Re: e10s-multi update and tests

2017-04-13 Thread Gabor Krizsanits 
Hi,

On Thu, Apr 13, 2017 at 6:09 AM, Fischer Liu  wrote:

> We are using dom.ipc.processCount to limit the count of process.
> After updating dom.ipc.processCount, do we still need to restart Firefox?
>
>
It depends. It will prevent any new processes to be launched that would go
beyond the new limit right away, but it will not shut down any existing
processes if the current number is above the limit. For that you must
either close tabs or restart the browser.


> As I know in PreallocatedProcessManagerImpl::AllocateNow[2] and
> PreallocatedProcessManagerImpl::RereadPrefs, there would read the pref to
> check the max count to allow creation or close process.
>
> [1]
> https://dxr.mozilla.org/mozilla-central/rev/f40e24f40b4c4556944c762d4764ea
> ce261297f5/dom/ipc/PreallocatedProcessManager.cpp#143
>  [2]
> https://dxr.mozilla.org/mozilla-central/rev/f40e24f40b4c4556944c762d4764ea
> ce261297f5/dom/ipc/PreallocatedProcessManager.cpp#200
>
>
The preallocated process manager is not enabled yet on any channels by
default, I'm still struggling with some failing tests and other issues with
it. Hopefully that'll be fixed soon (Bug 1341008
), I wish I had more
time for it... Once it's turned on, it will detect the change of the pref
and shut down the preallocated process if necessary just like you described.

Gabor
___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Re: e10s-multi update and tests

2017-04-12 Thread Fischer Liu 
Hi

We are using dom.ipc.processCount to limit the count of process.
After updating dom.ipc.processCount, do we still need to restart Firefox?

As I know in PreallocatedProcessManagerImpl::AllocateNow[2] and
PreallocatedProcessManagerImpl::RereadPrefs, there would read the pref to
check the max count to allow creation or close process.

[1]
https://dxr.mozilla.org/mozilla-central/rev/f40e24f40b4c4556944c762d4764eace261297f5/dom/ipc/PreallocatedProcessManager.cpp#143
 [2]
https://dxr.mozilla.org/mozilla-central/rev/f40e24f40b4c4556944c762d4764eace261297f5/dom/ipc/PreallocatedProcessManager.cpp#200

Thanks


On Fri, Mar 24, 2017 at 1:05 AM, Blake Kaplan  wrote:

> On Thu, Mar 23, 2017 at 3:32 AM, Andreas Tolfsen  wrote:
> > Do you use a tracking keyword of some sort for the disabled tests?
>
> This work is being tracked in
> .
> --
> Blake
> ___
> dev-platform mailing list
> dev-platform@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-platform
>
___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Re: e10s-multi update and tests

2017-03-23 Thread Blake Kaplan 
On Thu, Mar 23, 2017 at 3:32 AM, Andreas Tolfsen  wrote:
> Do you use a tracking keyword of some sort for the disabled tests?

This work is being tracked in
.
-- 
Blake
___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Re: e10s-multi update and tests

2017-03-23 Thread Gabor Krizsanits 
On Thu, Mar 23, 2017 at 8:56 AM, tapper  wrote:


> Hi is a11y working yet?
> Last time I tried it out it kept crashing.
>
> ___
> dev-platform mailing list
> dev-platform@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-platform
>

Hi, do you have a bug filed for that? As far as I can tell a11y only blocks
e10s on XP [1] on nightly, so 4 cp's should be enabled by default with a11y
users on any other platforms. If it's keep crashing that's a pretty big
deal, but so far we have not received any bug reports for that. So I would
like to think it's working.

[1]:
http://searchfox.org/mozilla-central/source/toolkit/xre/nsAppRunner.cpp#4950

Gabor
___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Re: e10s-multi update and tests

2017-03-23 Thread tapper 

On 22/03/2017 23:45, Blake Kaplan wrote:

Hello all,

As some of you might have noticed, we are now defaulting to 4 content
processes on Nightly builds! We're continuing to collect data and
planning on running experiments with different numbers of processes to
generate more data and allow us to fine-tune our defaults and
strategies for process allocation.

As part of our effort to enable e10s-multi on Nightly, we disabled a
few tests that were misbehaving. We've re-enabled most of them and are
on track to finish re-enabling them (after verifying that the problem
was in the test and not the underlying code).

As we get to the end of that effort, I'd like to ask everybody to
think about any areas that they feel are not tested with multiple
content processes that should be. Obviously, as we find regressions
related to multiple content processes we'll add regression tests, but
I would like to avoid having any last-minute panics over missing
tests.

Thanks.



Hi is a11y working yet?
Last time I tried it out it kept crashing.
___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Re: e10s-multi update and tests

2017-03-23 Thread Andreas Tolfsen 
Also sprach Blake Kaplan :

> As part of our effort to enable e10s-multi on Nightly, we disabled
> a few tests that were misbehaving. We've re-enabled most of them
> and are on track to finish re-enabling them (after verifying that
> the problem was in the test and not the underlying code).

Do you use a tracking keyword of some sort for the disabled tests?

___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Re: e10s-multi update and tests

2017-03-23 Thread Gabor Krizsanits 
On Thu, Mar 23, 2017 at 2:36 AM, Nicholas Nethercote  wrote:

> On Thu, Mar 23, 2017 at 12:12 PM, Andrew McCreight  >
> wrote:
>
> >
> > Though maybe you are asking which processes count against the limit of 4.
> >
>
> Yes, that's what I am asking.
>
> Nick
>
>
That is a good question, thanks for bringing it up. The GMP and other
non-content process types are  already addressed in previous messages, I
won't cover them here. The 4 content processes limit is for "web" type
content process.
We have different content process pools with different limits on them.
Currently we have these types: [1]

Web, File, Extension, Large allocation.

It's possible to set limit for all types by dom.ipc.processCount.[type].
(Note: dom.ipc.processCount sets limit for the default "web" type, we kept
this one for backward compatibility, but if you set
dom.ipc.processCount.web, dom.ipc.processCount will be ignored.)

[1]: http://searchfox.org/mozilla-central/source/dom/ipc/ContentParent.h#40

Gabor
___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Re: e10s-multi update and tests

2017-03-23 Thread bowen 
On Thursday, 23 March 2017 01:01:16 UTC, Nicholas Nethercote  wrote:
> Do we have a clear definition of "content process"? I.e. does/will it
> include:
> 
> - GMP processes (no?)
> - GPU process (probably not?)
> - file:// URL processes (probably should?)
> - Web Extensions processes (probably should?)
> - ServiceWorker processes (probably should?)

Each content process type (remote type) has its own pool of processes.
Remote types at the moment are: web (the normal content process), file, 
extension and webLargeAllocation.
GMP and GPU are their own GeckoProcessType.
(I'm not sure how the ServiceWorker one fits into this.)

Currently the pref dom.ipc.processCount acts as the default if 
dom.ipc.processCount. is not set.
(Mainly because this pref was already being used by a number of people, perhaps 
it should just have been left as a fallback for dom.ipc.processCount.web)

Current pref settings:
 * dom.ipc.processCount : 1 (Nightly 4)
 * dom.ipc.processCount.extension : 1
 * dom.ipc.processCount.webLargeAllocation : 10

So, on Nightly the web and file processes will both use up to 4 processes.
The file:// URL process is probably a fairly special case (although I don't 
have any stats on its usage), so we should probably add a specific setting of 1 
for that.

Maybe that 4 should be set on dom.ipc.processCount.web instead.

There has also been talk of setting an overall maximum although you would 
always have to allow at least 1 of each to start, even if we were at the 
maximum.

Cheers,
Bob

> Thanks.
> 
> Nick
> 
> On Thu, Mar 23, 2017 at 10:45 AM, Blake Kaplan  wrote:
> 
> > Hello all,
> >
> > As some of you might have noticed, we are now defaulting to 4 content
> > processes on Nightly builds! We're continuing to collect data and
> > planning on running experiments with different numbers of processes to
> > generate more data and allow us to fine-tune our defaults and
> > strategies for process allocation.
> >
> > As part of our effort to enable e10s-multi on Nightly, we disabled a
> > few tests that were misbehaving. We've re-enabled most of them and are
> > on track to finish re-enabling them (after verifying that the problem
> > was in the test and not the underlying code).
> >
> > As we get to the end of that effort, I'd like to ask everybody to
> > think about any areas that they feel are not tested with multiple
> > content processes that should be. Obviously, as we find regressions
> > related to multiple content processes we'll add regression tests, but
> > I would like to avoid having any last-minute panics over missing
> > tests.
> >
> > Thanks.
> > --
> > Blake
> > ___
> > firefox-dev mailing list
> > firefox-...@mozilla.org
> > https://mail.mozilla.org/listinfo/firefox-dev
> >

___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Re: e10s-multi update and tests

2017-03-22 Thread Chris Pearce 
On Thursday, March 23, 2017 at 2:37:18 PM UTC+13, Nicholas Nethercote wrote:
> On Thu, Mar 23, 2017 at 12:12 PM, Andrew McCreight
> wrote:
> 
> >
> > Though maybe you are asking which processes count against the limit of 4.
> >
> 
> Yes, that's what I am asking.
> 
> Nick

I think there's only one GPU process active at once.

GMP processes are not a content process, and should not count towards the 
content process limit.

At this stage I don't think we need to put a limit on how many GMP processes 
are active at once. All WebRTC's use of GMP shares a single GMP process. We do 
create a new GMP process for every origin doing EME, but in typical usage 
patterns I'd not expect many origins using EME concurrently. GMP processes also 
typically don't use more than 10MB of memory. So I don't think GMP processes 
will be an issue.

cpearce.
___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Re: e10s-multi update and tests

2017-03-22 Thread Nicholas Nethercote 
On Thu, Mar 23, 2017 at 12:12 PM, Andrew McCreight 
wrote:

>
> Though maybe you are asking which processes count against the limit of 4.
>

Yes, that's what I am asking.

Nick
___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Re: e10s-multi update and tests

2017-03-22 Thread Ben Kelly 
On Wed, Mar 22, 2017 at 9:12 PM, Andrew McCreight 
wrote:

> On Wed, Mar 22, 2017 at 6:00 PM, Nicholas Nethercote <
> n.netherc...@gmail.com
> > wrote:
>
> > Do we have a clear definition of "content process"? I.e. does/will it
> > include:
> >
> > - GMP processes (no?)
> > - GPU process (probably not?)
> > - file:// URL processes (probably should?)
> > - Web Extensions processes (probably should?)
> > - ServiceWorker processes (probably should?)
> >
>
> The most operational definition is that a content process is a process
> where XRE_IsContentProcess() returns true. From the enum GeckoProcessType,
> you can see the various other possibilities. So the answers to your
> questions are then probably that GMP and GPU are not, but the others are.
> Though maybe you are asking which processes count against the limit of 4.
>

FWIW, the intention is that the worker process would not count against the
limit.  It would not contain any windows and would be significantly smaller
in size.

Ben
___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Re: e10s-multi update and tests

2017-03-22 Thread Andrew McCreight 
On Wed, Mar 22, 2017 at 6:00 PM, Nicholas Nethercote  wrote:

> Do we have a clear definition of "content process"? I.e. does/will it
> include:
>
> - GMP processes (no?)
> - GPU process (probably not?)
> - file:// URL processes (probably should?)
> - Web Extensions processes (probably should?)
> - ServiceWorker processes (probably should?)
>

The most operational definition is that a content process is a process
where XRE_IsContentProcess() returns true. From the enum GeckoProcessType,
you can see the various other possibilities. So the answers to your
questions are then probably that GMP and GPU are not, but the others are.
Though maybe you are asking which processes count against the limit of 4.

Andrew


> Thanks.
>
> Nick
>
> On Thu, Mar 23, 2017 at 10:45 AM, Blake Kaplan  wrote:
>
> > Hello all,
> >
> > As some of you might have noticed, we are now defaulting to 4 content
> > processes on Nightly builds! We're continuing to collect data and
> > planning on running experiments with different numbers of processes to
> > generate more data and allow us to fine-tune our defaults and
> > strategies for process allocation.
> >
> > As part of our effort to enable e10s-multi on Nightly, we disabled a
> > few tests that were misbehaving. We've re-enabled most of them and are
> > on track to finish re-enabling them (after verifying that the problem
> > was in the test and not the underlying code).
> >
> > As we get to the end of that effort, I'd like to ask everybody to
> > think about any areas that they feel are not tested with multiple
> > content processes that should be. Obviously, as we find regressions
> > related to multiple content processes we'll add regression tests, but
> > I would like to avoid having any last-minute panics over missing
> > tests.
> >
> > Thanks.
> > --
> > Blake
> > ___
> > firefox-dev mailing list
> > firefox-...@mozilla.org
> > https://mail.mozilla.org/listinfo/firefox-dev
> >
> ___
> dev-platform mailing list
> dev-platform@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-platform
>
___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Re: e10s-multi update and tests

2017-03-22 Thread Nicholas Nethercote 
Do we have a clear definition of "content process"? I.e. does/will it
include:

- GMP processes (no?)
- GPU process (probably not?)
- file:// URL processes (probably should?)
- Web Extensions processes (probably should?)
- ServiceWorker processes (probably should?)

Thanks.

Nick

On Thu, Mar 23, 2017 at 10:45 AM, Blake Kaplan  wrote:

> Hello all,
>
> As some of you might have noticed, we are now defaulting to 4 content
> processes on Nightly builds! We're continuing to collect data and
> planning on running experiments with different numbers of processes to
> generate more data and allow us to fine-tune our defaults and
> strategies for process allocation.
>
> As part of our effort to enable e10s-multi on Nightly, we disabled a
> few tests that were misbehaving. We've re-enabled most of them and are
> on track to finish re-enabling them (after verifying that the problem
> was in the test and not the underlying code).
>
> As we get to the end of that effort, I'd like to ask everybody to
> think about any areas that they feel are not tested with multiple
> content processes that should be. Obviously, as we find regressions
> related to multiple content processes we'll add regression tests, but
> I would like to avoid having any last-minute panics over missing
> tests.
>
> Thanks.
> --
> Blake
> ___
> firefox-dev mailing list
> firefox-...@mozilla.org
> https://mail.mozilla.org/listinfo/firefox-dev
>
___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

e10s-multi update and tests

2017-03-22 Thread Blake Kaplan 
Hello all,

As some of you might have noticed, we are now defaulting to 4 content
processes on Nightly builds! We're continuing to collect data and
planning on running experiments with different numbers of processes to
generate more data and allow us to fine-tune our defaults and
strategies for process allocation.

As part of our effort to enable e10s-multi on Nightly, we disabled a
few tests that were misbehaving. We've re-enabled most of them and are
on track to finish re-enabling them (after verifying that the problem
was in the test and not the underlying code).

As we get to the end of that effort, I'd like to ask everybody to
think about any areas that they feel are not tested with multiple
content processes that should be. Obviously, as we find regressions
related to multiple content processes we'll add regression tests, but
I would like to avoid having any last-minute panics over missing
tests.

Thanks.
-- 
Blake
___
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform