Add minimal implementation of ingesting Parquet and CSV files - #327

Thanks @voonhous

I have two comments

1. Can you please submit all of the formatting and other changes separately? It makes it incredibly difficult to review your code if you submit 700+ lines of code difference.
2. Is there a good reason you wrap ingest with ingest_file? It seems like ingest has steps that both it and ingest_file should share, like validating the dataframe, getting the feature set, and inferring/applying the schema from the dataframe. These should not be rerun for every batch.

I think it would make more sense if you extended the existing ingest so that it took either a dataframe, path, or file-like object. You would then handle it just like in ingest_file, where you use different methods to read the source based on what the user provides, but ultimately it results in a table.

At the infer_schema step, a single batch can be materialized into a DF for creating the schema. Then the rest of the table is passed into ingest_kafka where all iteration can happen.

What do you think?

Also, would you mind updating the title of the PR to something more descriptive? We use the PRs as items in our change logs.

/retest

/test test-core-and-ingestion
Trying to figure out if the integration test in feast-ingestion is flaky i.e. the test can fail with the same test configuration, specifically this test feast.ingestion.ImportJobTest

I have created this code snippet which creates a testing dataframe. It has all the types besides lists of boolean. I think we should ensure that we can ingest this as both a parquet and a pandas dataframe.

https://gist.github.com/woop/d074ded542bc2b6ec5a0b5a96c72e9ab

/retest

/lgtm
/approve

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: voonhous, woop

The full list of commands accepted by this bot can be found here.

The pull request process is described here