STR: * Install Firefox beta * Make sure the add-on Mozilla Feedback/Test Pilot is installed and enabled or install it at https://addons.mozilla.org/en-US/firefox/addon/test-pilot/ 1. Run Mem Buster Test [http://random.pavlov.net/membuster/index.html] 2. Open about:memory 3. Click <minimize memory usage> 4. Take note of the memory used * Disable the add-on Mozilla Feedback/Test Pilot and restart from 1. * Compare the results between the two iterations.
Created attachment 582078 [details] Membuster - light version To reproduce it more easily, here's modified version of Mem Buster I wrote two years ago. It only takes about 70 seconds and opens each time http://www.google.com in maximum 10 windows. By the way, the memory benchmark and display feature doesn't work anymore, so if anyone knows why, you're welcome to tell me.
Can you spoil the ending for me and tell me what results you get? If the add-on causes more memory to be used, that's not necessarily a bug. Does it cause any zombie compartments? (https://developer.mozilla.org/en/Zombie_Compartments.)
Here are the values for the memory usage reported under explicit: Without any add-on: 42 43 44 45 45 With Feedback add-on: 65 93 117 142 169 (with 113 under heap-unclassified) I pasted the about:memory page below. The verbose version doesn't seem to be more interesting, but I saved it anyway. There's no zombie compartment, it seems more like a memory fragmentation issue. about:memory ------------ Main Process Explicit Allocations 168.88 MB (100.0%) -- explicit +--113.44 MB (67.17%) -- heap-unclassified +---43.33 MB (25.66%) -- js ¦ +--37.83 MB (22.40%) -- compartment([System Principal], 0x3178000) ¦ ¦ +--24.83 MB (14.70%) -- gc-heap ¦ ¦ ¦ +--10.15 MB (06.01%) -- shapes ¦ ¦ ¦ +---8.74 MB (05.18%) -- objects ¦ ¦ ¦ +---4.93 MB (02.92%) -- arena-unused ¦ ¦ ¦ +---1.00 MB (00.59%) -- (6 omitted) ¦ ¦ +---5.81 MB (03.44%) -- property-tables ¦ ¦ +---3.67 MB (02.17%) -- object-slots ¦ ¦ +---2.31 MB (01.37%) -- (8 omitted) ¦ ¦ +---1.20 MB (00.71%) -- script-data ¦ +---2.98 MB (01.76%) -- gc-heap-chunk-dirty-unused ¦ +---1.33 MB (00.79%) -- (8 omitted) ¦ +---1.19 MB (00.70%) -- compartment(atoms) ¦ +--1.19 MB (00.70%) -- (3 omitted) +----9.95 MB (05.89%) -- storage ¦ +--9.95 MB (05.89%) -- sqlite ¦ +--7.01 MB (04.15%) -- places.sqlite ¦ ¦ +--6.72 MB (03.98%) -- cache-used  ¦ ¦ +--0.29 MB (00.17%) -- (2 omitted) ¦ +--2.09 MB (01.24%) -- (12 omitted) ¦ +--0.85 MB (00.50%) -- other +----1.27 MB (00.75%) -- (6 omitted) +----0.89 MB (00.53%) -- xpti-working-set Other Measurements 0.00 MB -- gfx-d2d-surfacecache 0.00 MB -- gfx-d2d-surfacevram 0.00 MB -- gfx-surface-image 0.18 MB -- gfx-surface-win32 138.28 MB -- heap-allocated 161.67 MB -- heap-committed 1.63 MB -- heap-dirty 50.72 MB -- heap-unallocated 5 -- js-compartments-system 1 -- js-compartments-user 29.00 MB -- js-gc-heap 5.16 MB -- js-gc-heap-arena-unused 0.00 MB -- js-gc-heap-chunk-clean-unused 2.98 MB -- js-gc-heap-chunk-dirty-unused 28.05% -- js-gc-heap-unused-fraction 213.85 MB -- private 219.74 MB -- resident 333.87 MB -- vsize
AFAWCT Test Pilot is auto-installed on Aurora and Beta. If that's true, then this is bad. I'll take a look with DMD to see what the heap-unclassified is all about.
(In reply to Nicholas Nethercote [:njn] from comment #5) > Am I doing something wrong? netscape.security.PrivilegeManager.enablePrivilege("UniversalXPConnect") silently fails when the file calling it is not on the local filesystem. So you just have to save it to your disk and open it to see the dialog message. > Are your results in comment 3 from the attached test? Yes.
I can't get this to work. I tried a local copy and hit the "continue..." button 7 or 8 times, but I only see this: Memory in use at startup: NaN Memory in use at test end: Difference: Benchmark: 0 Result: undefined Is that what you mean by "the memory benchmark and display feature doesn't work anymore"? Am I just supposed to reload about:memory after it finishes running each time?
(In reply to Nicholas Nethercote [:njn] from comment #7) > Is that what you mean by "the memory benchmark and display feature doesn't > work anymore"? > > Am I just supposed to reload about:memory after it finishes running each > time? Indeed. Sorry that it wasn't clear.
I've had mixed results. I created a new profile, installed Test Pilot, and ran your test several times, getting the following "explicit" values: 65, 180, 207, 172, 176, 178. "heap-unclassified" was typically 80--90MB, i.e. roughly 50%, which is high. Then I restarted the browser and tried again. This time "explicit" stayed roughly around 80MB each time I reran. Subsequent restarts gave similar behaviour. Without the addon it was 60--70MB. I re-did the "create new profile" part 3 times and the result was consistent each time. So, apart from the first run, Test Pilot's impact is modest. Thomas, can you replicate my results?
(In reply to Nicholas Nethercote [:njn] from comment #9) > So, apart from the first run, Test Pilot's impact is modest. > > Thomas, can you replicate my results? Unfortunately not. The usage stays the same no matter the number of restarts. I just noticed that when opening the Test Pilot window, the "Current studies" tab keeps loading without displaying anything else than the loading message. Maybe it has something to do with the memory usage. Do you see the same thing?
> I just noticed that when opening the Test Pilot window, the "Current > studies" tab keeps loading without displaying anything else than the loading > message. > Maybe it has something to do with the memory usage. Do you see the same > thing? I see two entries "Firefox Beta User Background Survey" and "Broken Extensions Survey". Each one has some descriptive text and a "Take the Survey" button. It doesn't change when I navigate to other pages. So it sounds like Test Pilot might be broken for you in some way?
adding Gregg Lind to this thread.
I tried it again with the latest beta, and I got the same results. But I've noticed there's a new version (1.2) than the one shipped with Firefox 10 Beta 4 (1.1.2). This new version works correctly (the current studies show up in the window) and there's no memory usage problem anymore. So I suggest to just update the version shipped with Firefox Beta and Aurora.
> This new version works correctly (the current studies show up in the window) and there's no memory > usage problem anymore. Sounds like this is fixed! Thanks for helping us with this, Thomas.
Well, the new version 1.2 works correctly, but Firefox Beta and Aurora still ship with the old leaking and non working version 1.1.2. Changing the bug summary according to it.
Oh, well that's bad. Jinghua, Greg, what's the story with this?
We did some investigation on it in last weeks. We found that patches are reviewed and accepted since last summer, but somehow they are not landed. ~3M users are on TP 1.2 now, and they got it through AMO update, not FX central. Gregg, any new update on next steps? What did Mossop say?
(In reply to Jinghua zhang from comment #17) > We did some investigation on it in last weeks. We found that patches are > reviewed and accepted since last summer, but somehow they are not landed. Reviewed but not landed in mozilla-central? Are there bugs on file for these? > ~3M users are on TP 1.2 now, and they got it through AMO update, not FX > central. > > Gregg, any new update on next steps? What did Mossop say? We have bug 719455 on file to get the tree updated to 1.2, still waiting on a patch or set of patches necessary to update to 1.2
> ~3M users are on TP 1.2 now, and they got it through AMO update, not FX central. How many people are on the older version?
Anyway, sounds like we should dupe this bug to bug 719455.