Deprecation notice

Notify potential users of this library that it will stop working very soon

Use threads instead of processes in Dataset.summaries

See #242

@tdsmith saw performance issues when running on Databricks due to process contention. Switching to threads seems to be an overall performance improvement.

Regenerate protobuf class

To keep up with evolution of protocol buffers, we've regenerated the class with latest protoc (3.6.1) and not require protbuf>=3.6.0

Increase Pandas Version

Increase pandas version (#225)

Spark 2.3 requires Pandas 0.19.2 for some operations.
See bug 1484715

Release as a universal wheel

Previous releases only provided a wheel release specific to either python 2 or python 3. Starting with this version, we release a universal wheel that leads to more efficient installation for consuming projects.

See #223

Add max_concurrency param and new grouping to Dataset

Dataset.records now has a group_by method that can be set to 'equal_size' rather than the default 'greedy' (#217)

The Dataset constructor now accepts a max_concurrency parameter that can be used to manage the number of spawned processes (#221)

Python 3 support

Add support for python3 (#208)

and make python3 default in docker

Explicitly deprecate some legacy methods (and document them as such)

Bug 1463885 - Disrecommend get_one_ping_per_client (#212)

* Update docs to more clearly mark deprecated methods as... deprecated

* Bug 1463885 - Disrecommend get_one_ping_per_client even more

v0.9.0

Bug 1447851 - Add content decoding for landfill #209

This release adds support for decoding messages as seen from ingestion. These documents are lightly processed and stored as gzipped json documents with a meta and content field.

v0.8.14

Change deploy acct to frank (#203)