Title.

gRPC is known to have issues with server-side load balancing. (See Best Practices#load-balancing)

Since we can't control the type of load balancer and client-side load balancing is a much larger task that wouldn't even apply to most current use-cases of the Cantor gRPC client this issue aims to add logic to the client that periodically refreshes the gRPC channel to ensure actual load balancing.

The unit tests for Cantor implementation on top of MySQL requires an instance of MySQL to run. Change the bang.sh script to start/stop one in a docker container.

Currently Cantor creates a single thread to flush the buffer directories for every S3 bucket (See https://github.com/salesforce/cantor/blob/master/cantor-s3/src/main/java/com/salesforce/cantor/s3/EventsOnS3.java#L64). This doesn't scale well as any other EventsOnS3 initialized pointing to different buckets will block waiting to flush. We need a way to maintain a thread per bucket, so different instances aren't blocking flush cycles.

Instead of milliseconds, take in hours of TTL.

Instead of initializing internal database in the constructor, do that lazily so we don't throw exceptions in the constructor.

Implement utility functions for each data structure to allowing users to export/import namespaces.

Add maven plugin for code style check.

Write wrappers similar to the loggable wrappers, that record metrics for each method in cantor. Should record metrics such as frequency, duration, and potentially some characteristics of the response.

Currently we can contact s3 by prefixing s3- to the front of any namespace. Ideally we would have something a little more obvious like s3://, but our servlet was inhibiting that ability. Since S3 datacenter specific we don't actually need to support the servlet, so axing prod support could be a simple approach

To be able to support MySQL (not only Mariadb), replace SQL statements that contain INTERSECT with INNER JOINs.

Once this is complete, users should be able to run CantorOnMysql on top of MySQL as well as Mariadb.

Create a samples directory and include sample code for using various features of Cantor.

Nodes start as a stand alone JVM, connect to the master and report the namespaces they support. Requests to the master are routed to nodes based on the namespace lookup table maintained in the master.

Add maven plugin for code coverage.

We should add the capability to only archive events that match metadata/dimension queries. Add these two as parameters, and pass them through to the call to get events to archive.