Switch to shared Lock (SerializableLock if possible) for reading and …

…writing

Fixes pydata#1172

The serializable lock will be useful for dask.distributed or multi-processing
(xref pydata#798, pydata#1173, among others).

doc/whats-new.rst

@@ -56,7 +56,7 @@ Breaking changes
   By `Guido Imperiale <https://github.com/crusaderky>`_ and
   `Stephan Hoyer <https://github.com/shoyer>`_.
 - Pickling a ``Dataset`` or ``DataArray`` linked to a file on disk no longer
-  caches its values into memory before pickling :issue:`1128`. Instead, pickle
+  caches its values into memory before pickling (:issue:`1128`). Instead, pickle
   stores file paths and restores objects by reopening file references. This
   enables preliminary, experimental use of xarray for opening files with
   `dask.distributed <https://distributed.readthedocs.io>`_.
@@ -206,6 +206,10 @@ Bug fixes
 - Fixed a bug with facetgrid (the ``norm`` keyword was ignored, :issue:`1159`).
   By `Fabien Maussion <https://github.com/fmaussion>`_.
 
+- Resolved a concurrency bug that could cause Python to crash when
+  simultaneously reading and writing netCDF4 files with dask (:issue:`1172`).
+  By `Stephan Hoyer <https://github.com/shoyer>`_.
+
 .. _whats-new.0.8.2:
 
 v0.8.2 (18 August 2016)

xarray/backends/api.py

@@ -3,7 +3,6 @@
 from __future__ import print_function
 import gzip
 import os.path
-import threading
 from distutils.version import StrictVersion
 from glob import glob
 from io import BytesIO
@@ -12,7 +11,7 @@
 import numpy as np
 
 from .. import backends, conventions
-from .common import ArrayWriter
+from .common import ArrayWriter, GLOBAL_LOCK
 from ..core import indexing
 from ..core.combine import auto_combine
 from ..core.utils import close_on_error, is_remote_uri
@@ -55,9 +54,6 @@ def _normalize_path(path):
         return os.path.abspath(os.path.expanduser(path))
 
 
-_global_lock = threading.Lock()
-
-
 def _default_lock(filename, engine):
     if filename.endswith('.gz'):
         lock = False
@@ -71,9 +67,9 @@ def _default_lock(filename, engine):
             else:
                 # TODO: identify netcdf3 files and don't use the global lock
                 # for them
-                lock = _global_lock
+                lock = GLOBAL_LOCK
         elif engine in {'h5netcdf', 'pynio'}:
-            lock = _global_lock
+            lock = GLOBAL_LOCK
         else:
             lock = False
     return lock

xarray/backends/common.py

@@ -2,7 +2,6 @@
 from __future__ import division
 from __future__ import print_function
 import numpy as np
-import itertools
 import logging
 import time
 import traceback
@@ -12,7 +11,12 @@
 
 from ..conventions import cf_encoder
 from ..core.utils import FrozenOrderedDict
-from ..core.pycompat import iteritems, dask_array_type, OrderedDict
+from ..core.pycompat import iteritems, dask_array_type
+
+try:
+    from dask.utils import SerializableLock as Lock
+except ImportError:
+    from threading import Lock
 
 # Create a logger object, but don't add any handlers. Leave that to user code.
 logger = logging.getLogger(__name__)
@@ -21,6 +25,10 @@
 NONE_VAR_NAME = '__values__'
 
 
+# dask.utils.SerializableLock if available, otherwise just a threading.Lock
+GLOBAL_LOCK = Lock()
+
+
 def _encode_variable_name(name):
     if name is None:
         name = NONE_VAR_NAME
@@ -150,7 +158,7 @@ def sync(self):
             import dask.array as da
             import dask
             if StrictVersion(dask.__version__) > StrictVersion('0.8.1'):
-                da.store(self.sources, self.targets, lock=threading.Lock())
+                da.store(self.sources, self.targets, lock=GLOBAL_LOCK)
             else:
                 da.store(self.sources, self.targets)
             self.sources = []