split_client

# Changelog

## 0.4.0

* Adds ability to pass `cache: false` to `get_treatment` and `get_treatments` calls.
Can be helpful for admin tools that change a split's definition and want to see the new
evaluation immediately

## 0.3.0

* Changes the testing strategy to have evaluations default to a yaml file if the
test or process has not been setup with its own data source. This makes it easier
for tests to opt-in to enabling flags, only for the tests where it matters. This way
the rest of a test suite won't have to be concerned with split. It also makes it easier
for global processes that are hard/impossible to isolate can have some split evaluation
without throwing an error. The tradeoff here is that there won't be a loud error to let
engineers know when a process hasn't been configured with an isolated data_source.

## 0.2.1

* Fixes a bug where an empty yaml file would break evaluations

## 0.2.0

* Swaps out the custom cache for Cachex because of timeouts happening in `SplitClient.Boundary.Treatments` calls. The cache had a complicated api key consisting of key, split_name, and any attributes. This was likely overkill, made the cache too big, and made lookups too expensive which likely caused the timeouts. After 10,000 or so `get_treatment` calls with unique keys the server would timeout. Now we can run 100,000 `get_treatments` calls with with unique keys and see no timeouts.
* There's no more polling the split_evaluator, instead developers can configure a `:cache_ttl`