All Projects → kornelski → Http Cache Semantics

kornelski / Http Cache Semantics

Licence: bsd-2-clause
RFC 7234 in JavaScript. Parses HTTP headers to correctly compute cacheability of responses, even in complex cases

Programming Languages

javascript
184084 projects - #8 most used programming language

Projects that are alternatives of or similar to Http Cache Semantics

Cacheable Request
Wrap native HTTP requests with RFC compliant cache support
Stars: ✭ 218 (+28.99%)
Mutual labels:  rfc, cache
Http Core
Core HTTP Specifications
Stars: ✭ 165 (-2.37%)
Mutual labels:  rfc
Android Cache Fix Gradle Plugin
Gradle plugin to fix Android caching problems
Stars: ✭ 150 (-11.24%)
Mutual labels:  cache
Dataloader Php
DataLoaderPhp is a generic utility to be used as part of your application's data fetching layer to provide a simplified and consistent API over various remote data sources such as databases or web services via batching and caching.
Stars: ✭ 160 (-5.33%)
Mutual labels:  cache
Rfcs
This repository contains proposals, standards and documentations related to Nervos Network.
Stars: ✭ 154 (-8.88%)
Mutual labels:  rfc
Cachemanager
CacheManager is an open source caching abstraction layer for .NET written in C#. It supports various cache providers and implements many advanced features.
Stars: ✭ 2,049 (+1112.43%)
Mutual labels:  cache
Strapi Middleware Cache
🔌 A cache middleware for https://strapi.io
Stars: ✭ 146 (-13.61%)
Mutual labels:  cache
Holster
A place to keep useful golang functions and small libraries
Stars: ✭ 166 (-1.78%)
Mutual labels:  cache
Q Municate Ios
Q-municate iOS repository
Stars: ✭ 164 (-2.96%)
Mutual labels:  cache
Action Docker Layer Caching
🐳 Enable Docker layer caching in GitHub Actions
Stars: ✭ 160 (-5.33%)
Mutual labels:  cache
Study
全栈工程师学习笔记;Spring登录、shiro登录、CAS单点登录和Spring boot oauth2单点登录;Spring data cache 缓存,支持Redis和EHcahce; web安全,常见web安全漏洞以及解决思路;常规组件,比如redis、mq等;quartz定时任务,支持持久化数据库,动态维护启动暂停关闭;docker基本用法,常用image镜像使用,Docker-MySQL、docker-Postgres、Docker-nginx、Docker-nexus、Docker-Redis、Docker-RabbitMQ、Docker-zookeeper、Docker-es、Docker-zipkin、Docker-ELK等;mybatis实践、spring实践、spring boot实践等常用集成;基于redis的分布式锁;基于shared-jdbc的分库分表,支持原生jdbc和Spring Boot Mybatis
Stars: ✭ 159 (-5.92%)
Mutual labels:  cache
Layercache
Caching made simple for Android and Java.
Stars: ✭ 155 (-8.28%)
Mutual labels:  cache
Haproxy
HAProxy Load Balancer's development branch (mirror of git.haproxy.org)
Stars: ✭ 2,463 (+1357.4%)
Mutual labels:  cache
Libcache
A Lightweight in-memory key:value cache library for Go.
Stars: ✭ 152 (-10.06%)
Mutual labels:  cache
Springbootlearning
《Spring Boot教程》源码
Stars: ✭ 2,065 (+1121.89%)
Mutual labels:  cache
Jstarcraft Core
目标是提供一个通用的Java核心编程框架,作为搭建其它框架或者项目的基础. 让相关领域的研发人员能够专注高层设计而不用关注底层实现. 涵盖了缓存,存储,编解码,资源,脚本,监控,通讯,事件,事务9个方面.
Stars: ✭ 150 (-11.24%)
Mutual labels:  cache
Cacheable Response
An HTTP compliant route path middleware for serving cache response with invalidation support.
Stars: ✭ 157 (-7.1%)
Mutual labels:  cache
Lru Cache
💫 A feature complete LRU cache implementation in C++
Stars: ✭ 162 (-4.14%)
Mutual labels:  cache
Rrule
JavaScript library for working with recurrence rules for calendar dates as defined in the iCalendar RFC and more.
Stars: ✭ 2,249 (+1230.77%)
Mutual labels:  rfc
Hitchcock
The Master of Suspense 🍿
Stars: ✭ 167 (-1.18%)
Mutual labels:  cache

Can I cache this? Build Status

CachePolicy tells when responses can be reused from a cache, taking into account HTTP RFC 7234 rules for user agents and shared caches. It also implements RFC 5861, implementing stale-if-error and stale-while-revalidate. It's aware of many tricky details such as the Vary header, proxy revalidation, and authenticated responses.

Usage

Cacheability of an HTTP response depends on how it was requested, so both request and response are required to create the policy.

const policy = new CachePolicy(request, response, options);

if (!policy.storable()) {
    // throw the response away, it's not usable at all
    return;
}

// Cache the data AND the policy object in your cache
// (this is pseudocode, roll your own cache (lru-cache package works))
letsPretendThisIsSomeCache.set(
    request.url,
    { policy, response },
    policy.timeToLive()
);
// And later, when you receive a new request:
const { policy, response } = letsPretendThisIsSomeCache.get(newRequest.url);

// It's not enough that it exists in the cache, it has to match the new request, too:
if (policy && policy.satisfiesWithoutRevalidation(newRequest)) {
    // OK, the previous response can be used to respond to the `newRequest`.
    // Response headers have to be updated, e.g. to add Age and remove uncacheable headers.
    response.headers = policy.responseHeaders();
    return response;
}

It may be surprising, but it's not enough for an HTTP response to be fresh to satisfy a request. It may need to match request headers specified in Vary. Even a matching fresh response may still not be usable if the new request restricted cacheability, etc.

The key method is satisfiesWithoutRevalidation(newRequest), which checks whether the newRequest is compatible with the original request and whether all caching conditions are met.

Constructor options

Request and response must have a headers property with all header names in lower case. url, status and method are optional (defaults are any URL, status 200, and GET method).

const request = {
    url: '/',
    method: 'GET',
    headers: {
        accept: '*/*',
    },
};

const response = {
    status: 200,
    headers: {
        'cache-control': 'public, max-age=7234',
    },
};

const options = {
    shared: true,
    cacheHeuristic: 0.1,
    immutableMinTimeToLive: 24 * 3600 * 1000, // 24h
    ignoreCargoCult: false,
};

If options.shared is true (default), then the response is evaluated from a perspective of a shared cache (i.e. private is not cacheable and s-maxage is respected). If options.shared is false, then the response is evaluated from a perspective of a single-user cache (i.e. private is cacheable and s-maxage is ignored). shared: true is recommended for HTTP clients.

options.cacheHeuristic is a fraction of response's age that is used as a fallback cache duration. The default is 0.1 (10%), e.g. if a file hasn't been modified for 100 days, it'll be cached for 100*0.1 = 10 days.

options.immutableMinTimeToLive is a number of milliseconds to assume as the default time to cache responses with Cache-Control: immutable. Note that per RFC these can become stale, so max-age still overrides the default.

If options.ignoreCargoCult is true, common anti-cache directives will be completely ignored if the non-standard pre-check and post-check directives are present. These two useless directives are most commonly found in bad StackOverflow answers and PHP's "session limiter" defaults.

storable()

Returns true if the response can be stored in a cache. If it's false then you MUST NOT store either the request or the response.

satisfiesWithoutRevalidation(newRequest)

This is the most important method. Use this method to check whether the cached response is still fresh in the context of the new request.

If it returns true, then the given request matches the original response this cache policy has been created with, and the response can be reused without contacting the server. Note that the old response can't be returned without being updated, see responseHeaders().

If it returns false, then the response may not be matching at all (e.g. it's for a different URL or method), or may require to be refreshed first (see revalidationHeaders()).

responseHeaders()

Returns updated, filtered set of response headers to return to clients receiving the cached response. This function is necessary, because proxies MUST always remove hop-by-hop headers (such as TE and Connection) and update response's Age to avoid doubling cache time.

cachedResponse.headers = cachePolicy.responseHeaders(cachedResponse);

timeToLive()

Returns approximate time in milliseconds until the response becomes stale (i.e. not fresh).

After that time (when timeToLive() <= 0) the response might not be usable without revalidation. However, there are exceptions, e.g. a client can explicitly allow stale responses, so always check with satisfiesWithoutRevalidation(). stale-if-error and stale-while-revalidate extend the time to live of the cache, that can still be used if stale.

toObject()/fromObject(json)

Chances are you'll want to store the CachePolicy object along with the cached response. obj = policy.toObject() gives a plain JSON-serializable object. policy = CachePolicy.fromObject(obj) creates an instance from it.

Refreshing stale cache (revalidation)

When a cached response has expired, it can be made fresh again by making a request to the origin server. The server may respond with status 304 (Not Modified) without sending the response body again, saving bandwidth.

The following methods help perform the update efficiently and correctly.

revalidationHeaders(newRequest)

Returns updated, filtered set of request headers to send to the origin server to check if the cached response can be reused. These headers allow the origin server to return status 304 indicating the response is still fresh. All headers unrelated to caching are passed through as-is.

Use this method when updating cache from the origin server.

updateRequest.headers = cachePolicy.revalidationHeaders(updateRequest);

revalidatedPolicy(revalidationRequest, revalidationResponse)

Use this method to update the cache after receiving a new response from the origin server. It returns an object with two keys:

  • policy — A new CachePolicy with HTTP headers updated from revalidationResponse. You can always replace the old cached CachePolicy with the new one.
  • modified — Boolean indicating whether the response body has changed.
    • If false, then a valid 304 Not Modified response has been received, and you can reuse the old cached response body. This is also affected by stale-if-error.
    • If true, you should use new response's body (if present), or make another request to the origin server without any conditional headers (i.e. don't use revalidationHeaders() this time) to get the new resource.
// When serving requests from cache:
const { oldPolicy, oldResponse } = letsPretendThisIsSomeCache.get(
    newRequest.url
);

if (!oldPolicy.satisfiesWithoutRevalidation(newRequest)) {
    // Change the request to ask the origin server if the cached response can be used
    newRequest.headers = oldPolicy.revalidationHeaders(newRequest);

    // Send request to the origin server. The server may respond with status 304
    const newResponse = await makeRequest(newRequest);

    // Create updated policy and combined response from the old and new data
    const { policy, modified } = oldPolicy.revalidatedPolicy(
        newRequest,
        newResponse
    );
    const response = modified ? newResponse : oldResponse;

    // Update the cache with the newer/fresher response
    letsPretendThisIsSomeCache.set(
        newRequest.url,
        { policy, response },
        policy.timeToLive()
    );

    // And proceed returning cached response as usual
    response.headers = policy.responseHeaders();
    return response;
}

Yo, FRESH

satisfiesWithoutRevalidation

Used by

Implemented

  • Cache-Control response header with all the quirks.
  • Expires with check for bad clocks.
  • Pragma response header.
  • Age response header.
  • Vary response header.
  • Default cacheability of statuses and methods.
  • Requests for stale data.
  • Filtering of hop-by-hop headers.
  • Basic revalidation request
  • stale-if-error

Unimplemented

  • Merging of range requests, If-Range (but correctly supports them as non-cacheable)
  • Revalidation of multiple representations

Trusting server Date

Per the RFC, the cache should take into account the time between server-supplied Date and the time it received the response. The RFC-mandated behavior creates two problems:

  • Servers with incorrectly set timezone may add several hours to cache age (or more, if the clock is completely wrong).
  • Even reasonably correct clocks may be off by a couple of seconds, breaking max-age=1 trick (which is useful for reverse proxies on high-traffic servers).

Previous versions of this library had an option to ignore the server date if it was "too inaccurate". To support the max-age=1 trick the library also has to ignore dates that pretty accurate. There's no point of having an option to trust dates that are only a bit inaccurate, so this library won't trust any server dates. max-age will be interpreted from the time the response has been received, not from when it has been sent. This will affect only RFC 1149 networks.

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].