BUG: GH11786 Thread safety issue with read_csv

doc/source/whatsnew/v0.18.0.txt

@@ -480,8 +480,7 @@ Bug Fixes
 - Bug in value label reading for ``StataReader`` when reading incrementally (:issue:`12014`)
 - Bug in vectorized ``DateOffset`` when ``n`` parameter is ``0`` (:issue:`11370`)
 - Compat for numpy 1.11 w.r.t. ``NaT`` comparison changes (:issue:`12049`)
-
-
+- Bug in ``read_csv`` when reading from a ``StringIO`` in threads (:issue:`11790`)
 
 
 

pandas/io/tests/test_parsers.py

@@ -9,6 +9,8 @@
 import nose
 import platform
 
+from multiprocessing.pool import ThreadPool
+
 from numpy import nan
 import numpy as np
 from pandas.io.common import DtypeWarning
@@ -4128,6 +4130,87 @@ def test_bool_header_arg(self):
             with tm.assertRaises(TypeError):
                 pd.read_fwf(StringIO(data), header=arg)
 
+    def test_multithread_stringio_read_csv(self):
+        # GH 11786
+        max_row_range = 10000
+        num_files = 100
+
+        bytes_to_df = [
+            '\n'.join(
+                ['%d,%d,%d' % (i, i, i) for i in range(max_row_range)]
+            ).encode() for j in range(num_files)]
+        files = [BytesIO(b) for b in bytes_to_df]
+
+        # Read all files in many threads
+        pool = ThreadPool(8)
+        results = pool.map(pd.read_csv, files)
+        first_result = results[0]
+
+        for result in results:
+            tm.assert_frame_equal(first_result, result)
+
+    def construct_dataframe(self, num_rows):
+
+        df = DataFrame(np.random.rand(num_rows, 5), columns=list('abcde'))
+        df['foo'] = 'foo'
+        df['bar'] = 'bar'
+        df['baz'] = 'baz'
+        df['date'] = pd.date_range('20000101 09:00:00',
+                                   periods=num_rows,
+                                   freq='s')
+        df['int'] = np.arange(num_rows)
+        return df
+
+    def generate_multithread_dataframe(self, path, num_rows, num_tasks):
+
+        def reader(arg):
+            start, nrows = arg
+
+            if not start:
+                return pd.read_csv(path, index_col=0, header=0, nrows=nrows,
+                                   parse_dates=['date'])
+
+            return pd.read_csv(path,
+                               index_col=0,
+                               header=None,
+                               skiprows=int(start) + 1,
+                               nrows=nrows,
+                               parse_dates=[9])
+
+        tasks = [
+            (num_rows * i / num_tasks,
+             num_rows / num_tasks) for i in range(num_tasks)
+        ]
+
+        pool = ThreadPool(processes=num_tasks)
+
+        results = pool.map(reader, tasks)
+
+        header = results[0].columns
+        for r in results[1:]:
+            r.columns = header
+
+        final_dataframe = pd.concat(results)
+
+        return final_dataframe
+
+    def test_multithread_path_multipart_read_csv(self):
+        # GH 11786
+        num_tasks = 4
+        file_name = '__threadpool_reader__.csv'
+        num_rows = 100000
+
+        df = self.construct_dataframe(num_rows)
+
+        with tm.ensure_clean(file_name) as path:
+            df.to_csv(path)
+
+            final_dataframe = self.generate_multithread_dataframe(path,
+                                                                  num_rows,
+                                                                  num_tasks)
+            tm.assert_frame_equal(df, final_dataframe)
+
+
 class TestMiscellaneous(tm.TestCase):
 
     # for tests that don't fit into any of the other classes, e.g. those that

pandas/src/parser/io.c

@@ -117,12 +117,12 @@ void* buffer_rd_bytes(void *source, size_t nbytes,
 
     size_t length;
     rd_source *src = RDS(source);
+    state = PyGILState_Ensure();
 
     /* delete old object */
     Py_XDECREF(src->buffer);
     args = Py_BuildValue("(i)", nbytes);
 
-    state = PyGILState_Ensure();
     func = PyObject_GetAttrString(src->obj, "read");
     /* printf("%s\n", PyBytes_AsString(PyObject_Repr(func))); */