(WIP) RFC-024: Cache

[jerrykingxyz]

• Feature Name: rspack_cache
• Start Date: 2022/11/14
• Rspack Issue: support memory cache to speed up HMR performance #949

Summary

Cache support for rspack.

Motivation

Cache can improve build speed.

Guide-level explanation

There are two parts of the configuration that are exposed to user.

Cache

Use cache field can config storage strategy.

// disable cache
{ "cache": false }
// use memory cache
{ "cache": true | undefined }

// custom memory cache
{
  "cache": {
    "type": "memory",
    // - Define the lifespan of unused cache entries in the memory cache.
    "maxGenerations": Infinity
  }
}

// custom filesystem cache
{
  "cache": {
    "type": "filesystem",
    // - Collect unused memory allocated during deserialization
    // "allowCollectingMemory": boolean,
    // - Define the lifespan of unused cache entries in the memory cache.
    // "maxMemoryGenerations": number,
    // - The amount of time in milliseconds that unused cache entries are allowed to stay in the filesystem cache; defaults to one month.
    "maxAge": number,
    // - Track and log detailed timing information for individual cache items of type 'filesystem'.
    "profile": boolean,
    // - An object of arrays of additional code dependencies for the build.
    "buildDependencies": Object,
    // - Base directory for the cache. Defaults to node_modules/.cache/rspack.
    "cacheDirectory": string,
    // - Locations for the cache. Defaults to path.resolve(cache.cacheDirectory, cache.name)
    "cacheLocation": string,
    // - Algorithm used the hash generation.
    // "hashAlgorithm": "md4" | "md5" | ...,
    // - Compression type used for the cache files.
    // "compression": false | "gzip" | "brotli",
    // - Dnotes the time period after which the cache storing should happen. 
    // "idleTimeout": number,
    // "idleTimeoutForInitialStore": number,
    // "idleTimeoutAfterLargeChanges": number,
    // - Name for the cache
    "name": string,
    // - Store strategy
    // "store": "pack",
    // - Version of the cache data.
    "version": string
  }
}

Snapshot

Use snapshot options decide how the file system snapshots are created and invalidated.

{
  "snapshot": {
    // - Snapshots for resolving of build dependencies when using the persistent cache.
    "resolveBuildDependencies": { hash: true, timestamp: true },
    // - Snapshots for build dependencies when using the persistent cache.
    "buildDependencies": { hash: true, timestamp: true },
    // - Snapshots for resolving of requests.
    "resolve": { hash: true, timestamp: true },
    // - Snapshots for building modules.
    "module": { hash: true, timestamp: true },
    // - An array of paths that are managed by a package manager and contain a version or a hash in their paths.
    // "immutablePaths": Array<RegExp | string>,
    // - An array of paths that are managed by a package manager.
    // "managedPaths": Array<RegExp | string>,
  }
}

Reference-level explanation

Storage

We need to create a trait to abstract storage operations. Both memory cache & filesystem cache implement this trait.

trait Cache<Data> {
  fn get(id: &str, etag: &str) -> Data;
  fn set(id: &str, etag: &str, data: Data);
  fn begin_idle();
  fn end_idle();
  fn shutdown();
}

struct MemoryCache<D> {
  data: HashMap<String, (String, D)>
}
impl Cache<D> for MemoryCache<D> {}

struct MemoryWithGCCache<D> {
  data: HashMap<String, (String, D)>
}
impl Cache<D> for MemoryWithGCCache<D> {}

struct FileSystemCache<D> {
  data: HashMap<String, (String, D)>
}
impl Cache<D> for FileSystemCache<D> {}

Rspack will create and save cache instance in compiler, and shared to compilation.

Snapshot

The snapshot checks whether a file has changed using a timestamp or hash.

struct Snapshot {
  timestamp: u64,
  hash: String,
  filePath: String
  managedPath: Vec<String>
}

impl Snapshot {
  fn new() -> Self {}
  fn checkValid() -> bool {}
}

Use cache

• Build module
  Because of swc ast & context can not be serialize, the build module stage will not be cache ast info.

Dev Stage

1. Basically available

• support snapshot
• support memory cache
• use cache in compilation make

2. All compilation stages available

• use cache in compilation seal
• use cache in emit assets
• support unsafe cache config

3. Support persistent storage

4. improve cache platform

Drawbacks

Nope

Prior art

#1111

Future possibilities

Nope

[zackarychapple]

Resurrecting this, it is a very big improvement for large projects that are still taking 10-20 seconds with Rspack. While we work on improving first build, having cache for subsequent builds would be a huge win.

[Pablinho]

Any thoughts on when rspack will support the file system caching, please? Working on a large project and this is a blocker to migrate from webpack.