510266 - Electrolysis: when a content process crashes, the chrome process asserts and dies

Status: RESOLVED WORKSFORME

(Core :: IPC, defect)

Description

[Benjamin Smedberg]

When the content process crashes (currently due to a JS_ASSERT), the chrome process does this:

WARNING: Channel error, quitting IO loop!: file ../../../src/ipc/glue/AsyncChannel.cpp, line 163

[0813/140923:FATAL:../../../src/ipc/chromium/src/base/thread.cc(162)] Check failed: GetThreadWasQuitProperly().

bent said that the channel error shouldn't shut down the IO loop, IIRC.

Comment 1

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

Both parent and child processes share the same underlying channel implementation, AsyncChannel, that lives in ipc/glue. If there's a channel error in the child process we need to exit the IO loop (which is run on the main thread) so that the child process will die appropriately. The parent process, however, should most definitely not die in the event of a channel error. It should instead trigger some kind of "aw snap" notification.

Comment 2

[Jeremias Bosch (:jbos)]

i checked this on latest fennec. By killing the content process. 
Fennec keeps running and does not die. The Tap is dead but you can just close it and open a new one.

I see this: 

###!!! [Parent][AsyncChannel] Error: Channel error: cannot send/recv

But thats all.

Reported Bug is already fixed?

Comment 3

[Martijn Wargers (dead)]

This seems bad enough that it should block Fennec, right?

Comment 4

[Josh Matthews [:jdm]]

This does not happen in general.  New bugs should be filed for specific instances in which the chrome process also dies along with the content process.  I'm going to mark this WFM.

Comment 5

[Martijn Wargers (dead)]

Thanks for the explanation. I've seen Fennec sometimes crash when the content process is crashing. I'll give it a better look next time, when I see it and when I can reproduce, then I'll file a bug.