What's new on ZODB 3.8.0 ======================== General ------- - The ZODB Storage APIs have been documented and cleaned up. - ZODB versions are now officially deprecated and support for them will be removed in ZODB 3.9. (They have been widely recognized as deprecated for quite a while.) - Changed the automatic garbage collection when opening a connection to only apply the garbage collections on those connections in the pool that are closed. (This fixed issue 113932.) - (3.8.0b3) Document conflict resolution (see ZODB/ConflictResolution.txt). - (3.8.0b3) Bugfix the situation in which comparing persistent objects (for instance, as members in BTree set or keys of BTree) might cause data inconsistency during conflict resolution. - (3.8.0b3) Support multidatabase references in conflict resolution. - (3.8.0b3) Make it possible to examine oid and (in some situations) database name of persistent object references during conflict resolution. - (3.8.0b3) Added missing data attribute for conflict errors. - (3.8.0b5) Fixed bug 153316: persistent and BTrees gave errors on x86_64 Intel XEON platforms. ZEO --- - (???) Fixed bug in transaction buffer: a tuple was unpacked incorrectly in `clear`. - (???) Fixed bug in blob filesystem helper: the `isSecure` check was inversed. - (3.8.0b6) Bug #98275: Made ZEO cache more tolerant when invalidating current versions of objects. - (3.8.0b4, 3.8.0b5) Fixed a serious bug that could cause client I/O to stop (hang). This was accomonied by a critical log message along the lines of: "RuntimeError: dictionary changed size during iteration". (In b4, the bug was only partially fixed.) - (3.8a1) ZEO's strategoes for avoiding client cache verification were improved in the case that servers are restarted. Before, if transactions were committed after the restart, clients that were up to date or nearly up to date at the time of the restart and then connected had to verify their caches. Now, it is far more likely that a client that reconnects soon after a server restart won't have to verify its cache. - (3.8a1) Fixed a serious bug that could cause clients that disconnect from and reconnect to a server to get bad invalidation data if the server serves multiple storages with active writes. - (3.8a1) It is now theoretically possible to use a ClientStorage in a storage server. This might make it possible to offload read load from a storage server at the cost of increasing write latency. This should increase write throughput by offloading reads from the final storage server. This feature is somewhat experimental. It has tests, but hasn't been used in production. Transactions ------------ - (3.8a1) Add a doom() and isDoomed() interface to the transaction module. First step towards the resolution of http://www.zope.org/Collectors/Zope3-dev/655 A doomed transaction behaves exactly the same way as an active transaction but raises an error on any attempt to commit it, thus forcing an abort. Doom is useful in places where abort is unsafe and an exception cannot be raised. This occurs when the programmer wants the code following the doom to run but not commit. It is unsafe to abort in these circumstances as a following get() may implicitly open a new transaction. Any attempt to commit a doomed transaction will raise a DoomedTransaction exception. - (3.8a1) Clean up the ZODB imports in transaction. Clean up weird import dance with ZODB. This is unnecessary since the transaction module stopped being imported in ZODB/__init__.py in rev 39622. - (3.8a1) Support for subtransactions has been removed in favor of save points. Blobs ----- - (???) Fixed bug in Connection.TmpStore: load() would not defer to the backend storage for loading blobs. - (3.8b5) Fixed bug #130459: Packing was broken by uncommitted blob data. - (3.8b4) Fixed bug #127182: Blobs were subclassable which was not desired. - (3.8b3) Fixed bug #126007: tpc_abort had untested code path that was broken. - (3.8b3) Fixed bug #129921: getSize() function in BlobStorage could not deal with garbage files - (3.8b1) Updated the Blob implementation in a number of ways. Some of these are backward incompatible with 3.8a1: o The Blob class now lives in ZODB.blob o The blob openDetached method has been replaced by the committed method. - (3.8a1) Added new blob feature. See the ZODB/Blobs directory for documentation. ZODB now handles (reasonably) large binary objects efficiently. Useful to use from a few kilobytes to at least multiple hundred megabytes. BTrees ------ - (3.8a1) Added support for 64-bit integer BTrees as separate types. (For now, we're retaining compile-time support for making the regular integer BTrees 64-bit.) - (3.8a1) Normalize names in modules so that BTrees, Buckets, Sets, and TreeSets can all be accessed with those names in the modules (e.g., BTrees.IOBTree.BTree). This is in addition to the older names (e.g., BTrees.IOBTree.IOBTree). This allows easier drop-in replacement, which can especially be simplify code for packages that want to support both 32-bit and 64-bit BTrees. - (3.8a1) Describe the interfaces for each module and actually declare the interfaces for each. - (3.8a1) Fix module references so klass.__module__ points to the Python wrapper module, not the C extension. - (3.8a1) introduce module families, to group all 32-bit and all 64-bit modules.