Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

ARROW-404: [Python] Fix segfault caused by HdfsClient getting closed before an HdfsFile #230

Closed
wants to merge 3 commits into from
Closed
Show file tree
Hide file tree
Changes from 2 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
81 changes: 51 additions & 30 deletions python/pyarrow/io.pyx
Original file line number Diff line number Diff line change
Expand Up @@ -504,7 +504,7 @@ cdef class HdfsClient:

out.mode = mode
out.buffer_size = c_buffer_size
out.parent = self
out.parent = _HdfsFileNanny(self, out)
out.is_open = True
out.own_file = True

Expand All @@ -516,55 +516,76 @@ cdef class HdfsClient:
"""
write_queue = Queue(50)

f = self.open(path, 'wb')
with self.open(path, 'wb') as f:
done = False
exc_info = None
def bg_write():
try:
while not done or write_queue.qsize() > 0:
try:
buf = write_queue.get(timeout=0.01)
except QueueEmpty:
continue

done = False
exc_info = None
def bg_write():
try:
while not done or write_queue.qsize() > 0:
try:
buf = write_queue.get(timeout=0.01)
except QueueEmpty:
continue

f.write(buf)
f.write(buf)

except Exception as e:
exc_info = sys.exc_info()
except Exception as e:
exc_info = sys.exc_info()

writer_thread = threading.Thread(target=bg_write)
writer_thread.start()
writer_thread = threading.Thread(target=bg_write)
writer_thread.start()

try:
while True:
buf = stream.read(buffer_size)
if not buf:
break
try:
while True:
buf = stream.read(buffer_size)
if not buf:
break

write_queue.put_nowait(buf)
finally:
done = True
write_queue.put_nowait(buf)
finally:
done = True

writer_thread.join()
if exc_info is not None:
raise exc_info[0], exc_info[1], exc_info[2]
writer_thread.join()
if exc_info is not None:
raise exc_info[0], exc_info[1], exc_info[2]

def download(self, path, stream, buffer_size=None):
f = self.open(path, 'rb', buffer_size=buffer_size)
f.download(stream)
with self.open(path, 'rb', buffer_size=buffer_size) as f:
f.download(stream)


# ----------------------------------------------------------------------
# Specialization for HDFS

# ARROW-404: Helper class to ensure that files are closed before the
# client. During deallocation of the extension class, the attributes are
# decref'd which can cause the client to get closed first if the file has the
# last remaining reference
cdef class _HdfsFileNanny:
cdef:
object client
object file_handle

def __cinit__(self, client, file_handle):
self.client = client
self.file_handle = file_handle

def __dealloc__(self):
self.file_handle.close()
# try to avoid cyclic GC
self.file_handle = None
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could we get around the cyclic GC problem by having a weak reference here to the file handle? (Disclaimer: I have only used weak references in C++, not in Python nor Cython, so I have no idea if that would really work.)

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Perhaps, let me try

self.client = None


cdef class HdfsFile(NativeFile):
cdef readonly:
int32_t buffer_size
object mode
object parent

def __dealloc__(self):
self.parent = None

def read(self, int nbytes):
"""
Read indicated number of bytes from the file, up to EOF
Expand Down
23 changes: 23 additions & 0 deletions python/pyarrow/tests/test_hdfs.py
Original file line number Diff line number Diff line change
Expand Up @@ -98,6 +98,29 @@ def test_hdfs_ls(hdfs):
assert contents == [dir_path, f1_path]


def _make_test_file(hdfs, test_name, test_path, test_data):
base_path = pjoin(HDFS_TMP_PATH, test_name)
hdfs.mkdir(base_path)

full_path = pjoin(base_path, test_path)

f = hdfs.open(full_path, 'wb')
f.write(test_data)

return full_path


@libhdfs
def test_hdfs_orphaned_file():
hdfs = hdfs_test_client()
file_path = _make_test_file(hdfs, 'orphaned_file_test', 'fname',
'foobarbaz')

f = hdfs.open(file_path)
hdfs = None
f = None # noqa


@libhdfs
def test_hdfs_download_upload(hdfs):
base_path = pjoin(HDFS_TMP_PATH, 'upload-test')
Expand Down