next-boost
adds a cache layer to your SSR (Server-Side Rendering) applications. It was built originally for Next.js
and should work with any node.js http.Server
based application.
next-boost
achieves great performance by rendering webpages on worker_threads
while serving the cached on the main thread.
If you are familiar with Next.js
, next-boost
can be considered as an implementation of Incremental Static Regeneration which works with getServerSideProps
.
$ npm install next-boost --save
- Drop-in replacement for Next.js's production mode:
next start
- Greatly reducing the server TTFB (time-to-first-byte)
- Non-blocking main process for cache-serving and using
worker_threads
for SSR - Simultaneous requests with the first one rendered and the rest served from the cache
- Small footprint with less than 200 LOC
- Used in production with 300K pages cached
- As a database-disk-hybrid cache
- no memory capacity limit, and works great on cheap VPSs
- no need to add a cache layer server like varnish, nginx Cache and etc.
- great performance, and may even have better performace than pure file system cache
- portability on major platforms
After install the package, just change the start script from next start
to next-boost
. All next start
's command line arguments, like -p
for specifing the port, are compatible.
"scripts": {
...
"start": "next-boost", // previously `next start`
...
},
There's an example under examples/nodejs
, which works with a plain http.Server
.
To use it with express.js
and next.js
, please check examples/with-express
.
By using worker_threads
, the CPU-heavy SSR rendering will not blocking the main process from serving the cache.
Here are the comparision of using ApacheBench
on a blog post fetched from database. HTML prerendered and the db operation takes around 10~20ms. The page takes around 200ms for Next.js to render.
$ /usr/local/bin/ab -n 200 -c 8 http://127.0.0.1:3000/blog/posts/2020/3/postname
Not a scientific benchmark, but the improvements are visibly huge.
with next start
(data fetched with getServerSideProps
):
Document Length: 76424 bytes
Concurrency Level: 8
Time taken for tests: 41.855 seconds
Complete requests: 200
Failed requests: 0
Total transferred: 15325600 bytes
HTML transferred: 15284800 bytes
Requests per second: 4.78 [#/sec] (mean)
Time per request: 1674.185 [ms] (mean)
Time per request: 209.273 [ms] (mean, across all concurrent requests)
Transfer rate: 357.58 [Kbytes/sec] received
with the drop-in next-boost
cli:
Document Length: 78557 bytes
Concurrency Level: 8
Time taken for tests: 0.149 seconds
Complete requests: 200
Failed requests: 0
Total transferred: 15747600 bytes
HTML transferred: 15711400 bytes
Requests per second: 1340.48 [#/sec] (mean)
Time per request: 5.968 [ms] (mean)
Time per request: 0.746 [ms] (mean, across all concurrent requests)
Transfer rate: 103073.16 [Kbytes/sec] received
It even outperforms next.js's static generated page (getStaticProps
), handling 2~2.5x requests per seconds in my environment.
next-boost
implements a server-side cache in the manner of stale-while-revalidate. When an expired (stale
) page is accessed, the cache will be served and at the same time, a background process will fetch the latest version (revalidate
) of that page and save it to the cache.
There are 2 parameters to control the behavior of the cache:
ttl (time-to-live)
: Afterttl
, the cache will be revalidated. And a cached page'sttl
will be updated when a page is revalidated.tbd (time-before-deletion)
: When a page is not hit again inttl + tbd
seconds, it will be completely remove from cache.
By sending a GET with header x-cache-status:update
to the URL, the cache will be revalidated. And if the page doesn't exists anymore, the cache will be deleted.
curl -H x-cache-status:update https://the_server_name.com/path_a
If you want to delete mutiple pages at once, you can run SQL on the cache directly:
sqlite3 /cache_path/cache.db "update cache set ttl=0 where key like '%/url/a%';"
This will force all urls containing /url/a
to be revalidated when next time accessed.
Deleting cache_path
will remove all the caches.
By default, each page with different URLs will be cached separately. But in some cases you would like, /path_a?utm_source=twitter
to be served with the same contents of /path_a
. paramFilter
is for filtering the query parameters.
// in .next-boost.js
{
...
paramFilter: (p) => p !== 'utm_source'
}
If available, .next-boost.js
at project root will be used. If you use next-boost programmatically, the filename can be changed in options you passed to CachedHandler
.
tips: If you are using next-boost
cli with Next.js, you may want to use the config file.
And here's an example .next-boost.sample.js
in the repo.
By default, all URLs will be cached under the default rule .*
. You can change the rules programmatically or by .next-boost.js
:
module.exports = {
rules: [
{ regex: '^/blog.*', ttl: 300 },
],
}
Above: only caching pages with URL start with /blog
.
interface HandlerConfig {
filename?: string
quiet?: boolean
cache?: {
ttl?: number
tbd?: number
path?: string
}
rules?: Array<URLCacheRule>
paramFilter?: ParamFilter
}
interface URLCacheRule {
regex: string
ttl: number
}
type ParamFilter = (param: string) => boolean
Logging is enabled by default. If you use next-boost
programmatically, you can disable logs by passing the quiet
boolean flag as an option to CachedHandler
.
...
const cached = await CachedHandler(args, { quiet: true });
...
There's also a --quiet
flag if you are using the command line.
- For architecture with multiple load-balanced rendering servers, the benefit of using
next-boost
is limited. Until the url is hit on every backend server, it can still miss the cache. Use reverse proxy with cache support (nginx, varnish etc) for that. GET
andHEAD
requests only.worker_threads
is used and it is a node.js 12+ feature.
next-boost
works as an in-place replacement for next start
by using Next.js's custom server feature.
On the linked page above, you can see the following notice:
Before deciding to use a custom server please keep in mind that it should only be used when the integrated router of Next.js can't meet your app requirements. A custom server will remove important performance optimizations, like serverless functions and Automatic Static Optimization.
next-boost is meant to be used on cloud VPS or containers, so serverless function is not an issue here. As to Automatic Static Optimization
, because we are not doing any app.render
here, it still works, as perfect as always.
Here's the article about when not to use SQLite. And for next-boost's main purpuse: super faster SSR on low-cost VPSs, as far as I know, it is the best choice.
MIT. Copyright 2020 Rakuraku Jyo.