518788 - nsHttpHandler PruneDeadConnections works in TYPE_REPEATING_SLACK mode

Status: RESOLVED FIXED

(Core :: Networking, defect)

Description

[Oleg Romashin (:romaxa)]

Attachment: Stop PruneConnections timer on system-idle callback

Open URL or some other URL without active scripts, but http: protocol

Wait for some time (1 minute) to stop all short time activities after page loading.

set break point on 'nsBaseAppShell::OnDispatchedEvent'

EXP:
No  'nsBaseAppShell::OnDispatchedEvent' activity

ACTUAL:
Breakpoint 1, nsBaseAppShell::OnDispatchedEvent (this=0x8fd5258, thr=0x8d8fb98)
    at widget/src/xpwidgets/nsBaseAppShell.cpp:235
235	  if (mBlockNativeEvent)
(gdb) frame 3
#3  0xb74c5ae8 in nsTimerImpl::PostTimerEvent (this=0x8f5b898)
    at xpcom/threads/nsTimerImpl.cpp:554
554	  nsresult rv = mEventTarget->Dispatch(event, NS_DISPATCH_NORMAL);
(gdb) p *this
$1 = {<nsITimer> = {<nsISupports> = {_vptr.nsISupports = 0xb77401a8}, <No data fields>}, mRefCnt = {
    mValue = 2}, mEventTarget = {<nsCOMPtr_base> = {mRawPtr = 0x8d8fb98}, <No data fields>}, 
  mClosure = 0x0, mCallback = {c = 0x8f5479c, i = 0x8f5479c, o = 0x8f5479c}, 
  mTimerCallbackWhileFiring = {<nsCOMPtr_base> = {mRawPtr = 0x0}, <No data fields>}, 
  mCallbackType = 3 '\3', mType = 1 '\1', mFiring = 0 '\0', mArmed = 0, mCanceled = 0, 
  mGeneration = 5, mDelay = 15000, mTimeout = 4031807423}
(gdb) x/wa 0x8f5479c 
0x8f5479c:	0xb76a9b98 <_ZTV13nsHttpHandler+152>

PruneDeadConnectionsTimer created and firing timeouts all the time:
http://mxr.mozilla.org/mozilla-central/source/netwerk/protocol/http/src/nsHttpHandler.cpp#326

We certainly should suspend that timeout for "system-idle" state.
Also it would be nice to suspend that timer if there is no connections currently which are possible to prune. And activate timer only when something changed and appear some possibility to Prune connections.

Comment 1

[Christian :Biesinger (don't email me, ping me on IRC)]

What are the semantics of the system-idle notification? (And why are they not documented anywhere?)

Comment 2

[Christian :Biesinger (don't email me, ping me on IRC)]

Comment on attachment 402795 [details] [diff] [review]
Stop PruneConnections timer on system-idle callback

minusing pending an answer to the question in the previous comment

Comment 3

[Oleg Romashin (:romaxa)]

> What are the semantics of the system-idle notification? (And why are they not
> documented anywhere?)

I think because nobody did care about this problem before...
Doug should we have some general doc about Power Management notifications, handling, and rules about what to do to keep PMT in good shape?

Comment 4

[Patrick McManus [:mcmanus]]

the timeout tick is no longer active (for a long time now) when there are no live connections