feat(cache): plain object cache

[SukkaW]

Context: #158

Cache with plain object.

cc @dailyrandomphoto @seaoak

[coveralls]

Coverage increased (+0.2%) to 96.774% when pulling afdc7f8 on SukkaW:plain-object-cache into f80d6c6 on hexojs:master.

[SukkaW]

Benchmark result:

Node.js Version		Cold Process Time	Memory
8	Current master branch of Hexo	28.3s	587MB
8	Hexo with sukkaw/hexo-util#plain-object-cache	26.3s	589MB
10	Current master branch of Hexo	19.3s	596MB
10	Hexo with sukkaw/hexo-util#plain-object-cache	21.7s	600MB
12	Current master branch of Hexo	19.0s	582MB
12	Hexo with sukkaw/hexo-util#plain-object-cache	16.8s	580MB
13	Current master branch of Hexo	24.9s	587MB
13	Hexo with sukkaw/hexo-util#plain-object-cache	22.1s	590MB

[curbengh]

I notice repetitive codes, perhaps can refactor into a cache.js (a new utility)?

[SukkaW]

@curbengh I see no necessary.

I have bring up something like this:

// cache.js

'use strict';

module.exports = class Cache {
  constructor() {
    this.cache = {};
  }

  set(id, value) {
    this.cache[id] = value;
  }

  has(id) {
    return this.cache[id] != null;
  }

  get(id) {
    return this.cache[id];
  }

  flush() {
    this.cache = {};
  }
};

But the usage will end in something like:

const Cache = require('./cache');
const cache = new Cache();

if (cache.has(cacheId)) return cache.get(cacheId);
...
cache.set(cacheId, value);

Nothing changed...

[curbengh]

const Cache = require('./cache');
const cache = new Cache();

if (cache.has(cacheId)) return cache.get(cacheId);
...
cache.set(cacheId, value);

while it doesn't reduce the line, it does look more readable.

[SukkaW]

@curbengh I have bring up cache.js with set(), get(), has(), del(), flush() and apply() methods. cache.apply() has similar usage of fragment_cache. In fact we could utilize fragment_cache helper using cache.apply.

[SukkaW]

Performance with Cache():

Node.js Version		Cold Process Time	Memory
8	Current master branch of Hexo	25.2s	591MB
8	Hexo with sukkaw/hexo-util#plain-object-cache	20.0s	603MB
10	Current master branch of Hexo	18.7s	593MB
10	Hexo with sukkaw/hexo-util#plain-object-cache	22.9s	585MB
12	Current master branch of Hexo	18.4s	576MB
12	Hexo with sukkaw/hexo-util#plain-object-cache	17.2s	584MB
13	Current master branch of Hexo	18.8s	642MB
13	Hexo with sukkaw/hexo-util#plain-object-cache	16.9s	624MB

[curbengh]

lgtm

[seaoak]

I tasted this patch in a large blog (dummy 3000 posts).

The result shows this patch is 13% faster than master.
(no regression!)

---

Details:

Measure elapsed time of hexo generate --debug.
For each condition, measure 10 times and calculate the average.
Before each measurement, do hexo clean and rm debug.log.
Set CPU affinity 0x00010000 and priority HIGH for hexo command (using START command).

Result

Elapsed time:

Branch	Average	Spread (worst - best)
master	77.34sec	1.67sec
SukkaW	67.15sec (-13.18%)	2.44sec

Source code:

• master => 621c9e3
• SukkaW => SukkaW/hexo-util@afdc7f8

Environment:

• OS: Windows 10 64bit
• CPU: Ryzen Threadripper 2950X (16core/32thread, 3.5GHz)
• MEM: 32GB (8GB x4) DDR4 (3600MHz)
• SSD: Samsung EVO970 (1TB)
• Node v12.14.0

[seaoak]

Measure elapsed time of hexo generate --debug.

Memory usage is about 2.3GB as the worst case in both conditions.
(measured with tasklist command)