Rietveld Code Review Tool
Help | Bug tracker | Discussion group | Source code | Sign in
(211)

Issue 14604043: code review 14604043: runtime/pprof: disable flaky TestGoroutineSwitch on windows (Closed)

Can't Edit
Can't Publish+Mail
Start Review
Created:
10 years, 5 months ago by brainman
Modified:
10 years, 5 months ago
Reviewers:
rsc
CC:
golang-dev, rsc
Visibility:
Public.

Description

runtime/pprof: disable flaky TestGoroutineSwitch on windows Update issue 6417

Patch Set 1 #

Patch Set 2 : diff -r 52ce02a2718e https://go.googlecode.com/hg/ #

Patch Set 3 : diff -r 52ce02a2718e https://go.googlecode.com/hg/ #

Patch Set 4 : diff -r 09706910ff2c https://go.googlecode.com/hg/ #

Unified diffs Side-by-side diffs Delta from patch set Stats (+3 lines, -0 lines) Patch
M src/pkg/runtime/pprof/pprof_test.go View 1 1 chunk +3 lines, -0 lines 0 comments Download

Messages

Total messages: 6
brainman
Hello golang-dev@googlegroups.com, I'd like you to review this change to https://go.googlecode.com/hg/
10 years, 5 months ago (2013-10-10 23:00:02 UTC) #1
rsc
LGTM It's not so much flaky as that multiprocessor profiling doesn't work correctly on Windows, ...
10 years, 5 months ago (2013-10-14 13:45:42 UTC) #2
brainman
*** Submitted as https://code.google.com/p/go/source/detail?r=6f59535f589d *** runtime/pprof: disable flaky TestGoroutineSwitch on windows Update issue 6417 R=golang-dev, ...
10 years, 5 months ago (2013-10-15 02:00:12 UTC) #3
brainman
On 2013/10/14 13:45:42, rsc wrote: > > It's not so much flaky as that multiprocessor ...
10 years, 5 months ago (2013-10-15 02:01:10 UTC) #4
rsc
On Mon, Oct 14, 2013 at 10:01 PM, <alex.brainman@gmail.com> wrote: > On 2013/10/14 13:45:42, rsc ...
10 years, 5 months ago (2013-10-15 18:37:30 UTC) #5
brainman
10 years, 5 months ago (2013-10-16 00:51:35 UTC) #6
Message was sent while issue was closed.
On 2013/10/15 18:37:30, rsc wrote:
> 
> With a loop doing nothing but calling Gosched, I expect at least some of
> the profiling events to interrupt the scheduler. ...

I can see that happens. No problem here.

> ... Those should be recorded
> by sigprof as the actual PC with a parent of "System". ...

This won't happen in this case, because sigprof is not called:

if(gp != nil && gp != mp->g0 && gp->status != Gsyscall) ...

(in https://code.google.com/p/go/source/browse/src/pkg/runtime/os_windows.c#405)
is never true. We tried changing this condition, but that only made things
worse.

> ... The test is looking
> for at least one such record.

I think the test is wrong about assuming that. Unless you want to arrange
windows profilem to do something special here, I don't see how we can reliably
run this test on windows.

Alex
Sign in to reply to this message.

Powered by Google App Engine
RSS Feeds Recent Issues | This issue
This is Rietveld f62528b