/fred.rs

An async Redis client for Rust.

Primary LanguageRustApache License 2.0Apache-2.0

Fred

License License CircleCI Crates.io API docs

A high level async Redis client for Rust built on Tokio and Futures.

Example

use fred::prelude::*;

#[tokio::main]
async fn main() -> Result<(), RedisError> {
  let config = RedisConfig::default();
  let policy = ReconnectPolicy::default();
  let client = RedisClient::new(config);
  
  // connect to the server, returning a handle to the task that drives the connection
  let _ = client.connect(Some(policy));
  let _ = client.wait_for_connect().await?;
 
  // convert responses to many common Rust types
  let foo: Option<String> = client.get("foo").await?;
  assert!(foo.is_none());
  
  let _: () = client.set("foo", "bar", None, None, false).await?;
  // or use turbofish to declare response types
  println!("Foo: {:?}", client.get<String, _>("foo").await?);
  
  // or use a lower level interface for responses to defer parsing, etc
  let foo: RedisValue = client.get("foo").await?;
  assert_eq!(foo.as_str().unwrap(), "bar");
  
  let _ = client.quit().await?;
  Ok(())
}

See the examples for more.

Install

With cargo edit.

cargo add fred

Features

  • Supports RESP2 and RESP3 protocol modes.
  • Supports clustered, centralized, and sentinel Redis deployments.
  • Optional built-in reconnection logic with multiple backoff policies.
  • Publish-Subscribe and keyspace events interfaces.
  • Supports transactions.
  • Supports Lua scripts.
  • Supports streaming results from the MONITOR command.
  • Supports custom commands provided by third party modules.
  • Supports TLS connections.
  • Supports streaming interfaces for scanning functions.
  • Options to automatically pipeline requests when possible.
  • Automatically retry requests under bad network conditions.
  • Built-in tracking for network latency and payload size metrics.
  • An optional client pooling interface to round-robin requests among a pool of clients.
  • An optional sentinel client for interacting directly with sentinel nodes to manually fail over servers, etc.
  • An optional pubsub subscriber client that will automatically manage channel subscriptions.
  • Optional built in support for JSON values.

Note: Fred requires Tokio 1.x or above. Actix users must be using 4.x or above as a result.

Tracing

This crate supports tracing via the tracing crate. See the tracing info for more information.

This feature is disabled by default, but callers can opt-in via the full-tracing or partial-tracing features.

Logging

To enable logs use the environment variable RUST_LOG with a value of trace, debug, warn, error, or info. See the documentation for env_logger for more information.

When a client is initialized it will generate a unique client name with a prefix of fred-. This name will appear in all logging statements on the client in order to associate client and server operations if logging is enabled on both.

Compile Time Features

Name Default Description
enable-tls x Enable TLS support. This requires OpenSSL (or equivalent) dependencies.
vendored-tls Enable TLS support, using vendored OpenSSL (or equivalent) dependencies, if possible.
ignore-auth-error x Ignore auth errors that occur when a password is supplied but not required.
metrics Enable the metrics interface to track overall latency, network latency, and request/response sizes.
reconnect-on-auth-error A NOAUTH error is treated the same as a general connection failure and the client will reconnect based on the reconnection policy. This is recommended if callers are using ElastiCache.
pool-prefer-active x Prefer connected clients over clients in a disconnected state when using the RedisPool interface.
full-tracing Enable full tracing support. This can emit a lot of data so a partial tracing feature is also provided.
partial-tracing Enable partial tracing support, only emitting traces for top level commands and network latency. Note: this has a non-trivial impact on performance.
blocking-encoding Use a blocking task for encoding or decoding frames over a certain size. This can be useful for clients that send or receive large payloads, but will only work when used with a multi-thread Tokio runtime.
network-logs Enable TRACE level logging statements that will print out all data sent to or received from the server. These are the only logging statements that can ever contain potentially sensitive user data.
custom-reconnect-errors Enable an interface for callers to customize the types of errors that should automatically trigger reconnection logic.
monitor Enable an interface for running the MONITOR command.
sentinel-client Enable an interface for communicating directly with Sentinel nodes. This is not necessary to use normal Redis clients behind a sentinel layer.
sentinel-auth Enable an interface for using different authentication credentials to sentinel nodes.
subscriber-client Enable a higher level subscriber client that manages channel subscription state for callers.
serde-json Enable an interface to automatically convert Redis types to JSON.
no-client-setname Disable the automatic CLIENT SETNAME command used to associate server logs with client logs.

Environment Variables

Name Default Description
FRED_DISABLE_CERT_VERIFICATION false Disable certificate verification when using TLS features.
FRED_DISABLE_HOST_VERIFICATION false Disable host verification when using TLS features.

Pipelining

The caller can toggle pipelining via flags on the RedisConfig provided to a client to enable automatic pipelining for commands whenever possible. These settings can drastically affect performance on both the server and client, but further performance tuning may be necessary to avoid issues such as using too much memory on the client or server while buffering commands.

This module also contains a separate test application that can be used to demonstrate the effects of pipelining. This test application also contains some helpful information on how to use the tracing features.

ACL & Authentication

Prior to the introduction of ACL commands in Redis version 6 clients would authenticate with a single password. If callers are not using the ACL interface, or using Redis version <=5.x, they should configure the client to automatically authenticate by using the password field on the RedisConfig and leaving the username field as None.

If callers are using ACLs and Redis version >=6.x they can configure the client to automatically authenticate by using the username and password fields on the provided RedisConfig.

It is required that the authentication information provided to the RedisConfig allows the client to run CLIENT SETNAME and CLUSTER NODES. Callers can still change users via the AUTH command later, but it recommended to instead use the username and password provided to the RedisConfig so that the client can automatically authenticate after reconnecting.

If this is not possible callers need to ensure that the default user can run the two commands above. Additionally, it is recommended to move any calls to the AUTH or HELLO command inside the on_reconnect block.

Redis Sentinel

To use the Redis Sentinel interface callers should provide a ServerConfig::Sentinel variant when creating the client's RedisConfig. This should contain the host/port tuples for the known sentinel nodes when first creating the client.

The client will automatically update these values in place as sentinel nodes change whenever connections to the primary Redis server close. Callers can inspect these changes with the client_config function on any RedisClient that uses the sentinel interface.

Note: Sentinel connections will use the same TLS configuration options as the connections to the Redis servers. By default connections will also use the same authentication credentials as well unless the sentinel-auth feature is enabled.

Callers can also use the sentinel-client feature to communicate directly with Sentinel nodes.

Customizing Error Handling

The custom-reconnect-errors feature enables an interface on the globals to customize the list of errors that should automatically trigger reconnection logic (if configured).

In many cases applications respond to Redis errors by logging the error, maybe waiting and reconnecting, and then trying again. Whether to do this often depends on the prefix in the error message, and this interface allows callers to specify which errors should be handled this way.

Errors that trigger this can be seen with the on_error function.

Tests

See the testing documentation for more information.

Beware: the tests will periodically run flushall.

Contributing

See the contributing documentation for info on adding new commands.