TiKV deep dive
- 1. Introduction
- 2. Consensus algorithm
- 3. Key-value engine
- 4. Distributed transaction
- 5. Scalability
- 6. Remote Procedure Calls (RPC)
- 7. Resource scheduling
- 8. Distributed SQL over TiKV
- 8.1. Distributed SQL
Remote Procedure Calls (RPC)
Communication between services occurs over remote procedure calls. RPCs happen all the time in distributed systems. To obtain a webpage, your browser has to make at least one RPC to this website.
TiKV, as a distributed system involving a number of nodes, uses RPCs to communicate between nodes, as well as the Placement Driver and clients.
As it turns out, exposing functionality as a remote interface isn’t a trivial. Networks are unreliable, systems are diverse and evolving, a huge variety of languages and formats exist, and even things like encoding are hard!
On the shoulders of giants
Over the past decades the field of computing has largely settled on a few common standards.
The vast majority of services work over the HTTP or HTTP/2 network protocols. This solves problems such as:
TiKV uses HTTP/2. HTTP/2 is more performant and capable than HTTP/1 for TiKV uses.
With those abilities supported, there remains a need to work with structures of data. Commonly this ends up being an interface description language format like Protocol Buffers, which TiKV uses. Unlike an interchange format, an interface description language allows for the definition of services and RPCs in addition to just data interchange. This solves the following problems:
- Serialization/Deserialization format (Text-to-
structand vice versa)
- Backward/Forward compatibility (e.g. Structure fields changing, being added, removed)
- Service & RPC definition
Wrapping it all together
Simply having the pieces is not enough. Making it usable for all parties involved is another story. gRPC does a great job wrapping up the above technologies and providing usable interfaces.
Over the next chapter, we’ll look at each of these technologies and how they work.