perf: add string flattening to reduce retained memory (node only)

add a memory footprint benchmark
harttle · Oct 10, 2019 · 3ad512c · 3ad512c
1 parent 7bae5bc
commit 3ad512c
Show file tree

Hide file tree

Showing 7 changed files with 102 additions and 6 deletions.
diff --git a/benchmark/index.ts b/benchmark/index.ts
@@ -2,12 +2,14 @@ import { output } from './output'
 import { tag } from './tag'
 import { demo } from './demo'
 import { layout } from './layout'
+import { memory } from './memory'
 
 async function main () {
   await output()
   await tag()
   await demo()
   await layout()
+  await memory()
 }
 
 main()
diff --git a/benchmark/memory.ts b/benchmark/memory.ts
@@ -0,0 +1,32 @@
+import { join } from 'path'
+import { readFileSync } from 'fs'
+import { getHeapStatistics } from 'v8'
+import { Liquid } from '../src/liquid'
+
+const engineOptions = {
+  root: __dirname,
+  extname: '.liquid'
+}
+
+const engine = new Liquid(engineOptions)
+
+const templateString = readFileSync(join(__dirname, 'templates/lorem-html.liquid'), 'utf8')
+const templateSizeKb = templateString.length / 1024
+
+const NB_INSTANCES_TO_RETAIN = 250
+
+export function memory () {
+  console.log('--- memory ---')
+  const initialUsedHeapSize = getHeapStatistics().used_heap_size
+  const templates = []
+
+  for (let i = 0; i < NB_INSTANCES_TO_RETAIN; i++) {
+    templates.push(engine.parse(templateString))
+  }
+
+  const finalUsedHeapSize = getHeapStatistics().used_heap_size
+  const heapDifference = finalUsedHeapSize - initialUsedHeapSize
+  const avgRetainedPerTemplate = (heapDifference / NB_INSTANCES_TO_RETAIN) / 1024
+
+  console.log(`retained memory for a ${templateSizeKb}KB template: ${avgRetainedPerTemplate}KB (${NB_INSTANCES_TO_RETAIN} samples)`)
+}
diff --git a/benchmark/templates/lorem-html.liquid b/benchmark/templates/lorem-html.liquid
@@ -0,0 +1,38 @@
+<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"
+        "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
+<html xmlns="http://www.w3.org/1999/xhtml"><head>
+
+    <meta http-equiv="content-type" content="text/html; charset=utf-8">
+    <!-- You may use a different charset if needed -->
+
+    <title>Latin Lipsum -  Courtesy generator.lorem-ipsum.info</title>
+</head>
+
+<body>
+<p>Lorem ipsum dolor sit amet, falli detraxit facilisis sit ex. Nec vidit nihil ut, vis utamur definitionem in. Ancillae phaedrum ut eos. Ne eam errem cotidieque.
+</p>
+<p>
+    Sit amet vitae referrentur ad, oportere necessitatibus vituperatoribus vim ne. Audiam quaeque nusquam ei eos, mel ex adhuc prompta complectitur. Decore voluptua cotidieque vel ei, veri novum persecuti te sea. Nisl novum sea cu. Dicta delicatissimi ut est. Et has eripuit impedit mediocrem.
+</p>
+<p>
+    Eu vim evertitur argumentum, pro legendos iudicabit voluptatibus ne. Ne alii nullam deserunt duo, vis no animal verterem. Ne quo veniam euripidis. Pri discere mentitum vituperatoribus ne. Meliore corrumpit vim at, has ea admodum convenire.
+</p>
+<p>
+    Ne mea alii detraxit electram, his ne epicurei gloriatur consetetur. Ex pro aeque omnes appareat, possit integre accommodare eu quo. Qui in porro molestiae, no sit equidem petentium patrioque, omnes voluptaria vim te. Odio quot rebum ad sed, ei sale reque inani sea.
+</p>
+<p>
+    At mea vidit petentium, mei lorem pertinacia signiferumque ne. Per eu dicam postea probatus, sea pericula accusamus omittantur ad. Id mei munere ullamcorper. Oportere ocurreret cu vel. Pri ne dicant soluta graecis, duis ponderum corrumpit sed eu, pro ei rebum vivendum.
+</p>
+<p>
+    Eum ea exerci sententiae constituam, ut lorem putant volutpat pro. Vix ex iudicabit salutatus principes, pri nisl erat ut, ne cum persius verterem phaedrum. Vix mundi vitae eu. Vel summo vocent albucius in. Qui omnis partem possim ex, wisi doctus efficiantur cu has. Aperiri denique ea pri, cum dissentias signiferumque eu. Per posse dissentiet necessitatibus id, omnis eleifend te mel.
+</p>
+<p>
+    Phaedrum erroribus adolescens ut nec, duo mutat viris id. In tale purto dolores sit, ei consul inermis est, his ea movet expetenda. Cibo scribentur id eam, per an quem iracundia, has erat zril solet eu. Falli fierent ne cum, eam placerat facilisis suscipiantur te. Id quo dicat tractatos definiebas, est copiosae splendide id.
+</p>
+<p>
+    Constituto neglegentur qui ne, eum ne putent phaedrum, no quod nominati consulatu per. Mei cu accusam principes interesset, vim fugit abhorreant eu, ludus feugait an sed. Clita expetendis definitionem has in, meis pericula no mea, soluta nemore cotidieque eos cu. Summo populo signiferumque pri eu, eum no labores democritum posidonium, pro ignota adipisci in. Tollit molestie id mei.
+</p>
+<p>
+    Ipsum impetus et eam, ei his epicuri instructior. Impedit ullamcorper vim an. Impedit ullamcorper vim an. Et eum senserit sententiae reprimique, quem detracto sententiae at sed, no inermis ocurreret rationibus sea pericula no.
+
+</body></html>
diff --git a/rollup.config.ts b/rollup.config.ts
@@ -71,6 +71,11 @@ const umd = {
       delimiters: ['', ''],
       './fs/node': './fs/browser'
     }),
+    replace({
+      include: './src/parser/tokenizer.ts',
+      delimiters: ['', ''],
+      './flatten/node': './flatten/browser'
+    }),
     typescript({
       tsconfigOverride: {
         include: [ 'src' ],
@@ -99,6 +104,11 @@ const min = {
       delimiters: ['', ''],
       './fs/node': './fs/browser'
     }),
+    replace({
+      include: './src/parser/tokenizer.ts',
+      delimiters: ['', ''],
+      './flatten/node': './flatten/browser'
+    }),
     typescript({
       tsconfigOverride: {
         include: [ 'src' ],

diff --git a/src/parser/flatten/browser.ts b/src/parser/flatten/browser.ts
@@ -0,0 +1,3 @@
+export function flatten (str: string) {
+  return str
+}
diff --git a/src/parser/flatten/node.ts b/src/parser/flatten/node.ts
@@ -0,0 +1,10 @@
+/**
+ * This function forces a string to be re-instantiated in memory using a flat representation instead of a graph
+ * of concatenated strings. This is an optimization to reduce the memory footprint of token string fragments after
+ * they are parsed.
+ * This optimization targets the V8 javascript engine and only works on Node.js.
+ * @param {string} str
+ */
+export function flatten (str: string) {
+  return Buffer.from(str).toString()
+}
diff --git a/src/parser/tokenizer.ts b/src/parser/tokenizer.ts
@@ -5,6 +5,7 @@ import { Token } from './token'
 import { OutputToken } from './output-token'
 import { TokenizationError } from '../util/error'
 import { NormalizedFullOptions, applyDefault } from '../liquid-options'
+import { flatten } from './flatten/node'
 
 enum ParseState { HTML, OUTPUT, TAG }
 
@@ -36,15 +37,15 @@ export class Tokenizer {
       }
       if (state === ParseState.HTML) {
         if (input.substr(p, outputDelimiterLeft.length) === outputDelimiterLeft) {
-          if (buffer) tokens.push(new HTMLToken(buffer, input, line, col, file))
+          if (buffer) tokens.push(new HTMLToken(flatten(buffer), input, line, col, file))
           buffer = outputDelimiterLeft
           line = curLine
           col = p - lineBegin + 1
           p += outputDelimiterLeft.length
           state = ParseState.OUTPUT
           continue
         } else if (input.substr(p, tagDelimiterLeft.length) === tagDelimiterLeft) {
-          if (buffer) tokens.push(new HTMLToken(buffer, input, line, col, file))
+          if (buffer) tokens.push(new HTMLToken(flatten(buffer), input, line, col, file))
           buffer = tagDelimiterLeft
           line = curLine
           col = p - lineBegin + 1
@@ -57,7 +58,7 @@ export class Tokenizer {
         input.substr(p, outputDelimiterRight.length) === outputDelimiterRight
       ) {
         buffer += outputDelimiterRight
-        tokens.push(new OutputToken(buffer, buffer.slice(outputDelimiterLeft.length, -outputDelimiterRight.length), input, line, col, this.options, file))
+        tokens.push(new OutputToken(flatten(buffer), buffer.slice(outputDelimiterLeft.length, -outputDelimiterRight.length), input, line, col, this.options, file))
         p += outputDelimiterRight.length
         buffer = ''
         line = curLine
@@ -66,7 +67,7 @@ export class Tokenizer {
         continue
       } else if (input.substr(p, tagDelimiterRight.length) === tagDelimiterRight) {
         buffer += tagDelimiterRight
-        tokens.push(new TagToken(buffer, buffer.slice(tagDelimiterLeft.length, -tagDelimiterRight.length), input, line, col, this.options, file))
+        tokens.push(new TagToken(flatten(buffer), buffer.slice(tagDelimiterLeft.length, -tagDelimiterRight.length), input, line, col, this.options, file))
         p += tagDelimiterRight.length
         buffer = ''
         line = curLine
@@ -81,10 +82,10 @@ export class Tokenizer {
       const str = buffer.length > 16 ? buffer.slice(0, 13) + '...' : buffer
       throw new TokenizationError(
         `${t} "${str}" not closed`,
-        new Token(buffer, input, line, col, file)
+        new Token(flatten(buffer), input, line, col, file)
       )
     }
-    if (buffer) tokens.push(new HTMLToken(buffer, input, line, col, file))
+    if (buffer) tokens.push(new HTMLToken(flatten(buffer), input, line, col, file))
 
     whiteSpaceCtrl(tokens, this.options)
     return tokens