515354 - Implement "about:memory"

Status: VERIFIED FIXED

(Toolkit :: about:memory, enhancement)

Description

[RNicoletto]

Attachment: Google Chrome 3.x "about:memory" page

One of the greatest features of Google Chrome is the "about:memory" page and I would like to see something similar in Mozilla Firefox.

Refer to bug 515352 description for more details about this enhancement request.

Just for reference: Memory Usage Backgrounder - Chromium Memory Usage (http://dev.chromium.org/memory-usage-backgrounder)

Comment 1

[(mostly gone) XtC4UaLL [:xtc4uall]]

related Bugs:
Bug 392351 and dependencies
Bug 400120
Bug 453420 comment 3

Comment 2

[RNicoletto]

Added another bug dependecy:
Bug 167035

Comment 4

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Attachment: export jemalloc.h and jemalloc_stats symbol

This only changes the VC9 jemalloc patch; we need something else that will change the VC8 one.  This is a requirement for the following patch.

Comment 5

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Attachment: about:memory

Simple about:memory implementation.  Displays a page that basically dumps all the reporters when refreshed.  Should be good enough to build on top of.

We'll have to figure out localization; I'm thinking that we support a properties file where we look up the reporter path ("malloc/committed"), and if a string for that is present, we use that -- otherwise we use the path itself.

Comment 6

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Attachment: screenshot

Screenshot of what this looks like.  Note that jemalloc is giving some bogus numbers currently.

Comment 7

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Comment on attachment 400872 [details] [diff] [review]
export jemalloc.h and jemalloc_stats symbol

Ted, is exporting jemalloc.h in this way ok?  Seems to work fine.

Er, I have no idea why this patch removed the header from the _sample_.rc diff.  Ignore that hunk.

Comment 8

[Mike Connor [:mconnor]]

Comment on attachment 400875 [details] [diff] [review]
about:memory


>diff --git a/toolkit/components/aboutmemory/content/aboutMemory.js b/toolkit/components/aboutmemory/content/aboutMemory.js

code looks fine, there's random tab chars in places (noticed since diff indentation was wacky)

>+function doLoad()
>+{
>+    netscape.security.PrivilegeManager.enablePrivilege("UniversalXPConnect");

as noted, this isn't needed anymore.

this looks good overall, woo!

r=mconor

Comment 9

[(not currently active) Ted Mielczarek]

Comment on attachment 400872 [details] [diff] [review]
export jemalloc.h and jemalloc_stats symbol

Seems sane. Remember to put the crt patch license header back in before landing. Also, I'll get you a VC8 version of this soon, I promise.

Comment 10

[Mike Beltzner [:beltzner, not reading bugmail]]

Any chance of getting this as an XPI for 3.6? I'm almost willing to land this in all locales with en-US strings, really.

Comment 11

[(not currently active) Ted Mielczarek]

Attachment: VC8 CRT patch patch

This is an equivalent patch for the VC8 CRT patch. There's some extra changes (in there that I don't fully understand (but look harmless, and build fine), so if you want to be extra paranoid, I think you can actually just hand-edit crtvc8sp1.diff, and add that jemalloc_stats line directly below malloc_usable_size.

Comment 12

[(not currently active) Ted Mielczarek]

Might be stray whitespace junk from previous fiddling, I would recommend just hand-editing the VC8 patch. You can probably tryserver it to sanity check yourself.

Comment 13

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

So this all works great on Windows, but on Linux the libxul link fails -- it can't find jemalloc_stats.  Indeed, we don't actually link jemalloc into libxul, but instead link it directly to the browser app.  Is there any reason we do this?  Can we just make it part of libxul?

Comment 14

[Benjamin Smedberg]

Yes, the reason is that if you make jemalloc part of libxul you totally screw over all the embedders who will end up with mixed allocators (you end up replacing malloc/free midway through running their app).

Comment 15

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Checked in, disabled on Linux due to linkage issues.  On Mac, it reports the OSX allocator's zone 0 stats.

http://hg.mozilla.org/mozilla-central/rev/ed8c714d33ad

Comment 16

[Doug Turner (:dougt)]

vlad, can we get this for 192?

Comment 17

[u88484]

What are the values measure in? Bytes, MB, K, KB, GB?

Comment 18

[Jo Hermans]

Some of these figures are really bogus. I watched the 'memory mapped' counter going to 0, and then it became negative :

Memory mapped:	4,284,481,536
Memory in use:	100,894,418
	
malloc/allocated 100,903,002
malloc/mapped 4,284,481,536
malloc/committed 39,411,712
malloc/dirty 2,895,872

Comment 19

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

The bogus values are known -- there's a jemalloc stats bug, but it'll be fixed separately... bug 515556.

Comment 20

[PikeUK]

There's a typo in the tooltip text, commited instead of committed.

Comment 21

[Tyler Downer [He/Him]]

Committed is the EN-US spelling I believe. For EN-UK, it is probably commited.

Comment 22

[Reed Loden [:reed]]

(In reply to comment #20)
> There's a typo in the tooltip text, commited instead of committed.

Fixed -- http://hg.mozilla.org/mozilla-central/rev/8afae2d77f31

Comment 23

[Mike Beltzner [:beltzner, not reading bugmail]]

Vlad: any chance you can help describe in human terms what the values are so that the SUMO and MDC guys can write this up?

Comment 24

[Justin Dolske [:Dolske]]

Pushed to 192:
http://hg.mozilla.org/releases/mozilla-1.9.2/rev/47c8830c87ec
http://hg.mozilla.org/releases/mozilla-1.9.2/rev/27230e3ceea2

Comment 25

[Damon Sicore (:damons)]

Is this really resolved/fixed if it isn't implemented on Linux?

Comment 26

[Biju]

An enhancement request bug 521522 
-  about:memory should automatically refresh every 5 sec (configurable)

Comment 27

[Mike Beltzner [:beltzner, not reading bugmail]]

(In reply to comment #25)
> Is this really resolved/fixed if it isn't implemented on Linux?

A follow up should be levied against implementing it on all other platforms: Linux, Maemo, WinCE

Comment 28

[Chris Ilias [:cilias]]

I think this would be an ideal addition to the support article <https://support.mozilla.com/en-US/kb/High+memory+usage>, but when I look at about:memory, I have no idea how to use this feature. Can someone tell me?

Comment 29

[Eric Shepherd [:sheppy]]

Added a mention of this here:

https://developer.mozilla.org/en/Debugging_memory_leaks