Release 0.103.0

• Fixed error handling for paged queries
  • Error translation was not done for these so errors flew through without
    conversion to client-specific ones

Release 0.102.0

• Fixed bad range_end computation logic
  • This was not ready for keys whose last byte is 0xFF
  • It only implemented the happy path which worked fine when the keys / prefixes
    were strings (0xFF is not a 'normal' character). There is solid change for
    breakage when keys provided by caller are binary

Release 0.101.0

• No functional changes
• Project sanity
  • Dropped support for Python 3.9 as it is now EOL
  • Added support for Python 3.14; moved to dev environment to 3.14
• Move to protobuf >=6.33.1, including gencode bump
  • Gencode used in buf.build locked on 6.33.1
  • Runtime code >= 6.33.1
  • Runtime code now allows bump to the future 7.x which supports protobufs generated by 6.x
• Move to grpcio >= 1.75.1
  • First version to support Python 3.14

Release 0.100.0

• Enhance WatchResponse with max_mod_revision property
  • This should be used for purposes of tracking revision high watermark; using WatchResponse's header
    revision is unsafe for this purpose.

Release 0.99.0

• Expanded status() call to include all available information
  • Also includes cluster_id and member_id from status response's header

Release 0.98.0

• No functional changes
• Move to protobuf 5.27.1
  • Gencode used in buf.build locked on 5.27.1
  • Runtime code >= 5.27.1
• Move to grpcio >= 1.59
  • First grpcio version to support Python 3.12
  • Gencode used in buf.build locked on 1.59
  • Runtime code >= 1.59

Release 0.97.0

• Enhance WatchResponse with method to count events by type
  • Done without materialization of event classes
  • Useful especially for clients which analyze possibly large incoming event batches and
    look for those that only describe DELETEs.

Release 0.96.0

• Enhance WatchResponse to allow lazy event materialization
  • Old implementation read and created instances of Event for all events in the response
  • With large watch responses, this can contribute to memory usage spikes
  • The response now allows iterating events via generator, accessing events by index and determining
    number of events without the eager materialization
• Refactored Event and KVMetadata classes to not eagerly access fields from etcd response

0.95.1

• Fix for paged gets: do not send any sorting parameters as the default is already key-ascending order

0.95.0

• Allow limit argument on get_* calls
  • The value specified in limit is passed to etcd server; limiting happens server side
• Add support for paged gets of range of keys and keys starting with a prefix
  • This can be useful for clients that do stream processing on large number of key-value pairs
  • Without paging, the response from etcd can be quite large and lead to memory spikes in the client
  • With paging, results are obtained in smaller pages, so these memory spikes are not so significant.
  • The paged results are still presented as a single, continuous stream of key-value pairs