diff --git a/.travis.yml b/.travis.yml
new file mode 100644
index 0000000..7d99673
--- /dev/null
+++ b/.travis.yml
@@ -0,0 +1,13 @@
+language: python
+python:
+  - "2.7"
+  - "3.5"
+  - "3.6"
+# command to install dependencies
+install:
+  - pip install twisted
+  - pip install boto3
+  - pip install https://github.com/matrix-org/synapse/tarball/master
+# command to run tests
+script:
+  - PYTHONPATH=. trial test_s3
diff --git a/s3_storage_provider.py b/s3_storage_provider.py
index 1ec4951..ac75514 100644
--- a/s3_storage_provider.py
+++ b/s3_storage_provider.py
@@ -34,6 +34,9 @@ logger = logging.getLogger("synapse.s3")
 # The list of valid AWS storage class names
 _VALID_STORAGE_CLASSES = ('STANDARD', 'REDUCED_REDUNDANCY', 'STANDARD_IA')
 
+# Chunk size to use when reading from s3 connection in bytes
+READ_CHUNK_SIZE = 16 * 1024
+
 
 class S3StorageProviderBackend(StorageProvider):
     """
@@ -97,14 +100,8 @@ class _S3DownloadThread(threading.Thread):
         deferred (Deferred[_S3Responder|None]): If file exists
             resolved with an _S3Responder instance, if it doesn't
             exist then resolves with None.
-
-    Attributes:
-        READ_CHUNK_SIZE (int): The chunk size in bytes used when downloading
-            file.
     """
 
-    READ_CHUNK_SIZE = 16 * 1024
-
     def __init__(self, bucket, key, deferred):
         super(_S3DownloadThread, self).__init__(name="s3-download")
         self.bucket = bucket
@@ -125,59 +122,73 @@ class _S3DownloadThread(threading.Thread):
             reactor.callFromThread(self.deferred.errback, Failure())
             return
 
-        # Triggered by responder when more data has been requested (or
-        # stop_event has been triggered)
-        wakeup_event = threading.Event()
-        # Trigered by responder when we should abort the download.
-        stop_event = threading.Event()
-
-        producer = _S3Responder(wakeup_event, stop_event)
+        producer = _S3Responder()
         reactor.callFromThread(self.deferred.callback, producer)
+        _stream_to_producer(reactor, producer, resp["Body"], timeout=90.)
 
-        try:
-            body = resp["Body"]
 
-            while not stop_event.is_set():
-                # We wait for the producer to signal that the consumer wants
-                # more data (or we should abort)
-                wakeup_event.wait()
+def _stream_to_producer(reactor, producer, body, status=None, timeout=None):
+    """Streams a file like object to the producer.
 
-                # Check if we were woken up so that we abort the download
-                if stop_event.is_set():
-                    return
+    Correctly handles producer being paused/resumed/stopped.
 
-                chunk = body.read(self.READ_CHUNK_SIZE)
-                if not chunk:
-                    return
+    Args:
+        reactor
+        producer (_S3Responder): Producer object to stream results to
+        body (file like): The object to read from
+        status (_ProducerStatus|None): Used to track whether we're currently
+            paused or not. Used for testing
+        timeout (float|None): Timeout in seconds to wait for consume to resume
+            after being paused
+    """
 
-                # We clear the wakeup_event flag just before we write the data
-                # to producer.
-                wakeup_event.clear()
-                reactor.callFromThread(producer._write, chunk)
+    # Set when we should be producing, cleared when we are paused
+    wakeup_event = producer.wakeup_event
 
-        except Exception:
-            reactor.callFromThread(producer._error, Failure())
-            return
-        finally:
-            reactor.callFromThread(producer._finish)
-            if body:
-                body.close()
+    # Set if we should stop producing forever
+    stop_event = producer.stop_event
+
+    if not status:
+        status = _ProducerStatus()
+
+    try:
+        while not stop_event.is_set():
+            # We wait for the producer to signal that the consumer wants
+            # more data (or we should abort)
+            if not wakeup_event.is_set():
+                status.set_paused(True)
+                ret = wakeup_event.wait(timeout)
+                if not ret:
+                    raise Exception("Timed out waiting to resume")
+                status.set_paused(False)
+
+            # Check if we were woken up so that we abort the download
+            if stop_event.is_set():
+                return
+
+            chunk = body.read(READ_CHUNK_SIZE)
+            if not chunk:
+                return
+
+            reactor.callFromThread(producer._write, chunk)
+
+    except Exception:
+        reactor.callFromThread(producer._error, Failure())
+    finally:
+        reactor.callFromThread(producer._finish)
+        if body:
+            body.close()
 
 
 class _S3Responder(Responder):
     """A Responder for S3. Created by _S3DownloadThread
-
-    Args:
-        wakeup_event (threading.Event): Used to signal to _S3DownloadThread
-            that consumer is ready for more data (or that we've triggered
-            stop_event).
-        stop_event (threading.Event): Used to signal to _S3DownloadThread that
-            it should stop producing. `wakeup_event` must also be set if
-            `stop_event` is used.
     """
-    def __init__(self, wakeup_event, stop_event):
-        self.wakeup_event = wakeup_event
-        self.stop_event = stop_event
+    def __init__(self):
+        # Triggered by responder when more data has been requested (or
+        # stop_event has been triggered)
+        self.wakeup_event = threading.Event()
+        # Trigered by responder when we should abort the download.
+        self.stop_event = threading.Event()
 
         # The consumer we're registered to
         self.consumer = None
@@ -190,9 +201,10 @@ class _S3Responder(Responder):
         """See Responder.write_to_consumer
         """
         self.consumer = consumer
-        # We are a IPullProducer, so we expect consumer to call resumeProducing
-        # each time they want a new chunk of data.
-        consumer.registerProducer(self, False)
+        # We are a IPushProducer, so we start producing immediately until we
+        # get a pauseProducing or stopProducing
+        consumer.registerProducer(self, True)
+        self.wakeup_event.set()
         return self.deferred
 
     def __exit__(self, exc_type, exc_val, exc_tb):
@@ -200,18 +212,24 @@ class _S3Responder(Responder):
         self.wakeup_event.set()
 
     def resumeProducing(self):
-        """See IPullProducer.resumeProducing
+        """See IPushProducer.resumeProducing
         """
         # The consumer is asking for more data, signal _S3DownloadThread
         self.wakeup_event.set()
 
+    def pauseProducing(self):
+        """See IPushProducer.stopProducing
+        """
+        self.wakeup_event.clear()
+
     def stopProducing(self):
-        """See IPullProducer.stopProducing
+        """See IPushProducer.stopProducing
         """
         # The consumer wants no more data ever, signal _S3DownloadThread
         self.stop_event.set()
         self.wakeup_event.set()
-        self.deferred.errback(Exception("Consumer ask to stop producing"))
+        if not self.deferred.called:
+            self.deferred.errback(Exception("Consumer ask to stop producing"))
 
     def _write(self, chunk):
         """Writes the chunk of data to consumer. Called by _S3DownloadThread.
@@ -239,3 +257,24 @@ class _S3Responder(Responder):
 
         if not self.deferred.called:
             self.deferred.callback(None)
+
+
+class _ProducerStatus(object):
+    """Used to track whether the s3 download thread is currently paused
+    waiting for consumer to resume. Used for testing.
+    """
+
+    def __init__(self):
+        self.is_paused = threading.Event()
+        self.is_paused.clear()
+
+    def wait_until_paused(self, timeout=None):
+        is_paused = self.is_paused.wait(timeout)
+        if not is_paused:
+            raise Exception("Timed out waiting")
+
+    def set_paused(self, paused):
+        if paused:
+            self.is_paused.set()
+        else:
+            self.is_paused.clear()
diff --git a/test_s3.py b/test_s3.py
new file mode 100644
index 0000000..379d381
--- /dev/null
+++ b/test_s3.py
@@ -0,0 +1,181 @@
+# -*- coding: utf-8 -*-
+# Copyright 2018 New Vector Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from twisted.internet import defer
+from twisted.python.failure import Failure
+from twisted.test.proto_helpers import MemoryReactorClock
+from twisted.trial import unittest
+
+from queue import Queue
+from threading import Event, Thread
+
+from mock import Mock
+
+from s3_storage_provider import (
+    _stream_to_producer, _S3Responder, _ProducerStatus,
+)
+
+
+class StreamingProducerTestCase(unittest.TestCase):
+    def setUp(self):
+        self.reactor = ThreadedMemoryReactorClock()
+
+        self.body = Channel()
+        self.consumer = Mock()
+        self.written = ""
+
+        def write(data):
+            self.written += data
+
+        self.consumer.write.side_effect = write
+
+        self.producer_status = _ProducerStatus()
+        self.producer = _S3Responder()
+        self.thread = Thread(
+            target=_stream_to_producer,
+            args=(self.reactor, self.producer, self.body),
+            kwargs={
+                "status": self.producer_status,
+                "timeout": 1.0,
+            },
+        )
+        self.thread.daemon = True
+        self.thread.start()
+
+    def tearDown(self):
+        # Really ensure that we've stopped the thread
+        self.producer.stopProducing()
+
+    def test_simple_produce(self):
+        deferred = self.producer.write_to_consumer(self.consumer)
+
+        self.body.write("test")
+        self.wait_for_thread()
+        self.assertEqual("test", self.written)
+
+        self.body.write(" string")
+        self.wait_for_thread()
+        self.assertEqual("test string", self.written)
+
+        self.body.finish()
+        self.wait_for_thread()
+
+        self.assertTrue(deferred.called)
+        self.assertEqual(deferred.result, None)
+
+    def test_pause_produce(self):
+        deferred = self.producer.write_to_consumer(self.consumer)
+
+        self.body.write("test")
+        self.wait_for_thread()
+        self.assertEqual("test", self.written)
+
+        # We pause producing, but the thread will currently be blocked waiting
+        # to read data, so we wake it up by writing before asserting that
+        # it actually pauses.
+        self.producer.pauseProducing()
+        self.body.write(" string")
+        self.wait_for_thread()
+        self.producer_status.wait_until_paused(10.)
+        self.assertEqual("test string", self.written)
+
+        # If we write again we remain paused and nothing gets written
+        self.body.write(" second")
+        self.producer_status.wait_until_paused(10.)
+        self.assertEqual("test string", self.written)
+
+        # If we call resumeProducing the buffered data gets read and written.
+        self.producer.resumeProducing()
+        self.wait_for_thread()
+        self.assertEqual("test string second", self.written)
+
+        # We can continue writing as normal now
+        self.body.write(" third")
+        self.wait_for_thread()
+        self.assertEqual("test string second third", self.written)
+
+        self.body.finish()
+        self.wait_for_thread()
+
+        self.assertTrue(deferred.called)
+        self.assertEqual(deferred.result, None)
+
+    def test_error(self):
+        deferred = self.producer.write_to_consumer(self.consumer)
+
+        self.body.write("test")
+        self.wait_for_thread()
+        self.assertEqual("test", self.written)
+
+        excp = Exception("Test Exception")
+        self.body.error(excp)
+        self.wait_for_thread()
+
+        self.assertTrue(deferred.called)
+        self.assertIsInstance(deferred.result, Failure)
+
+    def wait_for_thread(self):
+        """Wait for something to call `callFromThread` and advance reactor
+        """
+        self.reactor.thread_event.wait(1)
+        self.reactor.thread_event.clear()
+        self.reactor.advance(0)
+
+
+class ThreadedMemoryReactorClock(MemoryReactorClock):
+    """
+    A MemoryReactorClock that supports callFromThread.
+    """
+
+    def __init__(self):
+        super(ThreadedMemoryReactorClock, self).__init__()
+        self.thread_event = Event()
+
+    def callFromThread(self, callback, *args, **kwargs):
+        """
+        Make the callback fire in the next reactor iteration.
+        """
+        d = defer.Deferred()
+        d.addCallback(lambda x: callback(*args, **kwargs))
+        self.callLater(0, d.callback, True)
+
+        self.thread_event.set()
+
+        return d
+
+
+class Channel(object):
+    """Simple channel to mimic a thread safe file like object
+    """
+    def __init__(self):
+        self._queue = Queue()
+
+    def read(self, _):
+        val = self._queue.get()
+        if isinstance(val, Exception):
+            raise val
+        return val
+
+    def write(self, val):
+        self._queue.put(val)
+
+    def error(self, err):
+        self._queue.put(err)
+
+    def finish(self):
+        self._queue.put(None)
+
+    def close(self):
+        pass