raise nag_max when running on windows

raise nag_max when running on windows

I tried 2x to "fetch chromium" on a Z620 Win8 and was unable to complete
successfully without something getting killed. I also tried 1x at nag_max=20,
but had to go higher to complete successfully. My network and disk IO are
probably high-end, so a larger buffer is probably reasonable.

R=szager@chromium.org

[szager1, 2013-05-31 17:08:47]

I'd like to understand this better before you do this.  Even on Windows, there
shouldn't be a process that goes 25 minutes without producing terminal output. 
On linux, I rarely see a git process go longer than 5 seconds without terminal
output.

I assume the process that's timing out is the clone of blink, yes?

[scottmg, 2013-05-31 17:16:23]

On 2013/05/31 17:08:47, szager1 wrote:
> I'd like to understand this better before you do this.  Even on Windows, there
> shouldn't be a process that goes 25 minutes without producing terminal output.

> On linux, I rarely see a git process go longer than 5 seconds without terminal
> output.
> 
> I assume the process that's timing out is the clone of blink, yes?

I saw both blink and ffmpeg get killed. (Log's gone now because of git svn fetch
long output).

The machine is basically unusable during the initial sync (for 5-10 minutes) so
I'm guessing it's just very heavy IO contention which makes the larger repos
thrash while resolving.

Potentially also output buffering so there _is_ output, but the script doesn't
see it for "a while"?

[szager1, 2013-05-31 17:19:49]

On 2013/05/31 17:16:23, scottmg wrote:

> The machine is basically unusable during the initial sync (for 5-10 minutes)
so
> I'm guessing it's just very heavy IO contention which makes the larger repos
> thrash while resolving.

This makes a lot of sense, and it explains why I haven't seen it in testing (I
rarely do a full clean checkout).

Before doing this, I'd like to try a different approach: limit the number of
full repo checkouts that gclient can run in parallel.

[scottmg, 2013-05-31 17:27:35]

On 2013/05/31 17:19:49, szager1 wrote:
> On 2013/05/31 17:16:23, scottmg wrote:
> 
> > The machine is basically unusable during the initial sync (for 5-10 minutes)
> so
> > I'm guessing it's just very heavy IO contention which makes the larger repos
> > thrash while resolving.
> 
> This makes a lot of sense, and it explains why I haven't seen it in testing (I
> rarely do a full clean checkout).
> 
> Before doing this, I'd like to try a different approach: limit the number of
> full repo checkouts that gclient can run in parallel.

sg.

[scottmg, 2013-06-20 19:48:08]

On 2013/05/31 17:27:35, scottmg wrote:
> On 2013/05/31 17:19:49, szager1 wrote:
> > On 2013/05/31 17:16:23, scottmg wrote:
> > 
> > > The machine is basically unusable during the initial sync (for 5-10
minutes)
> > so
> > > I'm guessing it's just very heavy IO contention which makes the larger
repos
> > > thrash while resolving.
> > 
> > This makes a lot of sense, and it explains why I haven't seen it in testing
(I
> > rarely do a full clean checkout).
> > 
> > Before doing this, I'd like to try a different approach: limit the number of
> > full repo checkouts that gclient can run in parallel.
> 
> sg.

Hi, after the latest thread on chromium-dev, I've gotten 2 more private mails
that "fetch chromium" is unusable on Windows.

Could we please figure out a mitigation if you don't like this one?

[iannucci1, 2013-06-20 20:11:00]

I'm going to be doing a compatibility pass over fetch/recipes for
windows/mac, I'll make sure to cover this today+tomorrow.


On Thu, Jun 20, 2013 at 12:48 PM, <scottmg@chromium.org> wrote:

> On 2013/05/31 17:27:35, scottmg wrote:
>
>> On 2013/05/31 17:19:49, szager1 wrote:
>> > On 2013/05/31 17:16:23, scottmg wrote:
>> >
>> > > The machine is basically unusable during the initial sync (for 5-10
>>
> minutes)
>
>> > so
>> > > I'm guessing it's just very heavy IO contention which makes the larger
>>
> repos
>
>> > > thrash while resolving.
>> >
>> > This makes a lot of sense, and it explains why I haven't seen it in
>> testing
>>
> (I
>
>> > rarely do a full clean checkout).
>> >
>> > Before doing this, I'd like to try a different approach: limit the
>> number of
>> > full repo checkouts that gclient can run in parallel.
>>
>
>  sg.
>>
>
> Hi, after the latest thread on chromium-dev, I've gotten 2 more private
> mails
> that "fetch chromium" is unusable on Windows.
>
> Could we please figure out a mitigation if you don't like this one?
>
>
https://codereview.chromium.**org/16004008/<https://codereview.chromium.org/1...
>