Roll isolate go to binaries as generated by infra @ 6e664e19.

Roll isolate go to binaries as generated by infra @ 6e664e19.

This should bring back much faster uploading because of faster
compression library, but built with 1.6 Go compiler, which seemed
to have caused regression in speed in August roll @ a7ce77dd and
was reverted in @ 07ac77f3 in September.

There are also changes to auth library and various other stability
improvements accumulated since February 2016.

Hashes taken from builds of 3 builders on chromium.infra waterfall:
  /builders/infra-continuous-trusty-64/builds/4280
  /builders/infra-continuous-win-64/builds/4127
  /builders/infra-continuous-mac-10.9-64/builds/4142

R=maruel@chromium.org,djd@chromium.org
BUG=653723

[tandrii(chromium), 2016-10-20 15:54:32]

The CQ bit was checked by tandrii@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2016-10-20 15:54:43]

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

[M-A Ruel, 2016-10-20 16:13:12]

lgtm

[tandrii(chromium), 2016-10-20 16:14:19]

The CQ bit was unchecked by tandrii@chromium.org

[tandrii(chromium), 2016-10-20 16:14:21]

The CQ bit was checked by tandrii@chromium.org

[commit-bot: I haz the power, 2016-10-20 16:14:32]

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

[tandrii(chromium), 2016-10-20 16:22:13]

The CQ bit was unchecked by tandrii@chromium.org

[tandrii(chromium), 2016-10-20 16:25:48]

i don't like this roll any more. This is isolate run of this CL on Mac Release
builder:
https://build.chromium.org/p/tryserver.chromium.mac/builders/mac_chromium_rel...

319285
======
Hits    : 35075 (1.79Gib)
Misses  :   144 (5.09Gib)
Duration: 3m40.472s

5 prior builds on the same builder (but not the same slave though) have all <2
minutes run with comparable upload size.

319284
======
Hits    : 33927 (1.63Gib)
Misses  :    98 (4.33Gib)
Duration: 1m33.862s

319283
======
Hits    : 33922 (1.62Gib)
Misses  :   123 (4.60Gib)
Duration: 1m30.231s

319282
======
Hits    : 33818 (1.57Gib)
Misses  :    96 (3.97Gib)
Duration: 1m24.56s

319280
======
Hits    : 35142 (1.67Gib)
Misses  :   183 (5.24Gib)
Duration: 1m54.706s

[tandrii(chromium), 2016-10-20 16:28:40]

and on the same *slave* and same builder 2 more builds:

319255
======
Hits    : 33849 (1.77Gib)
Misses  :    64 (3.76Gib)
Duration: 1m28.296s

319174
======
Hits    : 33943 (1.84Gib)
Misses  :    70 (4.09Gib)
Duration: 1m26.233s



So, no this should not land at all. Something is still very very broken OR I
rolled wrong binaries...

[tandrii(chromium), 2016-10-20 16:37:05]

on the other hand, linux_chromium_chromeos_rel_ng shows improvement:

this CL :
Hits    : 16698 (4.96Gib)
Misses  :    54 (8.41Gib)
Duration: 3m8.526s

/linux_chromium_chromeos_rel_ng/builds/299830

Hits    : 16690 (5.39Gib)
Misses  :    68 (7.98Gib)
Duration: 4m26.718s

/linux_chromium_chromeos_rel_ng/builds/299828/
Hits    : 16677 (4.42Gib)
Misses  :    72 (8.95Gib)
Duration: 4m43.932s


This doesn't make much sense to me :(

[tandrii(chromium), 2016-10-21 13:03:28]

i just had thought: mac builders are in Golo which has a limited pipe to outside
Google storage servers. linux slaves are on GCE which have HUGE pipes to GS.
Hence, linux servers should be optimized for sending more data faster, while
Macs and Windowses (while in Golo) should perhaps invest more into compression.
What really would be awesome is measure how many bytes actually get uploaded
with time vs just complete files. M-A, WDYT?

[M-A Ruel, 2016-10-21 13:07:52]

On 2016/10/21 13:03:28, tandrii(chromium) wrote:
> i just had thought: mac builders are in Golo which has a limited pipe to
outside
> Google storage servers. linux slaves are on GCE which have HUGE pipes to GS.
> Hence, linux servers should be optimized for sending more data faster, while
> Macs and Windowses (while in Golo) should perhaps invest more into
compression.
> What really would be awesome is measure how many bytes actually get uploaded
> with time vs just complete files. M-A, WDYT?

Likely but it is a very complex problem. Also, this is orthogonal to fixing the
regression.