Skip to content

Latest commit

 

History

History
48 lines (35 loc) · 2.45 KB

DEVELOPING.md

File metadata and controls

48 lines (35 loc) · 2.45 KB
Raw

Developing

Overview

The Cloud Bigtable client is composed of two layers:

  • an automatically generated GAPIC base layer
  • a handwritten layer

Generated GAPIC base layer

The GAPIC layer is generated from a combination of protobuf and GAPIC yaml definitions found in googleapis. However the API is too protocol focused and is too low level to be consumed by an application directly. Instead, it's intended to be an implementation detail of the handwritten layer.

Handwritten layer

The handwritten layer for the data clients is meant to wrap the GAPIC layer entirely. This is necessary because some APIs are too protocol focused (ReadRows chunks vs logical row). Furthermore by constraining the api, the surface api can be made more ergonomic:

  • method calls have an implicit context of an instance
  • timestamps can be generated on the client side to make mutations idempotent
  • a handwritten api is a lot less verbose than the autogenerated protobuf api

Structure

  • BigtableDataSettings: replaces BaseBigtableDataSettings. It's a simple wrapper of stub/EnhancedBigtableStubSettings. This is meant to be the public facing api of settings.
  • BigtableDataClient: replaces BaseBigtableDataClient. It provides an ergonomic wrapper around stub/EnhancedBigtableStub. This class is meant to be simple syntactic sugar around the stub.
  • stub/EnhancedBigtableStubSettings: wraps the autogenerated stub/BigtableStubSettings. This class exposes all of the necessary settings. The settings can be categorized as:
    • General settings that affect all RPCs and include things like the target instance and connection pooling
    • Method specific settings that allow the user to tune each RPC individually
  • stub/EnhancedBigtableStub: wraps the autogenerated GrpcBigtableStub. It layers logical structure on top of the protocol level requests and responses. It is responsible for adapting ReadRows chunks to logical rows and for converting between raw protobufs and user facing wrappers.
  • models/: contains user facing wrappers for the protobufs. Each wrapper has a toProto method that will return the corresponding proto.
  • internal/: contains internal implementation details. The end user doesn't directly interact with these classes. It contains things like RequestContext, that is used by all toProto methods to contextualize the user facing wrapper with the target instance and app profile.