Remove GlobalTransformCache.js

Reviewed By: davidaurelio

Differential Revision: D7628728

fbshipit-source-id: 6b0c54e3e84e74cf531c17b4644da382e35f3ed4
This commit is contained in:
Miguel Jimenez Esun 2018-04-20 07:24:23 -07:00 committed by Facebook Github Bot
parent 084ac30863
commit df53fb3de5
3 changed files with 0 additions and 644 deletions

View File

@ -1,510 +0,0 @@
/**
* Copyright (c) 2016-present, Facebook, Inc.
*
* This source code is licensed under the MIT license found in the
* LICENSE file in the root directory of this source tree.
*
* @flow
* @format
*/
'use strict';
/* global Buffer: true */
const BatchProcessor = require('./BatchProcessor');
const FetchError = require('node-fetch/lib/fetch-error');
const crypto = require('crypto');
const fetch = require('node-fetch');
const jsonStableStringify = require('json-stable-stringify');
const path = require('path');
const throat = require('throat');
import type {
CustomTransformOptions,
Options as TransformWorkerOptions,
TransformOptionsStrict,
} from '../JSTransformer/worker';
import type {LocalPath} from '../node-haste/lib/toLocalPath';
import type {CachedResult, GetTransformCacheKey} from './TransformCaching';
import type {Agent as HttpAgent} from 'http';
import type {Agent as HttpsAgent} from 'https';
type FetchOptions = {agent?: ?HttpAgent | HttpsAgent};
/**
* The API that a global transform cache must comply with. To implement a
* custom cache, implement this interface and pass it as argument to the
* application's top-level `Server` class.
*/
export type GlobalTransformCache = {
keyOf(props: FetchProps): string,
/**
* Synchronously determine if it is worth trying to fetch a result from the
* cache. This can be used, for instance, to exclude sets of options we know
* will never be cached.
*/
shouldFetch(props: FetchProps): boolean,
/**
* Try to fetch a result. It doesn't actually need to fetch from a server,
* the global cache could be instantiated locally for example.
*/
fetch(key: string): Promise<?CachedResult>,
/**
* Try to store a result. Callsites won't necessarily wait for the success or
* failure of the Promise, so errors may be handled internally, they may
* otherwise be silently ignored. The actual storage operation could be done
* at a later point if desired. It is recommended to have this function be a
* no-op in production, and only do the storage operation from a script
* running on a Continuous Integration platform.
*/
store(key: string, result: CachedResult): Promise<void>,
};
type FetchResultURIs = (keys: Array<string>) => Promise<Map<string, string>>;
type FetchResultFromURI = (uri: string) => Promise<?CachedResult>;
type StoreResults = (resultsByKey: Map<string, CachedResult>) => Promise<void>;
export type FetchProps = {
+localPath: LocalPath,
+sourceCode: string,
+getTransformCacheKey: GetTransformCacheKey,
+transformOptions: TransformWorkerOptions,
};
type URI = string;
/**
* We aggregate the requests to do a single request for many keys. It also
* ensures we do a single request at a time to avoid pressuring the I/O.
*/
class KeyURIFetcher {
_batchProcessor: BatchProcessor<string, ?URI>;
_fetchResultURIs: FetchResultURIs;
/**
* When a batch request fails for some reason, we process the error locally
* and we proceed as if there were no result for these keys instead. That way
* a build will not fail just because of the cache.
*/
async _processKeys(keys: Array<string>): Promise<Array<?URI>> {
const URIsByKey = await this._fetchResultURIs(keys);
return keys.map(key => URIsByKey.get(key));
}
async fetch(key: string): Promise<?string> {
return await this._batchProcessor.queue(key);
}
constructor(fetchResultURIs: FetchResultURIs) {
this._fetchResultURIs = fetchResultURIs;
this._batchProcessor = new BatchProcessor(
{
maximumDelayMs: 10,
maximumItems: 500,
concurrency: 2,
},
this._processKeys.bind(this),
);
}
}
type KeyedResult = {key: string, result: CachedResult};
class KeyResultStore {
_storeResults: StoreResults;
_batchProcessor: BatchProcessor<KeyedResult, void>;
_promises: Array<Promise<void>>;
async _processResults(keyResults: Array<KeyedResult>): Promise<Array<void>> {
const resultsByKey = new Map(
keyResults.map(pair => [pair.key, pair.result]),
);
await this._storeResults(resultsByKey);
return new Array(keyResults.length);
}
async store(key: string, result: CachedResult): Promise<void> {
await this._batchProcessor.queue({key, result});
}
constructor(storeResults: StoreResults) {
this._storeResults = storeResults;
this._batchProcessor = new BatchProcessor(
{
maximumDelayMs: 1000,
maximumItems: 100,
concurrency: 10,
},
this._processResults.bind(this),
);
this._promises = [];
}
}
export type TransformProfile = {
+dev: boolean,
+minify: boolean,
+platform: ?string,
};
function profileKey({dev, platform}: TransformProfile): string {
return jsonStableStringify({dev, platform});
}
/**
* We avoid doing any request to the server if we know the server is not
* going to have any key at all for a particular set of transform options.
*/
class TransformProfileSet {
_profileKeys: Set<string>;
constructor(profiles: Iterable<TransformProfile>) {
this._profileKeys = new Set();
for (const profile of profiles) {
this._profileKeys.add(profileKey(profile));
}
}
has(profile: TransformProfile): boolean {
return this._profileKeys.has(profileKey(profile));
}
}
type FetchFailedDetails =
| {
+statusCode: number,
+statusText: string,
+type: 'unhandled_http_status',
+uri: string,
}
| {+type: 'invalid_data'}
| {+type: 'invalid_key_data', key: string};
class FetchFailedError extends Error {
/** Separate object for details allows us to have a type union. */
+details: FetchFailedDetails;
constructor(details: FetchFailedDetails) {
super(FetchFailedError._getMessage(details));
(this: any).details = details;
}
static _getMessage(details: FetchFailedDetails): string {
if (details.type === 'unhandled_http_status') {
return (
`Unexpected HTTP status: ${details.statusCode} ` +
JSON.stringify(details.statusText) +
` while fetching \`${details.uri}\``
);
}
if (details.type === 'invalid_key_data') {
return `Invalid data was returned for key \`${details.key}\``;
}
return `Invalid or empty data was returned.`;
}
}
/**
* For some reason the result stored by the server for a key might mismatch what
* we expect a result to be. So we need to verify carefully the data.
*/
function validateCachedResult(cachedResult: mixed): ?CachedResult {
if (
cachedResult != null &&
typeof cachedResult === 'object' &&
typeof cachedResult.code === 'string' &&
Array.isArray(cachedResult.dependencies) &&
cachedResult.dependencies.every(dep => typeof dep === 'string')
) {
return (cachedResult: any);
}
return null;
}
class URIBasedGlobalTransformCache {
_fetcher: KeyURIFetcher;
_fetchResultFromURI: FetchResultFromURI;
_profileSet: TransformProfileSet;
_optionsHasher: OptionsHasher;
_store: ?KeyResultStore;
static FetchFailedError: Class<FetchFailedError>;
/**
* For using the global cache one needs to have some kind of central key-value
* store that gets prefilled using keyOf() and the transformed results. The
* fetching function should provide a mapping of keys to URIs. The files
* referred by these URIs contains the transform results. Using URIs instead
* of returning the content directly allows for independent and parallel
* fetching of each result, that may be arbitrarily large JSON blobs.
*/
constructor(props: {
fetchResultFromURI: FetchResultFromURI,
fetchResultURIs: FetchResultURIs,
profiles: Iterable<TransformProfile>,
rootPath: string,
storeResults: StoreResults | null,
}) {
this._fetcher = new KeyURIFetcher(props.fetchResultURIs);
this._profileSet = new TransformProfileSet(props.profiles);
this._fetchResultFromURI = props.fetchResultFromURI;
this._optionsHasher = new OptionsHasher(props.rootPath);
if (props.storeResults != null) {
this._store = new KeyResultStore(props.storeResults);
}
}
/**
* Return a key for identifying uniquely a source file.
*/
keyOf(props: FetchProps) {
const hash = crypto.createHash('sha1');
const {sourceCode, localPath, transformOptions} = props;
hash.update(
this._optionsHasher.getTransformWorkerOptionsDigest(transformOptions),
);
const cacheKey = props.getTransformCacheKey(transformOptions);
hash.update(JSON.stringify(cacheKey));
hash.update(JSON.stringify(localPath));
hash.update(
crypto
.createHash('sha1')
.update(sourceCode)
.digest('hex'),
);
const digest = hash.digest('hex');
return `${digest}-${path.basename(localPath)}`;
}
/**
* We may want to improve that logic to return a stream instead of the whole
* blob of transformed results. However the results are generally only a few
* megabytes each.
*/
static async _fetchResultFromURI(
uri: string,
options: FetchOptions,
): Promise<CachedResult> {
const response = await fetch(uri, {
agent: options.agent,
method: 'GET',
timeout: 8000,
});
if (response.status !== 200) {
throw new FetchFailedError({
statusCode: response.status,
statusText: response.statusText,
type: 'unhandled_http_status',
uri,
});
}
const unvalidatedResult = await response.json();
const result = validateCachedResult(unvalidatedResult);
if (result == null) {
throw new FetchFailedError({type: 'invalid_data'});
}
return result;
}
/**
* It happens from time to time that a fetch fails, we want to try these again
* a second time if we expect them to be transient. We might even consider
* waiting a little time before retring if experience shows it's useful.
*/
static _fetchResultFromURIWithRetry(
uri: string,
options?: FetchOptions = {},
): Promise<CachedResult> {
return URIBasedGlobalTransformCache._fetchResultFromURI(uri, options).catch(
error => {
if (!URIBasedGlobalTransformCache.shouldRetryAfterThatError(error)) {
throw error;
}
return URIBasedGlobalTransformCache._fetchResultFromURI(uri, options);
},
);
}
/**
* The exposed version uses throat() to limit concurrency, as making too many parallel requests
* is more likely to trigger server-side throttling and cause timeouts.
*/
static fetchResultFromURI: (
uri: string,
options?: FetchOptions,
) => Promise<CachedResult>;
/**
* We want to retry timeouts as they're likely temporary. We retry 503
* (Service Unavailable) and 502 (Bad Gateway) because they may be caused by a
* some rogue server, or because of throttling.
*
* There may be other types of error we'd want to retry for, but these are
* the ones we experienced the most in practice.
*/
static shouldRetryAfterThatError(error: mixed): boolean {
return (
(error instanceof FetchError && error.type === 'request-timeout') ||
(error instanceof FetchFailedError &&
error.details.type === 'unhandled_http_status' &&
(error.details.statusCode === 503 || error.details.statusCode === 502))
);
}
shouldFetch(props: FetchProps): boolean {
return this._profileSet.has(props.transformOptions);
}
/**
* This may return `null` if either the cache doesn't have a value for that
* key yet, or an error happened, processed separately.
*/
async fetch(key: string): Promise<?CachedResult> {
const uri = await this._fetcher.fetch(key);
if (uri == null) {
return null;
}
return await this._fetchResultFromURI(uri);
}
async store(key: string, result: CachedResult): Promise<void> {
if (this._store != null) {
await this._store.store(key, result);
}
}
}
URIBasedGlobalTransformCache.fetchResultFromURI = throat(
500,
URIBasedGlobalTransformCache._fetchResultFromURIWithRetry,
);
class OptionsHasher {
_rootPath: string;
_cache: WeakMap<TransformWorkerOptions, string>;
constructor(rootPath: string) {
this._rootPath = rootPath;
this._cache = new WeakMap();
}
getTransformWorkerOptionsDigest(options: TransformWorkerOptions): string {
const digest = this._cache.get(options);
if (digest != null) {
return digest;
}
const hash = crypto.createHash('sha1');
this.hashTransformWorkerOptions(hash, options);
const newDigest = hash.digest('hex');
this._cache.set(options, newDigest);
return newDigest;
}
/**
* This function is extra-conservative with how it hashes the transform
* options. In particular:
*
* * we need to hash paths as local paths, i.e. relative to the root, not
* the absolute paths, otherwise everyone would have a different cache,
* defeating the purpose of global cache;
* * we need to reject any additional field we do not know of, because
* they could contain absolute path, and we absolutely want to process
* these.
*
* Theorically, Flow could help us prevent any other field from being here by
* using *exact* object type. In practice, the transform options are a mix of
* many different fields including the optional Babel fields, and some serious
* cleanup will be necessary to enable rock-solid typing.
*/
hashTransformWorkerOptions(
hash: crypto$Hash,
transform: TransformWorkerOptions,
): crypto$Hash {
return this.hashTransformOptions(hash, transform);
}
/**
* The transform options contain absolute paths. This can contain, for
* example, the username if someone works their home directory (very likely).
* We get rid of this local data for the global cache, otherwise nobody would
* share the same cache keys. The project roots should not be needed as part
* of the cache key as they should not affect the transformation of a single
* particular file.
*/
hashTransformOptions(
hash: crypto$Hash,
options: TransformOptionsStrict,
): crypto$Hash {
const {
assetDataPlugins,
customTransformOptions,
enableBabelRCLookup,
dev,
hot,
inlineRequires,
minify,
platform,
projectRoot,
...unknowns
} = options;
const unknownKeys = Object.keys(unknowns);
if (unknownKeys.length > 0) {
const message = `these transform option fields are unknown: ${JSON.stringify(
unknownKeys,
)}`;
throw new CannotHashOptionsError(message);
}
/* eslint-disable no-bitwise */
hash.update(
new Buffer([
+dev |
(+hot << 2) |
(+inlineRequires << 3) |
(+enableBabelRCLookup << 4) |
(+minify << 5),
]),
);
/* eslint-enable no-bitwise */
hash.update(JSON.stringify(assetDataPlugins));
hash.update(JSON.stringify(platform));
hash.update(JSON.stringify(this.toLocalPath(projectRoot)));
hash.update(
JSON.stringify(this.sortTransformOptions(customTransformOptions || {})),
);
return hash;
}
pathsToLocal(filePaths: Array<string>): Array<string> {
return filePaths.map(this.toLocalPath, this);
}
toLocalPath(filePath: string): string {
return path.relative(this._rootPath, filePath);
}
sortTransformOptions(
options: CustomTransformOptions,
): Array<[string, mixed]> {
return Object.keys(options)
.sort()
.map(key => [key, options[key]]);
}
}
class CannotHashOptionsError extends Error {
constructor(message: string) {
super();
this.message = message;
}
}
URIBasedGlobalTransformCache.FetchFailedError = FetchFailedError;
module.exports = {URIBasedGlobalTransformCache, CannotHashOptionsError};

View File

@ -1,102 +0,0 @@
/**
* Copyright (c) 2016-present, Facebook, Inc.
*
* This source code is licensed under the MIT license found in the
* LICENSE file in the root directory of this source tree.
*
* @format
* @emails oncall+js_foundation
*/
'use strict';
jest.useRealTimers();
const mockFetch = jest.fn();
jest.mock('node-fetch', () => mockFetch);
const {URIBasedGlobalTransformCache} = require('../GlobalTransformCache');
const FetchError = require('node-fetch/lib/fetch-error');
const getTransformOptions = require('../../__fixtures__/getTransformOptions');
async function fetchResultURIs(
keys: Array<string>,
): Promise<Map<string, string>> {
return new Map(keys.map(key => [key, `http://globalcache.com/${key}`]));
}
async function fetchResultFromURI(uri: string): Promise<?CachedResult> {
return {
code: `/* code from ${uri} */`,
dependencies: [],
dependencyOffsets: [],
};
}
describe('GlobalTransformCache', () => {
it('fetches results', async () => {
const cache = new URIBasedGlobalTransformCache({
fetchResultFromURI,
fetchResultURIs,
profiles: [{dev: true, minify: false, platform: 'ios'}],
rootPath: '/root',
storeResults: null,
});
const transformOptions = await getTransformOptions();
const result = await Promise.all([
cache.fetch(
cache.keyOf({
localPath: 'some/where/foo.js',
sourceCode: '/* beep */',
getTransformCacheKey: () => 'abcd',
transformOptions,
}),
),
cache.fetch(
cache.keyOf({
localPath: 'some/where/else/bar.js',
sourceCode: '/* boop */',
getTransformCacheKey: () => 'abcd',
transformOptions,
}),
),
]);
expect(result).toMatchSnapshot();
});
describe('fetchResultFromURI', () => {
const defaultFetchMockImpl = async uri => ({
status: 200,
json: async () => ({
code: `/* code from ${uri} */`,
dependencies: [],
dependencyOffsets: [],
}),
});
beforeEach(() => {
mockFetch.mockReset();
});
it('fetches result', async () => {
mockFetch.mockImplementation(defaultFetchMockImpl);
const result = await URIBasedGlobalTransformCache.fetchResultFromURI(
'http://globalcache.com/foo',
);
expect(result).toMatchSnapshot();
});
it('retries once on timeout', async () => {
mockFetch.mockImplementation(async uri => {
mockFetch.mockImplementation(defaultFetchMockImpl);
throw new FetchError('timeout!', 'request-timeout');
});
const result = await URIBasedGlobalTransformCache.fetchResultFromURI(
'http://globalcache.com/foo',
);
expect(result).toMatchSnapshot();
});
});
});

View File

@ -1,32 +0,0 @@
// Jest Snapshot v1, https://goo.gl/fbAQLP
exports[`GlobalTransformCache fetchResultFromURI fetches result 1`] = `
Object {
"code": "/* code from http://globalcache.com/foo */",
"dependencies": Array [],
"dependencyOffsets": Array [],
}
`;
exports[`GlobalTransformCache fetchResultFromURI retries once on timeout 1`] = `
Object {
"code": "/* code from http://globalcache.com/foo */",
"dependencies": Array [],
"dependencyOffsets": Array [],
}
`;
exports[`GlobalTransformCache fetches results 1`] = `
Array [
Object {
"code": "/* code from http://globalcache.com/3b3b861b6b80dd038c51262ca8b8b9f76e353ada-foo.js */",
"dependencies": Array [],
"dependencyOffsets": Array [],
},
Object {
"code": "/* code from http://globalcache.com/087b5bc3467f9a0670e49ce9118e6f6791aa12c6-bar.js */",
"dependencies": Array [],
"dependencyOffsets": Array [],
},
]
`;