Protocol

• _since and _before now accepts an integer value between quotes ",
  as it would be returned in the ETag response header.
• A batch request now fails if one of the subrequests fails (#510)
  (see new feature about transactions)

Breaking changes

• For PostgreSQL backends, it is recommended to specify postgresql://.

New features

• A transaction now covers the whole request/response cycle (#510, Kinto/kinto#194).
  If an error occurs during the request processing, every operation performed
  is rolled back. Note: This is only enabled with PostgreSQL backends. In
  other words, the rollback has no effect on backends like Redis or Memory.
• Add the protocol_version to tell which protocol version is
  implemented by the service in the hello page. (#324)
• New settings for backends when using PostgreSQL: *_pool_maxoverflow,
  *_pool_recycle, *_pool_timeout to control connections pool
  behaviour.
• Add custom pool supporting a max_backlog parameter that limits the
  number of threads waiting for a connection (#509)
• Add impacted_records attribute on ResourceChanged event (#501)
  This also allows listeners to react on particular field change, since old and
  new version of records is provided.

Bug fixes

• Fix Service CORS not being set when plugins are included
• Fix crash with Redis backend if record parent/id is unicode (fixes #556)
• Fix principals of permission backend not being plugged by default (#573)
• Fix Redis error traces not being logged (#560)
• Fix principals of permission backend not being plugged by default. (#573)
• Maintain pagination offset to prevent pagination loop in some cases. (#366)

Internal changes

• Switch to SQLAlchemy for smarter connections pools.
• Added a simple end-to-end test on a Cliquet sample application, using
  Loads <http://github.com/loads/>_. (fixes #512)
• Switched to SQLAlchemy sessions instead of raw connections and cursors. (#510)
• Refactor Redis clients instantiation to avoid repeated defaults. (#567, #568)
• Initialize Service class attributes before including plugins. (#578)
• Add a statsd_count helper function to ease the usage of statsd. (#574)
• Mention SQLAlchemy on missing PostgreSQL dependencies. (#545)