Design documentation for adding a raw-FFI thread manager

docs/design-api.md

@@ -0,0 +1,114 @@
+API Design
+
+# Client Wrapper API design doc
+
+## API requirements:
+- The API will be thread-safe.
+- The API will accept as inputs all of [RESP2 types](https://github.com/redis/redis-specifications/blob/master/protocol/RESP2.md).
+- The API will attempt authentication, topology refreshes, reconnections, etc., automatically. In case of failures concrete errors will be returned to the user.
+
+## Command Interface
+
+### Unix Domain Socket solution
+For clients based on Unix Domain Sockets (UDS), we will simply use the existing protobuf messages for creating a connection, sending requests, and receiving responses. Supported commands are enumerated in the [protobuf definition for requests](../babushka-core/src/protobuf/redis_request.proto) and we may add more in the future, although the `CustomCommand` request type is also adequate for all commands. As defined in the [protobuf definition for responses](../babushka-core/src/protobuf/response.proto), client wrappers will receive data as a pointer, which can be passed to Rust to marshal the data back into the wrapper language’s native data types.
+
+Transactions will be handled by adding a list of `Command`s to the protobuf request. The response will be a `redis::Value::Bulk`, which should be handled in the same Rust function that marshals the data back into the wrapper language's native data types. This is handled by storing the results in a collection type native to the wrapper language.
+
+When running Redis in Cluster Mode, several routing options will be provided. These are all specified in the protobuf request. The various options are detaield below in the ["Routing Options" section](#routing-options). We will also provide a separate client for handling Cluster Mode responses, which will convert the list of values and nodes into a map, as is done in existing client wrappers.
+
+### Raw FFI solution
+For clients using a raw FFI solution, in Rust, we will expose a general command that is able to take any command and arguments as strings. 
+
+We have 2 options for passing the command, arguments, and any additional configuration to the Rust core from the wrapper language:
+
+#### Protobuf
+The wrapper language will pass the commands, arguments, and configuration as protobuf messages using the same definitions as in the UDS solution.
+
+Transactions will be handled by adding a list of `Command`s to the protobuf request. The response will be a `redis::Value::Bulk`, which can be marshalled into a C array of values before being passed from Rust to the wrapper language. The wrapper language is responsible for converting the array of results to its own native collection type.
+
+Cluster Mode support is the same here as in the UDS solution detailed above.
+
+Pros:
+- We get to reuse the protobuf definitions, meaning fewer files to update if we make changes to the protobuf definitions
+- May be simpler to implement compared to the C data types solution, since we do not need to define our own C data types
+
+Cons:
+- There is additional overhead from marshalling data to and from protobuf, which could impact performance significantly
+
+#### C Data Types
+The wrapper language will pass commands, arguments, and configuration as C data types. 
+
+Transactions will be handled by passing a C array of an array of arguments to Rust from the wrapper language. The response will be a `redis::Value::Bulk`, which can be marshalled in the same way as explained in the protobuf solution.
+
+For Cluster Mode support, [routing options](#routing-options) will be defined as C enums and structs. Like in the protobuf solution, we will provide a separate client for handling Cluster Mode responses, which will convert the list of values and nodes into a map.
+
+Pros:
+- No additional overhead from marshalling to and from protobuf, so this should perform better
+- May be simpler to implement compared to protobuf solution, since it can be tricky to construct protobuf messages in a performant way and we have to add a varint to the messages as well 
+
+Cons:
+- Would add an additional file to maintain containing the C definitions (only one file though, since we could share between all raw FFI solutions), which we would need to update every time we want to update the existing protobuf definitions
+
+We will be testing both approaches to see which is easier to implement, as well as the performance impact before deciding on a solution.
+
+To marshal Redis data types back into the corresponding types for the wrapper language, we will convert them into appropriate C types, which can then be translated by the wrapper language into its native data types. Here is what a Redis result might look like:
+```
+typedef struct redisValue {
+    enum {NIL, INT, DATA, STATUS, BULK, OKAY} kind;
+    union Payload {
+        long intValue;
+        unsigned char *dataValue;
+        char *statusValue;
+        struct redisValue *bulkValue;
+    } payload;
+} RedisValue
+```
+
+## Routing Options
+We will be supporting routing Redis requests to all nodes, all primary nodes, or a random node. For more specific routing to a node, we will also allow sending a request to a primary or replica node with a specified hash slot or key. When the wrapper given a key route, the key is passed to the Rust core, which will find the corresponding hash slot for it.
+
+## Supported Commands
+We will be supporting all Redis commands. Commands with higher usage will be prioritized, as determined by usage numbers from AWS ElastiCache usage logs.
+
+## Type Validation
+Two different methods of sending requests will be supported. 
+
+### No Redis Type Validation
+We will expose an `executeRaw` method that does no validation of the input types or command on the client side, leaving it up to Redis to reject the request should it be malformed. This gives the user the flexibility to send any type of request they want, including ones not officially supported yet.
+
+For example, if a user wants to implement support for the Redis ZADD command in Java, their implementation might look something like this:
+```java
+public Long zadd(K key, double score, V member) throws RequestException {
+    string[] args = { key.toString(), score.toString(), member.toString() };
+    return (Long) executeRaw(args);
+}
+```
+
+where `executeRaw` has the following signature:
+```java
+public Object executeRaw(string[] args) throws RequestException
+```
+
+### With Redis Type Validation
+We will expose separate methods for each supported command, and will attempt to validate the inputs for each of these methods. There will be a separate version of each method for transactions, as well as another version for Cluster Mode clients. We may leverage the compiler for the wrapper language to validate the types of the command arguments for statically typed languages. Since wrappers should be as lightweight as possible, we most likely will not be performing any checks for proper typing for non-statically typed languages.
+
+## Errors
+ClosingError: Errors that report that the client has closed and is no longer usable.
+
+RedisError: Errors that were reported during a request.
+
+TimeoutError: Errors that are thrown when a request times out.
+
+ExecAbortError: Errors that are thrown when a transaction is aborted.
+
+ConnectionError: Errors that are thrown when a connection disconnects. These errors can be temporary, as the client will attempt to reconnect.
+
+Errors returned are subject to change as we update the protobuf definitions.
+
+## Java Specific Details
+We will be using the UDS solution for communication between the wrapper and the Rust core. This thin layer is implemented using the [jni-rs library](https://github.com/jni-rs/jni-rs) to start the socket listener and marshal Redis values into native Java data types.
+
+Errors in Rust are represented as Algebraic Data Types, which are not supported in Java by default (at least not in the versions of Java we want to support). Instead, we utilise the [jni-rs library](https://github.com/jni-rs/jni-rs) to throw Java `Exception`s where we receive errors from Redis.
+
+## Golang Specific Details
+We will be using a raw FFI solution for communication between the wrapper and the Rust core. TODO: Add more details here

docs/design-raw-ffi.md

@@ -0,0 +1,235 @@
+# Babushka Core Wrappers
+
+## Summary
+
+The Babushka client allows Redis users to connect to Redis using a variety of commands through a thin-client optimized for 
+various languages.  The client uses a performant core to establish and manage connections and communicate with Redis. The thin-client
+wrapper talks to the core using an FFI (foreign function interface) to Rust. 
+
+The following document discusses two primary communication protocol architectures for wrapping the Babushka clients. Specifically, 
+it details how Java-Babushka and Go-Babushka each use a different protocol and describes the advantages of each language-specific approach.
+
+# Unix Domain Socket Manager Connector
+
+## High-Level Design 
+
+**Summary**: The Babushka "UDS" solution uses a socket listener to manage rust-to-wrapper worker threads, and unix domain sockets 
+to deliver command requests between the wrapper and redis-client threads.  This works well because we allow the unix sockets to pass messages and manage threads
+through the OS, and unix sockets are very performant. This results in simple/fast communication.  The risk to avoid is that 
+unix sockets can become a bottleneck for data-intensive commands, and the library can spend too much time waiting on I/O
+blocking operations.
+
+```mermaid
+stateDiagram-v2
+  direction LR
+
+  Wrapper: Wrapper
+  UnixDomainSocket: Unix Domain Socket
+  RustCore: Rust-Core
+
+  [*] --> Wrapper: User
+  Wrapper --> UnixDomainSocket
+  UnixDomainSocket --> Wrapper
+  RustCore --> UnixDomainSocket
+  UnixDomainSocket --> RustCore
+  RustCore --> Redis
+  Redis --> RustCore
+```
+
+## Decision to use UDS Sockets for a Java-Babushka Wrapper
+
+The decision to use Unix Domain Sockets (UDS) to manage the Java-wrapper to Babushka Redis-client communication was thus:  
+1. Java contains an efficient socket protocol library ([netty.io](https://netty.io/)) that provides a highly configurable environment to manage sockets. 
+2. Java objects serialization/de-serialization is an expensive operation, and a performing multiple io operations between raw-ffi calls would be inefficient. 
+3. The async FFI requests with callbacks requires that we manage multiple runtimes (Rust and Java Thread management), and JNI does not provide an out-of-box solution for this.
+
+### Decision Log
+
+| Protocol                                     | Details                                                     | Pros                        | Cons                                               |
+|----------------------------------------------|-------------------------------------------------------------|-----------------------------|----------------------------------------------------|
+| Unix Domain Sockets (jni/netty)              | JNI to submit commands; netty.io for message passing; async | netty.io standard lib;      | complex configuration; limited by socket interface |
+| Raw-FFI (JNA, uniffi-rs, j4rs, interoptopus) | FFI to submit commands; Rust for message processing         | reusable in other languages | slow performance and uses JNI under the hood       |
+| Panama/jextract                              | Performance similar to a raw-ffi using JNI                  | modern                      | lacks early Java support (JDK 18+); prototype      |
+
+### Sequence Diagram
+
+```mermaid
+sequenceDiagram
+
+participant Wrapper as Client-Wrapper
+participant ffi as FFI
+participant manager as Rust-Core
+participant worker as Tokio Worker
+participant SocketListener as Socket Listener
+participant Socket as Unix Domain Socket
+participant Client as Redis
+
+activate Wrapper
+activate Client
+Wrapper -)+ ffi: connect_to_redis
+ffi -)+ manager: start_socket_listener(init_callback)
+    manager -) worker: Create Tokio::Runtime (count: CPUs)
+        activate worker
+        worker ->> SocketListener: listen_on_socket(init_callback)
+        SocketListener ->> SocketListener: loop: listen_on_client_stream
+            activate SocketListener
+        SocketListener -->> manager: 
+    manager -->> ffi: socket_path
+ffi -->>- Wrapper: socket_path
+    SocketListener -->> Socket: UnixStreamListener::new
+    activate Socket
+    SocketListener -->> Client: BabushkaClient::new
+Wrapper ->> Socket: connect 
+    Socket -->> Wrapper: 
+loop single_request
+    Wrapper ->> ffi: java_arg_to_redis
+    ffi -->> Wrapper: 
+    Wrapper -> Wrapper: pack protobuf.redis_request
+    Wrapper ->> Socket: netty.writeandflush (protobuf.redis_request)
+    Socket -->> Wrapper: 
+    Wrapper ->> Wrapper: wait
+        SocketListener ->> SocketListener: handle_request
+        SocketListener ->> Socket: read_values_loop(client_listener, client)
+            Socket -->> SocketListener: 
+        SocketListener ->> Client: send(request)
+        Client -->> SocketListener: ClientUsageResult
+        SocketListener ->> Socket: write_result
+            Socket -->> SocketListener: 
+    Wrapper ->> Socket: netty.read (protobuf.response)
+        Socket -->> Wrapper: 
+    Wrapper ->> ffi: redis_value_to_java
+    ffi -->> Wrapper: 
+    Wrapper ->> Wrapper: unpack protobuf.response 
+end
+Wrapper ->> Socket: close() 
+Wrapper ->> SocketListener: shutdown
+    SocketListener ->> Socket: close()
+    deactivate Socket
+    SocketListener ->> Client: close()
+    SocketListener -->> Wrapper: 
+    deactivate SocketListener
+    deactivate worker
+    deactivate Wrapper
+    deactivate Client
+```
+
+### Discussion
+* `redis_value_to_java`: This ffi call is necessary to evaluate the Redis::Value response that Redis returns to Rust-core, 
+and needs to be converted to a `JObject` before it can be evaluated by Java. We are looking for alternatives to this call
+to avoid an unnecessary ffi call. 
+* `java_arg_to_redis`: This ffi call is currently unnecessary, because all arguments sent are Strings. 
+
+
+### Elements
+* **Wrapper**: Our Babushka wrapper that exposes a client API (java, python, node, etc)
+* **Babushka FFI**: Foreign Function Interface definitions from our wrapper to our Rust Babushka-Core
+* **Babushka impl**: public interface layer and thread manager
+* **Tokio Worker**: Tokio worker threads (number of CPUs)
+* **SocketListener**: listens for work from the Socket, and handles commands
+* **Unix Domain Socket**: Unix Domain Socket to handle incoming requests and response payloads between Rust-Core and Wrapper 
+* **Redis**: Our data store
+
+## Wrapper-to-Core Connector with raw-FFI calls
+
+**Summary**: Foreign Function Interface (FFI) calls are simple to implement, cross-language calls.  The setup between Golang and the Rust-core
+is fairly simple using the well-supported CGO library.  While sending language calls is easy, setting it up in an async manner
+requires that we handle async callbacks. Golang has a simple, light-weight solution to that, using goroutines and channels, 
+to pass callbacks and execution between the languages. 
+
+```mermaid
+stateDiagram-v2
+  direction LR
+
+  Wrapper: Golang Wrapper
+  FFI: Foreign Function Interface
+  RustCore: Rust-Core
+
+  [*] --> Wrapper: User
+  Wrapper --> FFI
+  FFI --> Wrapper
+  RustCore --> FFI
+  FFI --> RustCore
+  RustCore --> Redis
+```
+
+## Decision to use Raw-FFI calls directly to Rust-Core for Golang Wrapper
+
+### Decision Log
+
+The decision to use raw FFI request from Golang to Rust-core was straight forward:
+1. Golang contains goroutines as an alternative, lightweight, and performant solution serves as an obvious solution to pass request, even at scale. 
+
+Due to lightweight thread management solution, we chose a solution that scales quickly and requires less configuration to achieve a performant solution 
+on par with existing industrial standards ([go-redis](https://github.com/redis/go-redis)).
+
+| Protocol                 | Details | Pros                                                   | Cons                                 |
+|--------------------------|---------|--------------------------------------------------------|--------------------------------------|
+| Unix Domain Sockets      |         | UDS performance; consistent protocol between languages | complex configuration                |
+| Raw-FFI (CGO/goroutines) |         | simplified and light-weight interface                  | separate management for each request |
+
+## Sequence Diagram - Raw-FFI Client
+
+**Summary**: If we make direct calls through FFI from our Wrapper to Rust, we can initiate commands to Redis.  This allows us
+to make on-demand calls directly to Rust-core solution. Since the calls are async, we need to manage and populate a callback
+object with the response and a payload.  
+
+We will need to avoid busy waits while waiting on the async response. The wrapper and Rust-core languages independently track
+threads.  On the Rust side, they use a Tokio runtime to manage threads. When the Rust-core is complete, and returning a Response,
+we can use the Callback object to re-awake the wrapper thread manager and continue work.
+
+Go routines have a performant solution using light-weight go-routines and channels.  Instead of busy-waiting, we awaken by 
+pushing goroutines to the result channel once the Tokio threads send back a callback.  
+
+### Sequence Diagram
+
+
+```mermaid
+sequenceDiagram
+
+participant Wrapper as Client-Wrapper
+participant channel as Result Channel 
+participant ffi as Babushka FFI
+participant manager as Babushka impl
+participant worker as Tokio Worker
+participant Client as Redis
+
+activate Wrapper
+activate Client
+Wrapper -)+ ffi: create_connection(connection_settings)
+ffi ->>+ manager: start_thread_manager(init_callback)
+    manager ->> worker: Create Tokio::Runtime (count: CPUs)
+        activate worker
+    manager -->> Wrapper: Ok(BabushkaClient)
+        worker ->> Client: BabushkaClient::new
+        worker ->> worker: wait_for_work(init_callback)
+
+loop single_request
+Wrapper ->> channel: make channel
+    activate channel
+Wrapper -) ffi: command: single_command(protobuf.redis_request, &channel)
+Wrapper ->> channel: wait
+    ffi ->> manager: cmd(protobuf.redis_request)
+        manager ->> worker: command: cmd(protobuf.redis_request)
+            worker ->> Client: send(command, args)
+            Client -->> worker: Result
+        worker -->> ffi: Ok(protobuf.response<Redis::Value>)
+    ffi -->> channel: Ok(protobuf.response<Result>)
+channel ->> Wrapper: protobuf.response<Result>
+Wrapper ->> channel: close
+    deactivate channel
+end
+
+Wrapper -) worker: close_connection
+    worker -->> Wrapper: 
+    deactivate worker
+    deactivate Wrapper
+    deactivate Client
+```
+
+### Elements
+* **Client-Wrapper**: Our Babushka wrapper that exposes a client API (Go, etc)
+* **Result Channel**: Goroutine channel on the Babushka Wrapper 
+* **Babushka FFI**: Foreign Function Interface definitions from our wrapper to our Rust Babushka-Core
+* **Babushka impl**: public interface layer and thread manager
+* **Tokio Worker**: Tokio worker threads (number of CPUs)
+* **Redis**: Our data store