1183390 - 100+ MiB regression in explicit/images/content/raster/used after bug 1117607 landed

Status: RESOLVED FIXED

(Core :: Graphics: ImageLib, defect)

Description

[Eric Rahm [:erahm]]

+++ This bug was initially created as a clone of Bug #1117607 +++

AWSY regression range:
https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=8370d245d8aa&tochange=aea2836ce9fe

Snippet of diff:

> 105.02 MB (100.0%) -- explicit
> ├──103.92 MB (98.95%) -- images
> │  ├──103.92 MB (98.95%) -- content/raster
> │  │  ├──103.92 MB (98.95%) -- used
> │  │  │  ├───67.24 MB (64.02%) -- (565 tiny)
> │  │  │  │   ├───0.99 MB (00.94%) ++ image(776x330, http://localhost:8073/tp5/ucoz.ru/www.ucoz.ru/ucoz/img/bg_footer_top.jpg)
> │  │  │  │   ├───0.98 MB (00.93%) ── image(760x338, http://localhost:8081/tp5/zol.com.cn/icon.zol-img.com.cn/2011/bb_0417bg5.png)/unlocked/surface(760x338@0)/decoded-heap
> │  │  │  │   ├───0.95 MB (00.90%) ── image(970x255, http://localhost:8081/tp5/zol.com.cn/icon.zol-img.com.cn/2011/bg0324.png)/unlocked/surface(970x255@0)/decoded-heap
> │  │  │  │   ├───0.95 MB (00.90%) ── image(970x255, http://localhost:8081/tp5/zol.com.cn/icon.zol-img.com.cn/2011/bg0412.png)/unlocked/surface(970x255@0)/decoded-heap
> │  │  │  │   ├───0.90 MB (00.86%) ++ image(480x360, http://localhost:8099/tp5/youtube.com/i1.ytimg.com/vi/l2HD6hNNCNs/hqdefault.jpg)/unlocked
> │  │  │  │   ├───0.90 MB (00.86%) ++ image(480x360, http://localhost:8099/tp5/youtube.com/i2.ytimg.com/vi/qbhumzOeTQc/hqdefault.jpg)/unlocked
> │  │  │  │   ├───0.90 MB (00.86%) ++ image(480x360, http://localhost:8099/tp5/youtube.com/i3.ytimg.com/vi/jrN41AUXKo4/hqdefault.jpg)/unlocked
> │  │  │  │   ├───0.84 MB (00.80%) ── image(400x551, http://localhost:8072/tp5/terra.com.br/s1.trrsf.com.br/atm/3/pt/_img/ttl-section.png)/unlocked/surface(400x551@0)/decoded-heap
> │  │  │  │   ├───0.83 MB (00.79%) ++ image(100x2151, http://localhost:8092/tp5/telegraph.co.uk/www.telegraph.co.uk/template/ver1-0/i/sprite-icon.gif)
> │  │  │  │   ├───0.83 MB (00.79%) ++ image(950x226, http://localhost:8093/tp5/seesaa.net/blog.seesaa.jp/img/portal/fl_help_t.jpg)
> │  │  │  │   ├───0.80 MB (00.77%) ── image(300x700, http://localhost:8097/tp5/homeway.com.cn/img.hexun.com/www/2010/img/hx_tab.gif)/unlocked/surface(300x700@0)/decoded-heap
> │  │  │  │   ├───0.77 MB (00.73%) ── image(455x441, http://localhost:8077/tp5/myspace.com/c2.ac-images.myspacecdn.com/images02/148/l_b16f6adde34b4bea8a57d39ff757bd09.jpg)/unlocked/surface(455x441@0)/decoded-heap
> │  │  │  │   ├───0.75 MB (00.71%) ++ image(500x388, http://localhost:8090/tp5/web.de/img.web.de/v/hp10/bg/bg-tabs.png)
> │  │  │  │   ├───0.74 MB (00.70%) ++ image(500x384, http://localhost:8098/tp5/facebook.com/static.ak.fbcdn.net/rsrc.php/v1/zD/r/B4K_BWwP7P5.png)
> │  │  │  │   ├───0.72 MB (00.69%) ++ image(512x288, http://localhost:8075/tp5/wsj.com/m.wsj.net/video/20110408/040811hubpm/040811hubpm_512x288.jpg)/unlocked
> │  │  │  │   ├───0.72 MB (00.69%) ── image(525x360, http://localhost:8079/tp5/orange.fr/hp3.a.woopic.com/I/sprite-HPC.gif)/unlocked/surface(525x360@0)/decoded-heap
> │  │  │  │   ├───0.72 MB (00.69%) ── image(525x360, http://localhost:8079/tp5/orange.fr/hp3.a.woopic.com/I/sprite-HPC.gif@1)/unlocked/surface(525x360@0)/decoded-heap
> │  │  │  │   ├───0.70 MB (00.66%) -- image(200x909, http://localhost:8072/tp5/terra.com.br/s1.trrsf.com.br/atm/3/core/_img/ico-general.png)
> │  │  │  │   │   ├──0.70 MB (00.66%) ── unlocked/surface(200x909@0)/decoded-heap
> │  │  │  │   │   └──0.00 MB (00.00%) ── source [+]
> │  │  │  │   ├───0.70 MB (00.66%) ++ image(200x909, http://localhost:8072/tp5/terra.com.br/s2.trrsf.com.br/atm/3/core/_img/ico-general.png)
> │  │  │  │   ├───0.70 MB (00.66%) -- image(545x331, http://localhost:8086/tp5/xunlei.com/misc.web.xunlei.com/www_5_6/img/foucebox.png)
> │  │  │  │   │   ├──0.69 MB (00.66%) ── unlocked/surface(545x331@0)/decoded-heap
> │  │  │  │   │   └──0.00 MB (00.00%) ── source [+]

Given that this is all images, I feel comfortable pointing at bug 1117607.

Comment 1

[Eric Rahm [:erahm]]

Attachment: memory report before

Comment 2

[Eric Rahm [:erahm]]

Attachment: memory report after

Comment 3

[Seth Fowler [:seth] [:s2h]]

So nothing is being leaked here. All of those images are unlocked, so they'd be discarded within 60s. (Or maybe less, I think we may have modified the pref on AWSY?)

My best guess at this point is some sort of race that's making RequestDiscard fail.

Comment 4

[Eric Rahm [:erahm]]

(In reply to Seth Fowler [:seth] from comment #3)
> So nothing is being leaked here. All of those images are unlocked, so they'd
> be discarded within 60s. (Or maybe less, I think we may have modified the
> pref on AWSY?)

We set 'image.mem.surfacecache.min_expiration_ms' to 10000 back in Feb [1]. Are there other values we should modify as well?

[1] https://github.com/mozilla/areweslimyet/commit/4a9af37ae7429ad5674c55463e6bb6dac2ec106c

Comment 5

[Seth Fowler [:seth] [:s2h]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=aa06ca421166

Comment 6

[Seth Fowler [:seth] [:s2h]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=9110413f3fb1

Comment 7

[Seth Fowler [:seth] [:s2h]]

(In reply to Eric Rahm [:erahm] from comment #4)
> We set 'image.mem.surfacecache.min_expiration_ms' to 10000 back in Feb [1].
> Are there other values we should modify as well?

Nope, that should do it.

I've just pushed a speculative patch that prevents us from locking the image more than once if we launch multiple decoders for the same image before the first one has finished. If this is the issue, bug 1176124 will be the long term fix, but there'd be no reason not to go ahead and land this patch and resolve the issue faster.

Comment 8

[Seth Fowler [:seth] [:s2h]]

Here's the series:

https://areweslimyet.com/?series=bug1183390

Not seeing any improvement from that patch, unfortunately.

Comment 9

[Seth Fowler [:seth] [:s2h]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=018cc6ee196b

Comment 10

[Seth Fowler [:seth] [:s2h]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=a11703ef3f1b

Comment 11

[Seth Fowler [:seth] [:s2h]]

As foretold by comment 7, bug 1176124 seems to have resolved this completely.

Comment 12

[Nicholas Nethercote [inactive]]

yay