[mojo-edk] Close pending merges if the parent channel dies.

[mojo-edk] Close pending merges if the parent channel dies.

Posix behaviour for child process when a parent dies is... complicated.
ChildThreadImpl relies on the IPC channel error handler to detect when the
parent is gone and shut down. However, if the parent dies before the channel's
token/port is merged, the child never receives the channel error.

This change causes ports that have pending merges to close when the parent
channel is gone. This doesn't cover the case when a merge has been sent, but
hasn't yet been processed by the time the parent channel is gone. That's
a bit more complicated.

BUG=604282
Committed: https://crrev.com/121c8804d511e331b9098a06e425172917c34e06
Cr-Commit-Position: refs/heads/master@{#405047}

[Anand Mistry (off Chromium), 2016-07-12 03:55:05]

Description was changed from

==========
Test.

BUG=604282
==========

to

==========
[mojo-edk] Close pending merges if the parent channel dies.

Posix behaviour for child process when a parent dies is... complicated.
ChildThreadImpl relies on the IPC channel error handler to detect when the
parent is gone and shut down. However, if the paren dies before the channel's
token/port is merged, the child never receives the channel error.

This change causes ports that have pending merges to close when the parent
channel is gone. This doesn't cover the case when a merge has been sent, but
hasn't yet been processed by the time the parent channel is gone. That's
a bit more complicated.

BUG=604282
==========

[Anand Mistry (off Chromium), 2016-07-12 03:55:30]

Description was changed from

==========
[mojo-edk] Close pending merges if the parent channel dies.

Posix behaviour for child process when a parent dies is... complicated.
ChildThreadImpl relies on the IPC channel error handler to detect when the
parent is gone and shut down. However, if the paren dies before the channel's
token/port is merged, the child never receives the channel error.

This change causes ports that have pending merges to close when the parent
channel is gone. This doesn't cover the case when a merge has been sent, but
hasn't yet been processed by the time the parent channel is gone. That's
a bit more complicated.

BUG=604282
==========

to

==========
[mojo-edk] Close pending merges if the parent channel dies.

Posix behaviour for child process when a parent dies is... complicated.
ChildThreadImpl relies on the IPC channel error handler to detect when the
parent is gone and shut down. However, if the paren dies before the channel's
token/port is merged, the child never receives the channel error.

This change causes ports that have pending merges to close when the parent
channel is gone. This doesn't cover the case when a merge has been sent, but
hasn't yet been processed by the time the parent channel is gone. That's
a bit more complicated.

BUG=604282
==========

[Anand Mistry (off Chromium), 2016-07-12 03:55:41]

Description was changed from

==========
[mojo-edk] Close pending merges if the parent channel dies.

Posix behaviour for child process when a parent dies is... complicated.
ChildThreadImpl relies on the IPC channel error handler to detect when the
parent is gone and shut down. However, if the paren dies before the channel's
token/port is merged, the child never receives the channel error.

This change causes ports that have pending merges to close when the parent
channel is gone. This doesn't cover the case when a merge has been sent, but
hasn't yet been processed by the time the parent channel is gone. That's
a bit more complicated.

BUG=604282
==========

to

==========
[mojo-edk] Close pending merges if the parent channel dies.

Posix behaviour for child process when a parent dies is... complicated.
ChildThreadImpl relies on the IPC channel error handler to detect when the
parent is gone and shut down. However, if the parent dies before the channel's
token/port is merged, the child never receives the channel error.

This change causes ports that have pending merges to close when the parent
channel is gone. This doesn't cover the case when a merge has been sent, but
hasn't yet been processed by the time the parent channel is gone. That's
a bit more complicated.

BUG=604282
==========

[Anand Mistry (off Chromium), 2016-07-12 03:57:09]

amistry@chromium.org changed reviewers:
+ rockot@chromium.org

[Anand Mistry (off Chromium), 2016-07-12 03:57:09]

This is the cause of the test failure which caused the GPU ChannelMojo change to
be reverted. Posix is weird.

[Ken Rockot(use gerrit already), 2016-07-12 07:08:47]

lgtm - isn't the other case you mention in the description already covered by
node loss behavior? as soon as a merge request is sent, the local port is
proxied to the destination node. if that node connection is then dropped, we
should back propagate proxy disruption.

[Anand Mistry (off Chromium), 2016-07-12 07:31:40]

On 2016/07/12 07:08:47, Ken Rockot wrote:
> lgtm - isn't the other case you mention in the description already covered by
> node loss behavior? as soon as a merge request is sent, the local port is
> proxied to the destination node. if that node connection is then dropped, we
> should back propagate proxy disruption.

Unfortunately, no. A merge request to the parent only contains the port name.
The port itself isn't associated with the parent node (because both ends are
still local) and so it won't get caught by the node connection loss.

I won't submit this today because I'm not particularly happy about it. But the
current structure doesn't lead to a better approach I can think of at the
moment.

[Anand Mistry (off Chromium), 2016-07-13 04:16:12]

The CQ bit was checked by amistry@chromium.org

[commit-bot: I haz the power, 2016-07-13 04:17:26]

CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-07-13 05:39:41]

Description was changed from

==========
[mojo-edk] Close pending merges if the parent channel dies.

Posix behaviour for child process when a parent dies is... complicated.
ChildThreadImpl relies on the IPC channel error handler to detect when the
parent is gone and shut down. However, if the parent dies before the channel's
token/port is merged, the child never receives the channel error.

This change causes ports that have pending merges to close when the parent
channel is gone. This doesn't cover the case when a merge has been sent, but
hasn't yet been processed by the time the parent channel is gone. That's
a bit more complicated.

BUG=604282
==========

to

==========
[mojo-edk] Close pending merges if the parent channel dies.

Posix behaviour for child process when a parent dies is... complicated.
ChildThreadImpl relies on the IPC channel error handler to detect when the
parent is gone and shut down. However, if the parent dies before the channel's
token/port is merged, the child never receives the channel error.

This change causes ports that have pending merges to close when the parent
channel is gone. This doesn't cover the case when a merge has been sent, but
hasn't yet been processed by the time the parent channel is gone. That's
a bit more complicated.

BUG=604282
==========

[commit-bot: I haz the power, 2016-07-13 05:39:42]

Committed patchset #3 (id:40001)

[commit-bot: I haz the power, 2016-07-13 05:41:01]

Description was changed from

==========
[mojo-edk] Close pending merges if the parent channel dies.

Posix behaviour for child process when a parent dies is... complicated.
ChildThreadImpl relies on the IPC channel error handler to detect when the
parent is gone and shut down. However, if the parent dies before the channel's
token/port is merged, the child never receives the channel error.

This change causes ports that have pending merges to close when the parent
channel is gone. This doesn't cover the case when a merge has been sent, but
hasn't yet been processed by the time the parent channel is gone. That's
a bit more complicated.

BUG=604282
==========

to

==========
[mojo-edk] Close pending merges if the parent channel dies.

Posix behaviour for child process when a parent dies is... complicated.
ChildThreadImpl relies on the IPC channel error handler to detect when the
parent is gone and shut down. However, if the parent dies before the channel's
token/port is merged, the child never receives the channel error.

This change causes ports that have pending merges to close when the parent
channel is gone. This doesn't cover the case when a merge has been sent, but
hasn't yet been processed by the time the parent channel is gone. That's
a bit more complicated.

BUG=604282

Committed: https://crrev.com/121c8804d511e331b9098a06e425172917c34e06
Cr-Commit-Position: refs/heads/master@{#405047}
==========

[commit-bot: I haz the power, 2016-07-13 05:41:02]

Patchset 3 (id:??) landed as
https://crrev.com/121c8804d511e331b9098a06e425172917c34e06
Cr-Commit-Position: refs/heads/master@{#405047}

[waxmiguel, 2016-07-13 05:43:24]

waxmigs2902@gmail.com changed reviewers:
+ waxmigs2902@gmail.com

[waxmiguel, 2016-07-13 05:43:24]

cheers!

[waxmiguel, 2016-07-13 05:44:54]

On 2016/07/12 03:55:05, Anand Mistry wrote:
> Description was changed from
> 
> ==========
> Test.
> 
> BUG=604282
> ==========
> 
> to
> 
> ==========
> [mojo-edk] Close pending merges if the parent channel dies.
> 
> Posix behaviour for child process when a parent dies is... complicated.
> ChildThreadImpl relies on the IPC channel error handler to detect when the
> parent is gone and shut down. However, if the paren dies before the channel's
> token/port is merged, the child never receives the channel error.
> 
> This change causes ports that have pending merges to close when the parent
> channel is gone. This doesn't cover the case when a merge has been sent, but
> hasn't yet been processed by the time the parent channel is gone. That's
> a bit more complicated.
> 
> BUG=604282
> ==========

thanks

[waxmiguel, 2016-07-13 06:12:40]

On 2016/07/13 05:39:42, commit-bot: I haz the power wrote:
> Committed patchset #3 (id:40001)