Use apache on the Mac on V8-Latest builders as an experiment...

Use apache on the Mac on V8-Latest builders as an experiment
to see if http test flakiness goes away.

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=33130

[tony, 2009-11-25 19:57:37]

http://codereview.chromium.org/437063/diff/1/3
File webkit/tools/layout_tests/layout_package/apache_http_server.py (right):

http://codereview.chromium.org/437063/diff/1/3#newcode36
webkit/tools/layout_tests/layout_package/apache_http_server.py:36:
path_utils.MaybeMakeDirectory(os.path.join("tmp", "WebKit"))
I think this is going to make the directory tmp/WebKit relative to where you run
the script.  You need the leading /.

http://codereview.chromium.org/437063/diff/1/2
File webkit/tools/layout_tests/run_webkit_tests.py (right):

http://codereview.chromium.org/437063/diff/1/2#newcode1081
webkit/tools/layout_tests/run_webkit_tests.py:1081: options.use_apache =
(options.builder_name and
What makes this mac only?

Can you add a comment here?  Perhaps the comment can explain what to pass to the
command line if one wants to test apache locally.

http://codereview.chromium.org/437063/diff/1/5
File webkit/tools/layout_tests/test_expectations.txt (right):

http://codereview.chromium.org/437063/diff/1/5#newcode3030
webkit/tools/layout_tests/test_expectations.txt:3030: // Started failing with
move from Lighttp to Apache on Mac.
nit: missing d on lighttpd

http://codereview.chromium.org/437063/diff/1/5#newcode3031
webkit/tools/layout_tests/test_expectations.txt:3031: // Since we have some bots
on Lighttp and some on Apache, mark it as flaky for now.
Should comments be 80 cols?

[ojan, 2009-11-25 20:16:48]

All done. Added a commandline flag to use apache. Also, added some sleep
calls to starting the server.

On Wed, Nov 25, 2009 at 11:57 AM, <tony@chromium.org> wrote:

>
> http://codereview.chromium.org/437063/diff/1/3
> File webkit/tools/layout_tests/layout_package/apache_http_server.py
> (right):
>
> http://codereview.chromium.org/437063/diff/1/3#newcode36
> webkit/tools/layout_tests/layout_package/apache_http_server.py:36:
> path_utils.MaybeMakeDirectory(os.path.join("tmp", "WebKit"))
> I think this is going to make the directory tmp/WebKit relative to where
> you run the script.  You need the leading /.
>
> http://codereview.chromium.org/437063/diff/1/2
> File webkit/tools/layout_tests/run_webkit_tests.py (right):
>
> http://codereview.chromium.org/437063/diff/1/2#newcode1081
> webkit/tools/layout_tests/run_webkit_tests.py:1081: options.use_apache =
> (options.builder_name and
> What makes this mac only?
>

Ugh. I forgot to add that bit. Done and tested on try-server.


> Can you add a comment here?  Perhaps the comment can explain what to
> pass to the command line if one wants to test apache locally.
>
> http://codereview.chromium.org/437063/diff/1/5
> File webkit/tools/layout_tests/test_expectations.txt (right):
>
> http://codereview.chromium.org/437063/diff/1/5#newcode3030
> webkit/tools/layout_tests/test_expectations.txt:3030: // Started failing
> with move from Lighttp to Apache on Mac.
> nit: missing d on lighttpd
>
> http://codereview.chromium.org/437063/diff/1/5#newcode3031
> webkit/tools/layout_tests/test_expectations.txt:3031: // Since we have
> some bots on Lighttp and some on Apache, mark it as flaky for now.
> Should comments be 80 cols?
>
>
> http://codereview.chromium.org/437063
>

[tony, 2009-11-25 20:57:09]

http://codereview.chromium.org/437063/diff/11/13
File webkit/tools/layout_tests/layout_package/apache_http_server.py (right):

http://codereview.chromium.org/437063/diff/11/13#newcode102
webkit/tools/layout_tests/layout_package/apache_http_server.py:102:
time.sleep(3)
This seems flaky.  What happens if you start too soon?  Can't bind to the port?

Maybe you can read the error log after apache starts for error messages instead
of sleeping.

http://codereview.chromium.org/437063/diff/11/13#newcode107
webkit/tools/layout_tests/layout_package/apache_http_server.py:107:
time.sleep(3)
Can we poll every 500ms or something when using UrlIsAlive?  It should be less
flaky.

[ojan, 2009-11-25 21:38:08]

http://codereview.chromium.org/437063/diff/11/13
File webkit/tools/layout_tests/layout_package/apache_http_server.py (right):

http://codereview.chromium.org/437063/diff/11/13#newcode102
webkit/tools/layout_tests/layout_package/apache_http_server.py:102:
time.sleep(3)
On 2009/11/25 20:57:09, tony wrote:
> This seems flaky.  What happens if you start too soon?  Can't bind to the
port?
> 
> Maybe you can read the error log after apache starts for error messages
instead
> of sleeping.
> 

I now check if stderr had any text and assume that means it didn't start up
properly. Does that seem fine? It worked for the one case I was hitting locally
at least.

http://codereview.chromium.org/437063/diff/11/13#newcode107
webkit/tools/layout_tests/layout_package/apache_http_server.py:107:
time.sleep(3)
On 2009/11/25 20:57:09, tony wrote:
> Can we poll every 500ms or something when using UrlIsAlive?  It should be less
> flaky.

Good call. It now polls when doing the Popen and when checking the ports.

[tony, 2009-11-25 21:45:30]

LGTM

http://codereview.chromium.org/437063/diff/11/13
File webkit/tools/layout_tests/layout_package/apache_http_server.py (right):

http://codereview.chromium.org/437063/diff/11/13#newcode102
webkit/tools/layout_tests/layout_package/apache_http_server.py:102:
time.sleep(3)
On 2009/11/25 21:38:08, ojan wrote:
> I now check if stderr had any text and assume that means it didn't start up
> properly. Does that seem fine? It worked for the one case I was hitting
locally
> at least.

Seems fine for now.  We can add more stuff as it comes up.

http://codereview.chromium.org/437063/diff/5003/5005#newcode121
webkit/tools/layout_tests/layout_package/apache_http_server.py:121: 
Nit: There should still probably be some sleep here, like maybe .1?