1087518 - PHttpChannelChild::Send__delete__ called late in shutdown

Status: RESOLVED WORKSFORME

(Core :: Networking, defect, P3)

Description

[Andrew McCreight [:mccr8]]

In debug Linux e10s M5, we sometimes destroy an nsHTMLDocument during shutdown cycle collection, which ends up calling HttpChannelChild::Release, which ends up calling PHttpChannelChild::Send__delete__() when it is too late to actually send anything, so we exit the child process.  This prevents us from getting leak logs for the child process.

Comment 1

[Josh Matthews [:jdm]]

I wonder if we could just make the IPDL machinery know when it's too late to be able to send anything and silently ignore such requests? It seems like the only other solution is playing whack-a-mole with the ContentChild::IsAlive hammer.

Comment 2

[Andrew McCreight [:mccr8]]

That sounds fine to me, but I think Ben wants it like this.  Is that right, Ben?  We have a few of these on file already, and I need to file some more.

Comment 3

[Josh Matthews [:jdm]]

Also, this case is more complicated because the IPDL destructor will release the channel, thus preventing a leak. If we guard the Send__delete__ call, we don't run the release code and therefore leak the channel.

Comment 4

[Andrew McCreight [:mccr8]]

This is also happening sometimes in e10s debug M1.

Comment 5

[Ben Turner (not reading bugmail, use the needinfo flag!)]

We talked about this in some other bug, right?

Comment 6

[Andrew McCreight [:mccr8]]

Yeah, sorry, I should have cancelled the needinfo.

Comment 7

[Firefox Bug Husbandry Bot]

Bulk change to priority: https://bugzilla.mozilla.org/show_bug.cgi?id=1399258

Comment 8

[Firefox Bug Husbandry Bot]

Bulk change to priority: https://bugzilla.mozilla.org/show_bug.cgi?id=1399258

Comment 9

[Andrew McCreight [:mccr8]]

content process leak logs have been working for a long time, so I suppose this isn't relevant any more.