On Valgrind bots, print a message every 20 minutes

On Valgrind bots, print a message every 20 minutes

I believe that the timeout failure on the Valgrind bot occurs
because we skip many tests consecutively without printing any
output.  Skipping these tests requires non-trivial work.

Printing a message every 20 minutes will avoid timeouts.

BUG=skia:4740
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1554193002

Committed: https://skia.googlesource.com/skia/+/c149f0e36e0ca0ff35ec6e2f79419b58d529db40

[msarett, 2016-01-04 17:43:37]

Description was changed from

==========
On Valgrind bots, print a message on skipped benchmarks

I believe that the timeout failure on the Valgrind bot occurs
because we skip many tests consecutively without printing any
output.  Skipping these tests requires non-trivial work.

This is a quick fix to verify that this is the actual cause
of the failure.  I plan to follow up and allow us to skip
tests more efficiently.

BUG=skia:4740
==========

to

==========
On Valgrind bots, print a message on skipped benchmarks

I believe that the timeout failure on the Valgrind bot occurs
because we skip many tests consecutively without printing any
output.  Skipping these tests requires non-trivial work.

This is a quick fix to verify that this is the actual cause
of the failure.  I plan to follow up and allow us to skip
tests more efficiently.

BUG=skia:4740
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
==========

[msarett, 2016-01-04 17:45:54]

msarett@google.com changed reviewers:
+ borenet@google.com, egdaniel@google.com, scroggo@google.com

[scroggo, 2016-01-04 17:56:35]

Alternatively, why not do something similar to what DM does? See
start_keepalive() in DM.cpp. It prints a message every once in a while stating
what is currently running. Or will its extra thread interfere with nanobench
timing? I guess we could only do it in debug/only for valgrind.

[egdaniel, 2016-01-04 18:00:57]

On 2016/01/04 17:56:35, scroggo wrote:
> Alternatively, why not do something similar to what DM does? See
> start_keepalive() in DM.cpp. It prints a message every once in a while stating
> what is currently running. Or will its extra thread interfere with nanobench
> timing? I guess we could only do it in debug/only for valgrind.

Whatever printing we do should only be done in valgrind. Before when it was
always printing, there were a lot of complaints while developing cause of the
spam. Thus I support this method of it being behind a flag passed into
nanobench. Now if we do a DM style print every minute or this approach of just
print when skipping I don't really care as long as the later is enough prints to
get the valgrind bot working again.

[scroggo, 2016-01-04 18:15:32]

On 2016/01/04 18:00:57, egdaniel wrote:
> On 2016/01/04 17:56:35, scroggo wrote:
> > Alternatively, why not do something similar to what DM does? See
> > start_keepalive() in DM.cpp. It prints a message every once in a while
stating
> > what is currently running. Or will its extra thread interfere with nanobench
> > timing? I guess we could only do it in debug/only for valgrind.
> 
> Whatever printing we do should only be done in valgrind. Before when it was
> always printing, there were a lot of complaints while developing cause of the
> spam. Thus I support this method of it being behind a flag passed into
> nanobench. Now if we do a DM style print every minute or this approach of just
> print when skipping I don't really care as long as the later is enough prints
to
> get the valgrind bot working again.

My vote would be to do it every once in a while (as infrequently as necessary to
fix valgrind - it looks like it's every 300 seconds in DM). I suspect that would
print less extra lines, though I could be completely wrong (valgrind runs really
slowly...).

[msarett, 2016-01-04 18:31:08]

Patchset #3 (id:40001) has been deleted

[msarett, 2016-01-04 18:31:13]

Patchset #3 (id:60001) has been deleted

[msarett, 2016-01-04 18:38:39]

Patchset #3 (id:80001) has been deleted

[msarett, 2016-01-04 18:41:35]

I would agree that printing less is better.  Now we just
print a message every 20 minutes.

While the other fix was intended to be temporary, I think
that this fix is sufficiently non-annoying to be permanent.

I think having an extra thread is a bit strange, but is
probably fine if we only use it when running with valgrind.

I still plan to try to make the BRD bench setup more efficient as a follow-up.

[msarett, 2016-01-04 18:42:34]

Description was changed from

==========
On Valgrind bots, print a message on skipped benchmarks

I believe that the timeout failure on the Valgrind bot occurs
because we skip many tests consecutively without printing any
output.  Skipping these tests requires non-trivial work.

This is a quick fix to verify that this is the actual cause
of the failure.  I plan to follow up and allow us to skip
tests more efficiently.

BUG=skia:4740
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
==========

to

==========
On Valgrind bots, print a message on skipped benchmarks

I believe that the timeout failure on the Valgrind bot occurs
because we skip many tests consecutively without printing any
output.  Skipping these tests requires non-trivial work.

Printing a message every 20 minutes will avoid timeouts.

BUG=skia:4740
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
==========

[msarett, 2016-01-04 18:42:53]

Description was changed from

==========
On Valgrind bots, print a message on skipped benchmarks

I believe that the timeout failure on the Valgrind bot occurs
because we skip many tests consecutively without printing any
output.  Skipping these tests requires non-trivial work.

Printing a message every 20 minutes will avoid timeouts.

BUG=skia:4740
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
==========

to

==========
On Valgrind bots, print a message every 20 minutes

I believe that the timeout failure on the Valgrind bot occurs
because we skip many tests consecutively without printing any
output.  Skipping these tests requires non-trivial work.

Printing a message every 20 minutes will avoid timeouts.

BUG=skia:4740
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
==========

[scroggo, 2016-01-04 19:18:52]

lgtm

[msarett, 2016-01-04 19:19:27]

The CQ bit was checked by msarett@google.com

[commit-bot: I haz the power, 2016-01-04 19:19:30]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1554193002/100001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1554193002/100001

[commit-bot: I haz the power, 2016-01-04 19:35:46]

Description was changed from

==========
On Valgrind bots, print a message every 20 minutes

I believe that the timeout failure on the Valgrind bot occurs
because we skip many tests consecutively without printing any
output.  Skipping these tests requires non-trivial work.

Printing a message every 20 minutes will avoid timeouts.

BUG=skia:4740
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
==========

to

==========
On Valgrind bots, print a message every 20 minutes

I believe that the timeout failure on the Valgrind bot occurs
because we skip many tests consecutively without printing any
output.  Skipping these tests requires non-trivial work.

Printing a message every 20 minutes will avoid timeouts.

BUG=skia:4740
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...

Committed:
https://skia.googlesource.com/skia/+/c149f0e36e0ca0ff35ec6e2f79419b58d529db40
==========

[commit-bot: I haz the power, 2016-01-04 19:35:47]

Committed patchset #3 (id:100001) as
https://skia.googlesource.com/skia/+/c149f0e36e0ca0ff35ec6e2f79419b58d529db40

[mtklein, 2016-01-05 02:44:21]

mtklein@google.com changed reviewers:
+ mtklein@google.com

[mtklein, 2016-01-05 02:44:21]

For what it's worth, I agree this is a good way to go.  That extra thread will
never cause problems in nanobench, even on non-Valgrind runs.  (Not unless it's
something weird like a benchmark of SkDebugf...)