Increate test server start timeout.

Increate test server start timeout.

BUG=106543
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=113729

[sky, 2011-12-09 01:22:29]

LGTM

[Evan Stade, 2011-12-09 02:13:24]

rubber stamp

[Paweł Hajdan Jr., 2011-12-09 11:20:20]

Please read
https://groups.google.com/a/chromium.org/group/chromium-dev/msg/14257a8f1b932...
where I describe some problems with this change.

At the _very_ least please make the timeout consistent on POSIX.

[Sergey Ulanov, 2011-12-09 19:24:53]

I think it's a bug that the test server takes so much time to start. I don't
think we need the same change for POSIX because the service starts much faster
on mac and linux. 
I've opened http://crbug.com/107011 to track this issue.

On 2011/12/09 11:20:20, Paweł Hajdan Jr. wrote:
> Please read
>
https://groups.google.com/a/chromium.org/group/chromium-dev/msg/14257a8f1b932...
> where I describe some problems with this change.
> 
> At the _very_ least please make the timeout consistent on POSIX.

[Paweł Hajdan Jr., 2011-12-12 17:15:24]

On 2011/12/09 19:24:53, sergeyu wrote:
> I think it's a bug that the test server takes so much time to start.

Right, this is what this whole thing is about: increasing the timeouts does not
actually fix anything.

> I don't
> think we need the same change for POSIX because the service starts much faster
> on mac and linux. 

I don't see any reason why Linux/Mac wouldn't be affected, at least until we can
understand why the problem happens on Windows and why it appears/disappears
"mysteriously".

If in doubt, please see net/test/OWNERS.

By the way, could you please extract the part that determines how long the
timeout is to the common part of the code? Something like "int
TestServer::start_timeout_ms() { return TestTimeouts::action_max_timeout_ms();
}".

[Sergey Ulanov, 2011-12-12 18:33:29]

On Mon, Dec 12, 2011 at 9:15 AM, <phajdan.jr@chromium.org> wrote:

> On 2011/12/09 19:24:53, sergeyu wrote:
>
>> I think it's a bug that the test server takes so much time to start.
>>
>
> Right, this is what this whole thing is about: increasing the timeouts
> does not
> actually fix anything.


That's why I opened http://crbug.com/107011 .


>
>
>  I don't
>> think we need the same change for POSIX because the service starts much
>> faster
>> on mac and linux.
>>
>
> I don't see any reason why Linux/Mac wouldn't be affected, at least until
> we can
> understand why the problem happens on Windows and why it appears/disappears
> "mysteriously".
>

It doesn't appears/disappears mysteriously. Test server is always slow to
start on Windows. Currently on the build bot it takes about 10 seconds to
start (the start time most probably has Gaussian distribution). Many tests
succeed because test server manages to start under 10 seconds (e.g. 9.8
seconds), other don't because the server fails to start in 10 seconds. All
tests that use test server still take more than 10 seconds to complete on
Windows, usually from 12 to 15 seconds. The same tests complete on Mac in
under 5 seconds.



> If in doubt, please see net/test/OWNERS.
>
> By the way, could you please extract the part that determines how long the
> timeout is to the common part of the code? Something like "int
> TestServer::start_timeout_ms() { return TestTimeouts::action_max_**
> timeout_ms();
> }".
>
>
http://codereview.chromium.**org/8872050/<http://codereview.chromium.org/8872...
>

[Paweł Hajdan Jr., 2011-12-14 08:46:02]

On 2011/12/12 18:33:29, sergeyu wrote:
> That's why I opened http://crbug.com/107011 .

It's closed as a duplicate. That's why I said it doesn't fix anything: in fact,
the old problem is now again there, _and_ the timeouts are not consistent for
POSIX and Windows.

> It doesn't appears/disappears mysteriously. Test server is always slow to
> start on Windows.

Is it really always or "when I looked at the logs"? It makes sense anyway, but
still doesn't explain why the test server started failing a lot on that
particular bot and not others, then stopped for a while possibly (I don't
remember) and started again.

> Many tests succeed because test server manages to start under 10 seconds (e.g.
9.8
> seconds), other don't because the server fails to start in 10 seconds. All
> tests that use test server still take more than 10 seconds to complete on
> Windows, usually from 12 to 15 seconds. The same tests complete on Mac in
> under 5 seconds.

Whatever happens on non-Windows, I'd really like the timeouts to be consistent
and predictable. I'll think more about better solution.

Please finally *do* fix the review comment made by the owner of code you've
modified.