864202 - large memory leak downloading fonts

Status: RESOLVED WORKSFORME

(Core :: Graphics: Text, defect)

Description

[John Daggett (:jtd)]

While working on a testcase to reproduce a crash in the downloadable font cache (bug 838105), I discovered that we have a rather large leak in our font downloading code.

Steps to reproduce (OSX):

1. Download http://people.mozilla.org/~jdaggett/fontstresstest.zip
2. Unzip fontstresstest.zip into ~/Sites directory
3. Enable .htaccess usage
  a. Open /etc/apache2/users/<username>.conf
  b. Change "AllowOverride None" to "AllowOverride All"
  c. Restart Apache: sudo apachectl restart
4. Load via HTTP "fontstress.html"
5. Open Console.app and verify in apache logs that fonts are
   always pulled (shouldn't see 304 responses to font requests)
6. Run over a long period (e.g. all night)

Result: when running the test with Nightly on Win7, after two days about:memory shows:

  2535MB explicit 
  -- 2435MB heap-unclassified

I'm assuming the cause is in gfx somewhere but it might also be in CORS or related loader code.

Comment 1

[John Daggett (:jtd)]

Attachment: memory report after running for two days with testcase

Comment 2

[John Daggett (:jtd)]

Kimura-san, why doesn't this qualify as a MemShrink bug?

Comment 3

[Andrew McCreight [:mccr8]]

probably just a bugzilla midair

Comment 4

[Masatoshi Kimura [:emk]]

Yeah, sorry for the confusion.
(Bugzilla didn't warn the mid-air somehow...)

Comment 5

[John Daggett (:jtd)]

Attachment: logfile from DMD build, running testcase under OSX

Running a quick test with a DMD build on OSX, it looks like we're
leaking resources related to the font load.  The stack below stands
out in the attached report.  I need to track down what object is not
getting free'd here.

Unreported: 7,052 blocks in stack trace record 1 of 1,929
 28,884,992 bytes (28,884,992 requested / 0 slop)
 27.71% of the heap (27.71% cumulative);  38.46% of unreported (38.46% cumulative)
 Allocated at
   replace_malloc[/builds/mozcentral/obj-dmd/memory/replace/dmd/libdmd.dylib +0x1E48] 0x100019e48
   malloc_impl[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/libmozglue.dylib +0x1AA0] 0x100077aa0
   zone_malloc[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/libmozglue.dylib +0x1389] 0x100077389
   malloc_zone_malloc[/usr/lib/system/libsystem_c.dylib +0x2D153] 0x7fff8bf68153
   malloc[/usr/lib/system/libsystem_c.dylib +0x2DBA7] 0x7fff8bf68ba7
   moz_xmalloc[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/libmozalloc.dylib +0x1685] 0x1000f0685
   NS_Alloc[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/XUL +0x2AC75B5] 0x104bc75b5
   nsMemoryImpl::Alloc(unsigned long)[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/XUL +0x2AC7599] 0x104bc7599
   nsSegmentedBuffer::AppendNewSegment()[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/XUL +0x2A80940] 0x104b80940
   nsStorageStream::Write(char const*, unsigned int, unsigned int*)[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/XUL +0x2A81AAC] 0x104b81aac
   non-virtual thunk to nsStorageStream::Write(char const*, unsigned int, unsigned int*)[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/XUL +0x2A81C6D] 0x104b81c6d
   nsCacheEntryDescriptor::nsOutputStreamWrapper::Write_Locked(char const*, unsigned int, unsigned int*)[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/XUL +0x212047] 0x102312047
   nsCacheEntryDescriptor::nsOutputStreamWrapper::Write(char const*, unsigned int, unsigned int*)[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/XUL +0x211F47] 0x102311f47
   nsInputStreamTeeWriteEvent::Run()[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/XUL +0x2A70BB4] 0x104b70bb4
   nsThread::ProcessNextEvent(bool, bool*)[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/XUL +0x2AAEDA4] 0x104baeda4
   NS_ProcessNextEvent(nsIThread*, bool)[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/XUL +0x2A133FF] 0x104b133ff
   nsThread::ThreadFunc(void*)[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/XUL +0x2AAD837] 0x104bad837
   _pt_root[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/libnss3.dylib +0x21FB23] 0x10141fb23
   _pthread_start[/usr/lib/system/libsystem_c.dylib +0x14742] 0x7fff8bf4f742

It also looks like a fair amount of memory is used by the CGFont associated with Mac font entries:

Unreported: ~1,879 blocks from ~66 stack trace records in stack frame record 22 of 6,667
 ~17,151,645 bytes (~15,492,354 requested / ~1,659,291 slop)
 16.46% of the heap;  22.84% of unreported
 PC is
   MacOSFontEntry::GetFontRef()[/builds/mozcentral/obj-dmd/dist/NightlyDebug.app/Contents/MacOS/XUL +0x2C67CAA] 0x104d67caa

Comment 6

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Looks like this is the cache's fault?

John, how bad is this during normal usage?

Comment 7

[Michal Novotny [:michal]]

I've tried to run the test and I can see the increasing memory usage but the memory isn't used by the memory cache. nsStorageStream doesn't leak memory and we don't leak nsStorageStream in nsMemoryCacheDevice. BTW, almost all the allocated memory (ca 90%) is released when I close the tab with the test or when I click on "Minimize memory usage" button. Re-running the test again released the remaining 10% in my case.

Comment 8

[Patrick McManus [:mcmanus]]

michal is there still something to do here? should it be wfm? Given back to fonts?

Comment 9

[Michal Novotny [:michal]]

I don't see leak or any other problem in the cache.

Comment 10

[Hani Yacoub, Desktop QA]

I couldn't manage to reproduce this issue. Since the bug was logged 9 years ago, most likely isn't reproducible anymore for the reporter also.
Closing this bug as resolved: Worksforme.