Allow batch_size=None in Table.scan() to avoid filter incompatibilities

Allow None as a valid value for the batch_size argument to Table.scan(),
since HBase does not support specifying a batch size when some scanner
filters are used.

Fixes issue #54.

3 files changed, 33 insertions, 12 deletions

diff --git a/NEWS.rst b/NEWS.rst
index caac0a3..6fd875c 100644
--- a/NEWS.rst
+++ b/NEWS.rst
@@ -14,6 +14,11 @@ Release date: *not yet released*
   to :py:meth:`Table.scan` (`issue #39
   <https://github.com/wbolster/happybase/issues/39>`_).
 
+* Allow `None` as a valid value for the `batch_size` argument to
+  :py:meth:`Table.scan`, since HBase does not support specifying a batch size
+  when some scanner filters are used. (`issue #54
+  <https://github.com/wbolster/happybase/issues/54>`_).
+
 
 HappyBase 0.7
 -------------
diff --git a/happybase/table.py b/happybase/table.py
index ec67092..c81f4ce 100644
--- a/happybase/table.py
+++ b/happybase/table.py
@@ -254,6 +254,15 @@ class Table(object):
         this to a low value (or even 1) if your data is large, since a low
         batch size results in added round-trips to the server.
 
+        .. warning::
+
+           Not all HBase filters can be used in combination with a batch
+           size. Explicitly specify `None` for the `batch_size` argument
+           in those cases to override the default value. Failure to do
+           so can result in hard to debug errors (not HappyBase's
+           fault), such as a non-responsive connection. The HBase logs
+           may contain more useful information in these situations.
+
         **Compatibility notes:**
 
         * The `filter` argument is only available when using HBase 0.92
@@ -280,11 +289,11 @@ class Table(object):
         :return: generator yielding the rows matching the scan
         :rtype: iterable of `(row_key, row_data)` tuples
         """
-        if batch_size < 1:
-            raise ValueError("'batch_size' must be >= 1")
+        if batch_size is not None and batch_size < 1:
+            raise ValueError("'batch_size' must be >= 1 (or None)")
 
         if limit is not None and limit < 1:
-            raise ValueError("'limit' must be >= 1")
+            raise ValueError("'limit' must be >= 1 (or None)")
 
         if sorted_columns and self.connection.compat < '0.96':
             raise NotImplementedError(
@@ -349,16 +358,16 @@ class Table(object):
         n_returned = n_fetched = 0
         try:
             while True:
-                if limit is None:
-                    how_many = batch_size
+                if batch_size is None:
+                    how_many = 1
                 else:
-                    how_many = min(batch_size, limit - n_returned)
+                    how_many = batch_size
 
-                if how_many == 1:
-                    items = self.connection.client.scannerGet(scan_id)
-                else:
-                    items = self.connection.client.scannerGetList(
-                        scan_id, how_many)
+                if limit is not None:
+                    how_many = min(how_many, limit - n_returned)
+
+                items = self.connection.client.scannerGetList(
+                    scan_id, how_many)
 
                 n_fetched += len(items)
 
diff --git a/tests/test_api.py b/tests/test_api.py
index 2d22717..ed78f81 100644
--- a/tests/test_api.py
+++ b/tests/test_api.py
@@ -357,7 +357,7 @@ def test_scan():
         list(table.scan(row_prefix='foobar', row_start='xyz'))
 
     with assert_raises(ValueError):
-        list(table.scan(batch_size=None))
+        list(table.scan(batch_size=0))
 
     if connection.compat == '0.90':
         with assert_raises(NotImplementedError):
@@ -446,6 +446,13 @@ def test_scan_sorting():
         row.items())
 
 
+def test_scan_filter_and_batch_size():
+    # See issue #54
+    filter = "SingleColumnValueFilter ('cf1', 'qual1', =, 'binary:val1')"
+    for k, v in table.scan(filter=filter, batch_size=None):
+        print v
+
+
 def test_delete():
     row_key = 'row-test-delete'
     data = {'cf1:col1': 'v1',