17591 - drive webshell leaks to zarro

Status: VERIFIED DUPLICATE of bug 22068

(SeaMonkey :: General, defect, P1)

Description

[chris hofmann]

ooks like we have regressed again..

win32 startup and shutdown leaks 4 webshells.
running browser buster for a few minutes gets
XXX WARNING: Number of webshells being leaked: 36

linux doen't look like it reports webshells leaked on exit any more.  I'm I
missing voodo?

looks like we never found an owner for driving leaked webshells to zero.
and its not on the radar as part of the webshell re-design.   is it a
porkjockey task?

the [e-mail] team will start reporting webshells leaked as part
of
daily smoke-test reporting next week.

chris h.

Chris Hofmann wrote on Sun, 10 Oct 1999 14:03:13 -0700 (PDT):


  We are looking much better.  In a build made from early this morning.
  I ran for about 20 minutes, loaded around thirty pages, and after that  exit
and see
  7 appshells leaked..

  ...
  Document http://komodo.mozilla.org/buster/test_url_25.html loaded successfully
  Document: Done (14.78 secs)
  Error loading URL http://komodo.mozilla.org/buster/test_url_25.html
  Document: Done (27.08 secs)
  Error loading URL http://komodo.mozilla.org/buster/test_url_25.html
  Document: Done (33.5 secs)
  # of builders: 7
  Document http://komodo.mozilla.org/buster/test_url_25.html loaded successfully
  Document: Done (14.39 secs)
  # of builders: 1
  # of builders: 1
  # of builders: 7
  Shut down app shell component {4a85a5d0-cddd-11d2-b7f6-00805f05ffa5},
rv=0x00000
  000
  Shut down app shell component {18c2f989-b09f-11d2-bcde-00805f0e1353},
rv=0x00000
  000
  *** XPInstall Component destroyed
  XXX WARNING: Number of webshells being leaked: 7


  chris h.

  Steve Clark wrote:

    I don't think the webshell redesign will significantly change the ownership
pattern for webshell.  If some object is
    leaking a webshell today, that same object will probably still leak a
webshell when we're done.  I don't think
    there's any reason for people to wait for us before they start tracking
these leaks down.  In fact, that would
    probably help us considerably if the leaks were at 0 before we started, so
we can guarantee we don't introduce
    any new leaks.

    steve

    Chris Hofmann wrote:


     who would be on top of driving this number to zero, and how quickly might
there be progress?

     rpotts, travis, other?

     is there low hanging fruit that could reduce the number quickly,
     or is the travis webshell redesign, implementation, and landing required?

     chris h.

     Chris Hofmann wrote:

       leger/paulmac/build-accept team,   lets at this report on the number of
webshell's leaked to the browser
       daily verification smoke test report and track how we are doing to get to
zero.

       under "top sites"  it would be great if we id'ed the number of sites
loaded and then
       the number of webshell's leaked at the end of the test session.  this
info can be found
       as one of the last entries in the console window.

       this data would be good for intl. smoke test as well...

       thx

       chris h.

       "Kipp E.B. Hickman" wrote:

       > Chris Hofmann wrote:
       >
       > > As a baseline I ran the Friday night m10 build and let it cycle for a
7 or 8
       > > page loads on browser buster.
       > >
       > > debug | browserbuster
       > > http://komodo.mozilla.org/buster/test_url_25.html
       > >
       > > then file |quiit and somewhere in the neighborhood of 30-45 webshells
being leaked.
       > > here is a sample....
       > >
       > > session 1
       > > XXX WARNING: Number of webshells being leaked: 32
       > >
       > > session 2
       > > XXX WARNING: Number of webshells being leaked: 39
       > >
       > > session 3
       > > XXX WARNING: Number of webshells being leaked: 44
       > >
       > > I also ran this build overnight and loaded about 110 pages
       > > before I ran out of GDI resources and hung up response on my system.
       > >
       > > Questions:
       > >
       > > 1) how good is the webshell leak accounting in m10 and current trunk?
       >
       > 100% accurate. The way that it works is that the contructor for
webshells increments a counter, and
       > the destructor decrements the counter.
       >
       > >
       > > 2) when might we be ready to develop a good metric on this to plug in
to the verification criteria?
       >
       > Zero is the correct metric. And should be done ASAP.
       >
       > >
       > > 3) do we know how much these leaks are costing us?
       >
       > Huge amounts of memory and other system specific resources (e.g. in
windoze, alot of GDI resources).
       > In particular, you usually end up leaking the entire document, frame
model, style model, etc. In
       > addition to font caches and therefore the underlying system font
resources! The tree should turn
       > orange when a webshell leaks, IMO.
       >
       > The viewer doesn't leak webshells, at least for simple documents (I
haven't checked framesets/iframes
       > in awhile).
       >
       > --

Comment 1

[leger]

Not for dogfood...putting on PDT- radar.

Comment 2

[chris hofmann]

If I run browser buster for a few minutes I shutdown most webshells
but I still leak a few.   over to travis for investigation.


...
890
! nsSecureBrowserUIImpl found: http://www.nai.com/
change lock icon to unlock ?(http 0)
WEBSHELL- = 19
WEBSHELL- = 18
WEBSHELL- = 17
WEBSHELL- = 16
WEBSHELL- = 15
WEBSHELL- = 14
WEBSHELL- = 13
WEBSHELL- = 12
WEBSHELL- = 11
WEBSHELL- = 10
WEBSHELL- = 9
WEBSHELL- = 8
WEBSHELL- = 7
WEBSHELL- = 6
WEBSHELL- = 5
WEBSHELL- = 4
WEBSHELL- = 3
Hey : You are in QFA Shutdown
XXX WARNING: Number of webshells being leaked: 3
WEBSHELL- = 2

Comment 3

[don]

A noble task for M15.

Comment 4

[leger]

Putting dogfood in the keyword field.

Comment 5

[travis]

Ok, leaks in general seem to be down.  Can someone give me a case for the 
quickest most directy way to get leak?  By most direct, I mean where the number 
of WEBSHELL+ = is the fewest.  If someone can give me a good leak case with the 
number of WEBSHELL+ = in the 7 or so range, I can probably track this down 
really quickly.

Comment 6

[Brad Roberts]

*** This bug has been marked as a duplicate of 22068 ***

Comment 7

[Alan S. Jones]

verifed dupe