feat: endpoint plugin

[aaronchung-bitquill]

Summary

Aurora Limitless Database is a sharded database. It has a DB Shard Group that contains multiple transaction routers that clients are ultimately routed to interact with.

This PR introduces a new plugin that put simply, does client side load balancing whilst connecting to an Aurora Limitless Database.

The plugin does this by polling an API for available transaction routers and their load levels, caching this info, and then choosing which transaction router to connect to based on the load info.

Description

The overall architecture is very similar to the EFM plugin. To poll for available transaction routers and their load levels, there is a LimitlessRouterMonitor class and its accompanying monitor service class LimitlessRouterService class.

To choose hosts, there are modifications to the RoundRobinHostSelector strategy, and a new HighestWeightHostSelector strategy class.

The LimitlessConnectionPlugin uses the RoundRobinHostSelector strategy with weights to select which Transaction Router to connect to. The weights are essentially the inverse of the transaction routers load. So a more loaded transaction router will be selected less, and a less loaded transaction router will be selected more. Modifications to the RoundRobinHostSelector strategy are to allow cached weights to be cleared when there is a change in hosts or weights.

In the case the selected transaction router is down and the driver fails to connect to it, the plugin has connection retry logic. It will mark the last selected router as UNAVAILABLE, then select another transaction router to connect to based on load using the new HighestWeightHostSelector strategy. It will continue to select and try to connect to the highest weight available router until a successful connection is made, or until the max attempts are reached.

Additional Reviewers

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.