1689863 - Huge memory usage of zone / strings on main process with multi-process browser toolbox open

Status: NEW

(DevTools :: General, defect, P3)

Description

[Henrik Skupin [:whimboo][⌚️UTC+2] (away 10/03 - 10/13)]

Attachment: memory-report.json.gz

Mozilla/5.0 (Macintosh; Intel Mac OS X 10.16; rv:87.0) Gecko/20100101 Firefox/87.0 ID:20210127093943

In a test profile that is open for a while and I'm using to test a temporary webextension I had a couple of tabs open. Then I noticed a quite large memory usage of the main process (~1.3GB). As such I triggered a memory report (find it as attached). A lot of the memory is actually part of the zone / strings section. Closing all tabs and triggering a minimize memory action doesn't change anything. The strings as used by some websites stay around and accumulate the main process' memory. Interestingly the memory usage even grows by measuring memory via about:memory, which means it's now at 2GB.

The high memory use for window(chrome://browser/content/webext-panels.xhtml) might warrant a fix in my webextension, which doesn't handle unloading yet. But none of the strings are mentioned above seem to be related to the web extension.

Andrew, do you have an idea?

Comment 1

[Henrik Skupin [:whimboo][⌚️UTC+2] (away 10/03 - 10/13)]

Here an excerpt of the zone / strings:

Main Process (pid 54927)
Explicit Allocations

2,034.63 MB (100.0%) -- explicit
├────818.29 MB (40.22%) -- js-non-window
│    ├──788.47 MB (38.75%) -- zones
│    │  ├──774.45 MB (38.06%) -- zone(0x10e37c000)
│    │  │  ├──457.33 MB (22.48%) -- strings
│    │  │  │  ├──436.31 MB (21.44%) -- (944 tiny)
│    │  │  │  │  ├───16.00 MB (00.79%) ++ string(length=1215037, copies=4, "!function(e){function t(t){for(var n,o,a=t[0],i=t[1],s=0,l=[];s<a.length;s++)o=a[s],Object.prototype.hasOwnProperty.call(r,o)&&r[o]&&l.push(r[o][0]),r[o]=0;for(n in i)Object.prototype.hasOwnProperty.call(i,n)&&(e[n]=i[n]);for(u&&u(t);l.length;)l.shift()()}var n={},r={0:0};var o={};var a={273:function(){return{}}};function i(t){if(n[t])return n[t].exports;var r=n[t]={i:t,l:!1,exports:{}};return e[t].call(r.exports,r,r.exports,i),r.l=!0,r.exports}i.e=function(e){var t=[],n=r[e];if(0!==n)if(n)t.push(n[2]);else{var s=new Promise((function(t,o){n=r[e]=[t,o]}));t.push(n[2]=s);var l,c=document.createElement("script");c.charset="utf-8",c.timeout=120,i.nc&&c.setAttribute("nonce",i.nc),c.src=function(e){return i.p+""+e+".820a88dfe935a6797ae9.bundle.js"}(e);var u=new Error;l=function(t){c.onerror=c.onload=null,clearTimeout(d);var n=r[e];if(0!==n){if(n){var o=t&&("load"===t.type?"missing":t.type),a=t&&t.target&&t.target.src;u.message="Loading chunk "+e+" failed.//n("+o+": "+a+")",u.name="ChunkLoadError",u.type=o,u.reque" (truncated))
│    │  │  │  │  ├────8.00 MB (00.39%) ++ string(length=1048576, copies=8, "!function(e){function t(t){for(var n,o,a=t[0],i=t[1],s=0,l=[];s<a.length;s++)o=a[s],Object.prototype.hasOwnProperty.call(r,o)&&r[o]&&l.push(r[o][0]),r[o]=0;for(n in i)Object.prototype.hasOwnProperty.call(i,n)&&(e[n]=i[n]);for(u&&u(t);l.length;)l.shift()()}var n={},r={0:0};var o={};var a={273:function(){return{}}};function i(t){if(n[t])return n[t].exports;var r=n[t]={i:t,l:!1,exports:{}};return e[t].call(r.exports,r,r.exports,i),r.l=!0,r.exports}i.e=function(e){var t=[],n=r[e];if(0!==n)if(n)t.push(n[2]);else{var s=new Promise((function(t,o){n=r[e]=[t,o]}));t.push(n[2]=s);var l,c=document.createElement("script");c.charset="utf-8",c.timeout=120,i.nc&&c.setAttribute("nonce",i.nc),c.src=function(e){return i.p+""+e+".820a88dfe935a6797ae9.bundle.js"}(e);var u=new Error;l=function(t){c.onerror=c.onload=null,clearTimeout(d);var n=r[e];if(0!==n){if(n){var o=t&&("load"===t.type?"missing":t.type),a=t&&t.target&&t.target.src;u.message="Loading chunk "+e+" failed.//n("+o+": "+a+")",u.name="ChunkLoadError",u.type=o,u.reque" (truncated))
│    │  │  │  │  ├────4.22 MB (00.21%) ++ string(length=733900, copies=6, "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" (truncated))
│    │  │  │  │  ├────4.00 MB (00.20%) ++ string(length=331866, copies=6, "// zee.js: zlib compiled to js/n/nvar Zee = (function() {/n/nfunction adjustedTotalMemory(targetVal,stackSize){var totalMemory=64*1024;while(totalMemory<targetVal||totalMemory<2*stackSize){if(totalMemory<16*1024*1024){totalMemory*=2}else{totalMemory+=16*1024*1024}}return totalMemory}function createGzModule(heapMemorySize){var Module={noExitRuntime:true,noImageDecoding:true,TOTAL_STACK:262144,TOTAL_MEMORY:adjustedTotalMemory(heapMemorySize,262144)};var Module;if(!Module)Module=(typeof Module!=="undefined"?Module:null)||{};var moduleOverrides={};for(var key in Module){if(Module.hasOwnProperty(key)){moduleOverrides[key]=Module[key]}}var ENVIRONMENT_IS_WEB=typeof window==="object";var ENVIRONMENT_IS_WORKER=typeof importScripts==="function";var ENVIRONMENT_IS_NODE=typeof process==="object"&&typeof require==="function"&&!ENVIRONMENT_IS_WEB&&!ENVIRONMENT_IS_WORKER;var ENVIRONMENT_IS_SHELL=!ENVIRONMENT_IS_WEB&&!ENVIRONMENT_IS_NODE&&!ENVIRONMENT_IS_WORKER;if(ENVIRONMENT_IS_NODE){if(!Module["print"])Module["print"]=fu" (truncated))
│    │  │  │  │  ├────4.00 MB (00.20%) ++ string(length=579147, copies=2, "<!DOCTYPE html>/n<html class="ima-sdk-frame">/n  <head>/n    <title></title>/n    <style type="text/css">.videoAdUiActionCta{position:absolute;bottom:30px;box-sizing:border-box;padding:5px;max-width:75%}.uac-android-on-3p-instream.videoAdUiActionCta,.uac-ios-on-3p-instream.videoAdUiActionCta{background:linear-gradient(to top,rgba(0,0,0,0),rgba(0,0,0,.4) 100%);bottom:unset;max-width:100%;padding:12px;top:0;width:100%}.newProgressBarVariant .videoAdUiActionCta{bottom:46px}.videoAdUiActionCtaInset{font-family:arial,sans-serif;font-weight:500;font-size:16px;background:white;display:flex;flex-direction:row;border-radius:8px;box-shadow:0 1px 3px rgba(60,64,67,0.5);overflow:hidden;height:50px;pointer-events:auto}.uac-android-on-3p-instream .videoAdUiActionCtaInset,.uac-ios-on-3p-instream .videoAdUiActionCtaInset{background:transparent;box-shadow:none}.videoAdUiActionCtaIconContainer{min-width:50px;z-index:1000}.uac-android-on-3p-instream .videoAdUiActionCtaIconContainer,.uac-ios-on-3p-instream .videoAdUiActionCtaIc" (truncated))
│    │  │  │  │  ├────3.38 MB (00.17%) ++ string(length=7092, copies=432, "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" (truncated))
│    │  │  │  │  ├────3.00 MB (00.15%) ++ string(length=285054, copies=3, "/*! aniview version '6.1.2+25'/nAll rights reserved to ANIVIEW LTD 2021 *//n!function(){var t,e,n={676:function(t,e,n){"use strict";function i(t,e){(null==e||e>t.length)&&(e=t.length);for(var n=0,i=new Array(e);n<e;n++)i[n]=t[n];return i}n.d(e,{Z:function(){return i}})},9968:function(t,e,n){"use strict";function i(t){if(Array.isArray(t))return t}n.d(e,{Z:function(){return i}})},6156:function(t,e,n){"use strict";function i(t,e,n){return e in t?Object.defineProperty(t,e,{value:n,enumerable:!0,configurable:!0,writable:!0}):t[e]=n,t}n.d(e,{Z:function(){return i}})},692:function(t,e,n){"use strict";function i(t,e){if("undefined"!=typeof Symbol&&Symbol.iterator in Object(t)){var n=[],i=!0,r=!1,o=void 0;try{for(var a,s=t[Symbol.iterator]();!(i=(a=s.next()).done)&&(n.push(a.value),!e||n.length!==e);i=!0);}catch(t){r=!0,o=t}finally{try{i||null==s.return||s.return()}finally{if(r)throw o}}return n}}n.d(e,{Z:function(){return i}})},8970:function(t,e,n){"use strict";function i(){throw new TypeError("Invalid attempt to de" (truncated))
│    │  │  │  │  ├────3.00 MB (00.15%) ++ string(length=330500, copies=3, "/* prebid.js v3.22.0-pre/nUpdated : 2020-11-16 *//n!function(u){var s=window.pbjsChunk;window.pbjsChunk=function(e,t,n){for(var r,o,i,a=0,c=[];a<e.length;a++)o=e[a],d[o]&&c.push(d[o][0]),d[o]=0;for(r in t)Object.prototype.hasOwnProperty.call(t,r)&&(u[r]=t[r]);for(s&&s(e,t,n);c.length;)c.shift()();if(n)for(a=0;a<n.length;a++)i=f(f.s=n[a]);return i};var n={},d={306:0};function f(e){if(n[e])return n[e].exports;var t=n[e]={i:e,l:!1,exports:{}};return u[e].call(t.exports,t,t.exports,f),t.l=!0,t.exports}f.m=u,f.c=n,f.d=function(e,t,n){f.o(e,t)||Object.defineProperty(e,t,{configurable:!1,enumerable:!0,get:n})},f.n=function(e){var t=e&&e.__esModule?function(){return e.default}:function(){return e};return f.d(t,"a",t),t},f.o=function(e,t){return Object.prototype.hasOwnProperty.call(e,t)},f.p="",f.oe=function(e){throw console.error(e),e},f(f.s=817)}([function(e,t,n){"use strict";Object.defineProperty(t,"__esModule",{value:!0}),n.d(t,"internal",function(){return k}),n.d(t,"bind",function(){return N}),t.getUniqueIdentif" (truncated))

Comment 2

[Andrew McCreight [:mccr8]]

That's odd. I'm not sure what kind of code would cause those strings to hang around. Are you doing anything with devtools? That can cause weird leakiness.

Short of that, the steps here would be to get a GC log for the main process, grep it to find the address of one of those strings (because the GC log contains part of the value of the string), then run find_roots.py on the GC log to see what is holding the string alive.

Comment 3

[Henrik Skupin [:whimboo][⌚️UTC+2] (away 10/03 - 10/13)]

(In reply to Andrew McCreight [:mccr8] from comment #2)

That's odd. I'm not sure what kind of code would cause those strings to hang around. Are you doing anything with devtools? That can cause weird leakiness.

Hm, for the given Firefox process I have a browser toolbox open. But that runs in it's own instance. Nevertheless it could mean that the devtools server code as run in the Firefox process leaks memory? Is that a known problem? Nicholas or Alexandre?

Short of that, the steps here would be to get a GC log for the main process, grep it to find the address of one of those strings (because the GC log contains part of the value of the string), then run find_roots.py on the GC log to see what is holding the string alive.

I can have a look at this tomorrow. Looks like it will take a bit of time to get some results:
https://developer.mozilla.org/en-US/docs/Mozilla/Performance/GC_and_CC_logs#Analyzing_GC_and_CC_logs

Comment 4

[Henrik Skupin [:whimboo][⌚️UTC+2] (away 10/03 - 10/13)]

I closed the browser toolbox and indeed all the extra memory as mentioned above gets cleaned-up when minimizing the memory. So it's clearly a memory footprint issue with devtools.

Comment 5

[Henrik Skupin [:whimboo][⌚️UTC+2] (away 10/03 - 10/13)]

(In reply to Henrik Skupin (:whimboo) [⌚️UTC+2] (away 01/15 - 01/31) from comment #3)

Hm, for the given Firefox process I have a browser toolbox open.

To be clear this is about the new multi-process browser toolbox.

Comment 6

[Henrik Skupin [:whimboo][⌚️UTC+2] (away 10/03 - 10/13)]

Note that cleaning the console from all messages doesn't help. The string instances still stay around also after an additional GC.

Comment 7

[Henrik Skupin [:whimboo][⌚️UTC+2] (away 10/03 - 10/13)]

This is actually pretty easy to reproduce. Here are the steps:

1. Start Firefox and open the multi-process browser toolbox
2. Open about:memory in a tab and create an initial measurement
3. Open several tabs and start navigating through various pages across these tabs
4. Switch back to about:memory and create another measurement

After step 4 you will see that zones / strings has been grown, and contains strings from web pages that are no longer open. Continue to navigate if needed to see that it infinitely grows. The only chance to get rid of all these strings is to actually close the multi-process browser toolbox.

I actually won't find the time to do any GC analysis. Maybe someone else working on devtools can actually do that, or even knows what might go wrong here.

Comment 8

[Julian Descottes [:jdescottes]]

Also tentatively blocking the MBT enabling bug. Leaks are a potential cause for the overall slowness related to MBT.

Comment 9

[Nicolas Chevobbe [:nchevobbe]]

Not sure what's going on, but the strings that are leaking look like sources and sourcemaps.
On the MBT, I don't know when/if we clear the sources actor. I would guess when the associated target is destroyed, but it would be nice to check what we're doing about that.

Comment 10

[Alexandre Poirot [:ochameau]]

Memory usage of browser console and toolbox should have been drastically improved, especially when running in parent-process only and/or with keeping network inspection off by default.

Comment 11

[Henrik Skupin [:whimboo][⌚️UTC+2] (away 10/03 - 10/13)]

I think that I saw that again as part of bug 842979 comment 21. I'll keep an eye open for that particular issue.