path_glob):
if not is_uri(path_glob):
super(HadoopFilesystem, self).rm(path_glob)
# hadoop fs -rmr will print something like:
# Moved to trash: hdfs://hdnamenode:54310/user/dave/asdf
# to STDOUT, which we don't care about.
#
# if we ask to delete a path that doesn't exist, it prints
# to STDERR something like:
# rmr: <path>
# which we can safely ignore
try:
self.invoke_hadoop(
['fs', '-rmr', path_glob],
return_stdout=True, ok_stderr=[HADOOP_RMR_NO_SUCH_FILE])
except CalledProcessError:
raise IOError("Could not rm %s" % path_glob)
def test_copy_files_with_rename_to_local_wd_mirror(self):
# see test_upload_files_with_rename() in test_local for comparison
fish_path = self.makefile('fish', b'salmon')
fowl_path = self.makefile('fowl', b'goose')
# use _LOCAL_CLUSTER_MASTER because the default master (local[*])
# doesn't have a working directory
job = MRSparkOSWalk(['-r', 'spark',
'--spark-master', _LOCAL_CLUSTER_MASTER,
'--file', fish_path + '#ghoti',
'--file', fowl_path])
job.sandbox()
file_sizes = {}
with job.make_runner() as runner:
runner.run()
# check working dir mirror
wd_mirror = runner._wd_mirror()
self.assertIsNotNone(wd_mirror)
self.assertFalse(is_uri(wd_mirror))
self.assertTrue(exists(wd_mirror))
# only files which needed to be renamed should be in wd_mirror
self.assertTrue(exists(join(wd_mirror, 'ghoti')))
self.assertFalse(exists(join(wd_mirror, 'fish')))
self.assertFalse(exists(join(wd_mirror, 'fowl')))
for line in to_lines(runner.cat_output()):
path, size = safeeval(line)
file_sizes[path] = size
# check that files were uploaded to working dir
self.assertIn('fowl', file_sizes)
self.assertEqual(file_sizes['fowl'], 5)
self.assertIn('ghoti', file_sizes)
self.assertEqual(file_sizes['ghoti'], 6)
# fish was uploaded as "ghoti"
self.assertNotIn('fish', file_sizes)
