client: fix cache invalidation during a load from a storage for the same oid

This fixes the following an invalidation bug:

ERROR ZODB.Connection Couldn't load state for 0x504e
Traceback (most recent call last):
  File "ZODB/Connection.py", line 851, in setstate
    self._setstate(obj)
  File "ZODB/Connection.py", line 916, in _setstate
    self._load_before_or_conflict(obj)
  File "ZODB/Connection.py", line 931, in _load_before_or_conflict
    if not self._setstate_noncurrent(obj):
  File "ZODB/Connection.py", line 954, in _setstate_noncurrent
    assert end is not None
AssertionError

neo/client/app.py

@@ -18,6 +18,7 @@ from cPickle import dumps, loads
 from zlib import compress as real_compress, decompress
 from neo.lib.locking import Empty
 from random import shuffle
+from thread import get_ident
 import heapq
 import time
 import os
@@ -97,6 +98,7 @@ class Application(object):
         # no self-assigned UUID, primary master will supply us one
         self.uuid = None
         self._cache = ClientCache()
+        self._loading = {}
         self.new_oid_list = []
         self.last_oid = '\0' * 8
         self.storage_event_handler = storage.StorageEventHandler(self)
@@ -408,17 +410,29 @@ class Application(object):
         # TODO:
         # - rename parameters (here? and in handlers & packet definitions)
 
+        acquire = self._cache_lock_acquire
+        release = self._cache_lock_release
         self._load_lock_acquire()
+        try:
+            acquire()
             try:
                 result = self._loadFromCache(oid, tid, before_tid)
-            if not result:
-                result = self._loadFromStorage(oid, tid, before_tid)
-                self._cache_lock_acquire()
+                if result:
+                    return result
+                loading_key = oid, get_ident()
+                self._loading[loading_key] = None
+                release()
                 try:
-                    self._cache.store(oid, *result)
+                    result = self._loadFromStorage(oid, tid, before_tid)
                 finally:
-                    self._cache_lock_release()
+                    acquire()
+                    invalidated = self._loading.pop(loading_key)
+                if invalidated and not result[2]:
+                    result = result[0], result[1], invalidated
+                self._cache.store(oid, *result)
                 return result
+            finally:
+                release()
         finally:
             self._load_lock_release()
 
@@ -450,15 +464,11 @@ class Application(object):
         """
         Load from local cache, return None if not found.
         """
-        self._cache_lock_acquire()
-        try:
         if at_tid:
             result = self._cache.load(oid, at_tid + '*')
             assert not result or result[1] == at_tid
             return result
         return self._cache.load(oid, before_tid)
-        finally:
-            self._cache_lock_release()
 
     @profiler_decorator
     def tpc_begin(self, transaction, tid=None, status=' '):
@@ -772,6 +782,7 @@ class Application(object):
             self._cache_lock_acquire()
             try:
                 cache = self._cache
+                loading = self._loading
                 for oid, data in cache_dict.iteritems():
                     if data is CHECKED_SERIAL:
                         # this is just a remain of
@@ -779,7 +790,12 @@ class Application(object):
                         # was modified).
                         continue
                     # Update ex-latest value in cache
+                    try:
                         cache.invalidate(oid, tid)
+                    except KeyError:
+                        for k in loading:
+                            if k[0] == oid and not loading[k]:
+                                loading[k] = tid
                     if data is not None:
                         # Store in cache with no next_tid
                         cache.store(oid, data, tid, None)

neo/client/cache.py

@@ -179,8 +179,11 @@ class ClientCache(object):
         if size < max_size:
             item = self._load(oid, next_tid)
             if item:
-                assert not (item.data or item.level)
                 assert item.tid == tid and item.next_tid == next_tid
+                if item.level: # already stored
+                    assert item.data == data
+                    return
+                assert not item.data
                 self._history_size -= 1
             else:
                 item = CacheItem()
@@ -221,12 +224,31 @@ class ClientCache(object):
 
     def invalidate(self, oid, tid):
         """Mark data record as being valid only up to given tid"""
-        try:
         item = self._oid_dict[oid][-1]
-        except KeyError:
-            pass
-        else:
         if item.next_tid is None:
             item.next_tid = tid
         else:
             assert item.next_tid <= tid, (item, oid, tid)
+
+
+def test(self):
+    cache = ClientCache()
+    self.assertEqual(cache.load(1, 10), None)
+    self.assertEqual(cache.load(1, None), None)
+    self.assertRaises(KeyError, cache.invalidate, 1, 10)
+    data = 'foo', 5, 10
+    # 2 identical stores happens if 2 threads got a cache miss at the same time
+    cache.store(1, *data)
+    cache.store(1, *data)
+    self.assertEqual(cache.load(1, 10), data)
+    self.assertEqual(cache.load(1, None), None)
+    data = 'bar', 10, None
+    cache.store(1, *data)
+    self.assertEqual(cache.load(1, None), data)
+    cache.invalidate(1, 20)
+    self.assertEqual(cache.load(1, 20), ('bar', 10, 20))
+
+if __name__ == '__main__':
+    import unittest
+    unittest.TextTestRunner().run(type('', (unittest.TestCase,), {
+        'runTest': test})())

neo/client/handlers/master.py

@@ -113,8 +113,14 @@ class PrimaryNotificationsHandler(BaseHandler):
         app._cache_lock_acquire()
         try:
             invalidate = app._cache.invalidate
+            loading = app._loading
             for oid in oid_list:
+                try:
                     invalidate(oid, tid)
+                except KeyError:
+                    for k in loading:
+                        if k[0] == oid and not loading[k]:
+                            loading[k] = tid
             db = app.getDB()
             if db is not None:
                 db.invalidate(tid, oid_list)

neo/tests/client/testClientApp.py

@@ -20,6 +20,7 @@ from mock import Mock, ReturnValues
 from ZODB.POSException import StorageTransactionError, UndoError, ConflictError
 from .. import NeoUnitTestBase, buildUrlFromString, ADDRESS_TYPE
 from neo.client.app import Application
+from neo.client.cache import test as testCache
 from neo.client.exception import NEOStorageError, NEOStorageNotFoundError
 from neo.client.exception import NEOStorageDoesNotExistError
 from neo.lib.protocol import NodeTypes, Packet, Packets, Errors, INVALID_TID, \
@@ -160,6 +161,8 @@ class ClientApplicationTests(NeoUnitTestBase):
         #self.assertEqual(calls[0].getParam(0), conn)
         #self.assertTrue(isinstance(calls[0].getParam(2), Queue))
 
+    testCache = testCache
+
     def test_registerDB(self):
         app = self.getApp()
         dummy_db = []

neo/tests/threaded/test.py

@@ -511,26 +511,52 @@ class Test(NEOThreadedTest):
                 l1.release()
                 l2.acquire()
             orig(conn, packet, kw, handler)
+        def _loadFromStorage(orig, *args):
+            try:
+                return orig(*args)
+            finally:
+                l1.release()
+                l2.acquire()
         cluster = NEOCluster()
         try:
             cluster.start()
             t1, c1 = cluster.getTransaction()
-            c1.root()['x'] = x = PCounter()
+            c1.root()['x'] = x1 = PCounter()
             t1.commit()
             t1.begin()
-            x.value = 1
+            x1.value = 1
             t2, c2 = cluster.getTransaction()
-            x = c2.root()['x']
+            x2 = c2.root()['x']
             p = Patch(cluster.client, _handlePacket=_handlePacket)
             try:
                 t = self.newThread(t1.commit)
                 l1.acquire()
                 t2.abort()
+            finally:
+                del p
                 l2.release()
             t.join()
+            self.assertEqual(x2.value, 1)
+
+            return # Following is disabled due to deadlock
+                   # caused by client load lock
+            t1.begin()
+            x1.value = 0
+            x2._p_deactivate()
+            cluster.client._cache.clear()
+            p = Patch(cluster.client, _loadFromStorage=_loadFromStorage)
+            try:
+                t = self.newThread(x2._p_activate)
+                l1.acquire()
+                t1.commit()
+                t1.begin()
             finally:
                 del p
-            self.assertEqual(x.value, 1)
+                l2.release()
+            t.join()
+            x1._p_deactivate()
+            self.assertEqual(x2.value, 1)
+            self.assertEqual(x1.value, 0)
         finally:
             cluster.stop()
 
@@ -540,12 +566,13 @@ class Test(NEOThreadedTest):
             cluster.start()
             # Initialize objects
             t1, c1 = cluster.getTransaction()
-            c1.root()['x'] = x = PCounter()
+            c1.root()['x'] = x1 = PCounter()
             c1.root()['y'] = y = PCounter()
             y.value = 1
             t1.commit()
             # Get pickle of y
             t1.begin()
+            x = c1._storage.load(x1._p_oid)[0]
             y = c1._storage.load(y._p_oid)[0]
             # Start the testing transaction
             # (at this time, we still have x=0 and y=1)
@@ -557,30 +584,67 @@ class Test(NEOThreadedTest):
             client.setPoll(1)
             txn = transaction.Transaction()
             client.tpc_begin(txn)
-            client.store(x._p_oid, x._p_serial, y, '', txn)
+            client.store(x1._p_oid, x1._p_serial, y, '', txn)
             # Delay invalidation for x
             master_client = cluster.master.filterConnection(cluster.client)
             try:
                 master_client.add(lambda conn, packet:
                     isinstance(packet, Packets.InvalidateObjects))
-                client.tpc_finish(txn, None)
-                client.close()
+                tid = client.tpc_finish(txn, None)
                 client.setPoll(0)
                 cluster.client.setPoll(1)
                 # Change to x is committed. Testing connection must ask the
                 # storage node to return original value of x, even if we
                 # haven't processed yet any invalidation for x.
-                x = c2.root()['x']
+                x2 = c2.root()['x']
                 cluster.client._cache.clear() # bypass cache
-                self.assertEqual(x.value, 0)
+                self.assertEqual(x2.value, 0)
             finally:
                 master_client()
-            x._p_deactivate()
+            x2._p_deactivate()
             t1.abort() # process invalidation and sync connection storage
-            self.assertEqual(x.value, 0)
+            self.assertEqual(x2.value, 0)
             # New testing transaction. Now we can see the last value of x.
             t2.abort()
-            self.assertEqual(x.value, 1)
+            self.assertEqual(x2.value, 1)
+
+            # Now test cache invalidation during a load from a storage
+            l1 = threading.Lock(); l1.acquire()
+            l2 = threading.Lock(); l2.acquire()
+            def _loadFromStorage(orig, *args):
+                try:
+                    return orig(*args)
+                finally:
+                    l1.release()
+                    l2.acquire()
+            x2._p_deactivate()
+            cluster.client._cache.clear()
+            p = Patch(cluster.client, _loadFromStorage=_loadFromStorage)
+            try:
+                t = self.newThread(x2._p_activate)
+                l1.acquire()
+                # At this point, x could not be found the cache and the result
+                # from the storage (which is <value=1, next_tid=None>) is about
+                # to processed.
+                # Now modify x to receive an invalidation for it.
+                cluster.client.setPoll(0)
+                client.setPoll(1)
+                txn = transaction.Transaction()
+                client.tpc_begin(txn)
+                client.store(x2._p_oid, tid, x, '', txn)
+                tid = client.tpc_finish(txn, None)
+                client.close()
+                client.setPoll(0)
+                cluster.client.setPoll(1)
+                t1.abort() # make sure invalidation is processed
+            finally:
+                del p
+                # Resume processing of answer from storage. An entry should be
+                # added in cache for x=1 with a fixed next_tid (i.e. not None)
+                l2.release()
+            t.join()
+            self.assertEqual(x2.value, 1)
+            self.assertEqual(x1.value, 0)
         finally:
             cluster.stop()