Bug Fixed:

"Blob temp file get's removed before it should",
https://bugs.launchpad.net/zodb/+bug/595378

A way this to happen is that a transaction is aborted after the
commit process has started. I don't know how this would happen in
the wild.

In 3.10.3, the ZEO tpc_abort call to the server is changed to be
synchronous, which should address this case. Maybe there's another
case.

src/CHANGES.txt

@@ -13,6 +13,18 @@ Bugs Fixed
 
   https://bugs.launchpad.net/zodb/+bug/737198
 
+- "Blob temp file get's removed before it should",
+  https://bugs.launchpad.net/zodb/+bug/595378
+
+  A way this to happen is that a transaction is aborted after the
+  commit process has started. I don't know how this would happen in
+  the wild.
+
+  In 3.10.3, the ZEO tpc_abort call to the server is changed to be
+  synchronous, which should address this case. Maybe there's another
+  case.
+
+
 Performance enhancements
 ------------------------
 

src/ZEO/ServerStub.py

@@ -252,7 +252,7 @@ class StorageServer:
         return self.rpc.call('tpc_finish', id)
 
     def tpc_abort(self, id):
-        self.rpc.callAsync('tpc_abort', id)
+        self.rpc.call('tpc_abort', id)
 
     def history(self, oid, length=None):
         if length is None:

src/ZEO/tests/testZEO.py

@@ -1498,6 +1498,44 @@ def unix_domain_sockets():
     >>> c.root.x = 1
     >>> transaction.commit()
     >>> c.close()
+    """
+
+def gracefully_handle_abort_while_storing_many_blobs():
+    r"""
+
+    >>> import logging, sys
+    >>> old_level = logging.getLogger().getEffectiveLevel()
+    >>> logging.getLogger().setLevel(logging.ERROR)
+    >>> handler = logging.StreamHandler(sys.stdout)
+    >>> logging.getLogger().addHandler(handler)
+
+    >>> addr, _ = start_server(port='./sock', blob_dir='blobs')
+    >>> c = ZEO.connection(addr, blob_dir='cblobs')
+    >>> c.root.x = ZODB.blob.Blob('z'*(1<<20))
+    >>> c.root.y = ZODB.blob.Blob('z'*(1<<2))
+    >>> t = c.transaction_manager.get()
+    >>> c.tpc_begin(t)
+    >>> c.commit(t)
+
+We've called commit, but the blob sends are queued.  We'll call abort
+right away, which will delete the temporary blob files.  The queued
+iterators will try to open these files.
+
+    >>> c.tpc_abort(t)
+
+Now we'll try to use the connection, mainly to wait for everything to
+get processed. Before we fixed this by making tpc_finish a synchronous
+call to the server. we'd get some sort of error here.
+
+    >>> _ = c._storage._server.loadEx('\0'*8)
+
+    >>> c.close()
+
+    >>> logging.getLogger().removeHandler(handler)
+    >>> logging.getLogger().setLevel(old_level)
+
+
+
     """
 
 if sys.platform.startswith('win'):