Significantly increase timeouts to reduce flakiness...

Significantly increase timeouts to reduce flakiness

Several tests now "plan for success" and will terminate rapidly if all goes
well (and will hang if there really is a problem).

One test I just upped by a factor of 5, hoping that I'll get a timem slice
within a 5 second window, rather than assuming that I'll get one within
a 1 second window.  I could go further, but the test now takes over 5  seconds
to run, which seems like a lot of time for a trivial test.

bug=5723
r=maruel

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=7744

[M-A Ruel, 2008-12-31 13:36:21]

Can you replace the long sleep by a semi-busy loop? (E.g. loop with a very short
sleep) The goal is to have the test complete as fast as possible. The rest lgtm.

[M-A Ruel, 2009-01-02 21:04:25]

http://codereview.chromium.org/16495/diff/202/4
File base/watchdog_unittest.cc (right):

http://codereview.chromium.org/16495/diff/202/4#newcode100
Line 100: WatchdogCounter watchdog(TimeDelta::FromSeconds(5), "Enabled", true);
Ok I understand the idea and why you did this this way. Sorry I was on my phone
when I did the first review, it's a bit awkward. :)

I have one idea to fix it and keep the timer shorter, tell me what you think:

- Retrieve the current time
- Arm the watchdog for 1 second
- Sleep 100 ms
- Look how much time it actually slept, if the time diff < 0.9 sec, expect no
alarm, otherwise don't bother, it'll probably be flaky anyway.
- Sleep for 1500 (?)
- Expect alarm
- Continue the rest of the test...

Maybe it'd be flaky in its own way too. What do you think?
I'd like to reduce the 1500 sleep too but for now I think it'd be better this
way.

http://codereview.chromium.org/16495/diff/202/4#newcode111
Line 111: // It should almost instantly go off, but certainly in less than a
second.
5 minutes

[jar (doing other things), 2009-01-07 00:32:05]

http://codereview.chromium.org/16495/diff/202/4
File base/watchdog_unittest.cc (right):

http://codereview.chromium.org/16495/diff/202/4#newcode100
Line 100: WatchdogCounter watchdog(TimeDelta::FromSeconds(5), "Enabled", true);
I could do something of that sort.  The problem is that I'll abandon this test
"when the test system is slow."  The good news is that I can make it not be
flaky.... the bad news is that I *might* be moving into the "never test"
situation.  In fact, I'll probably get into a mode where a failure might not be
noted for several runs, causing confusion about the cause of the failure :-/.

IMO, it is not that clear such would be an improvement.  In addition, although I
don't think we can afford an extra 5 seconds on every test... we can afford it
on the few tests that are multi-threaded, and hence are fundamentally vulnerable
to race/flaky failures.

On 2009/01/02 21:04:25, M-A wrote:
> Ok I understand the idea and why you did this this way. Sorry I was on my
phone
> when I did the first review, it's a bit awkward. :)
> 
> I have one idea to fix it and keep the timer shorter, tell me what you think:
> 
> - Retrieve the current time
> - Arm the watchdog for 1 second
> - Sleep 100 ms
> - Look how much time it actually slept, if the time diff < 0.9 sec, expect no
> alarm, otherwise don't bother, it'll probably be flaky anyway.
> - Sleep for 1500 (?)
> - Expect alarm
> - Continue the rest of the test...
> 
> Maybe it'd be flaky in its own way too. What do you think?
> I'd like to reduce the 1500 sleep too but for now I think it'd be better this
> way.

http://codereview.chromium.org/16495/diff/202/4#newcode111
Line 111: // It should almost instantly go off, but certainly in less than a
second.
On 2009/01/02 21:04:25, M-A wrote:
> 5 minutes

Done.

[M-A Ruel, 2009-01-07 16:28:57]

lgtm