Mailing List Archive

On 5/29/21 8:05 PM, Ricardo Signes wrote:
> I was looking at https://perl5.test-smoke.org/
> <https://perl5.test-smoke.org/> earlier today, and looking at some of
> our branches, and just sort of generally looking around earlier, and I
> came to wonder:  Is anybody still smoking the smoke-me branches?
>
> They were useful.  Can someone who has been paying attention the whole
> time fill me (and us) in as to their status?
>
> Thanks!
>

Yes, but it's complicated.

IIRC, perl5.test-smoke.org died last September. Disk failure. Many
years of data lost. Eventually a replacement was put together, but it's
not receiving (or, perhaps more precisely, displaying) all the reports
being generated.

And many fewer testers are sending reports. Go to TonyC's aggregator,
http://perl.develop-help.com/. Note how the overwhelming majority of
reports are coming from one tester: Carlos Guevara. And I believe that
one consequence of the disk failure last September is that develop-help
is only picking up reports transmitted via email, that is, sent to some
perl mailing list. develop-help is *not* picking up reports that are
being transmitted via HTTP::Tiny. test-smoke.org *is* picking up reports
transmitted via HTTP::Tiny, but I don't know whether it is picking up
reports transmitted via email.

Let me put this concretely. I have never been able to figure out how to
send reports via email, so all my reports have been transmitted via
http. Up until September, my reports would appear on *both*
develop-help.com and test-smoke.org. Now they *only* appear on
test-smoke.org.

Also, the "new" test-smoke.org is (I have been told) just a VM somewhere
and is often subject to 504 timeout errors. The people who maintain
that have, I believe, time constraints. Recommend consulting with TonyC
first, then Tux.

Thank you very much.
Jim Keenan

On Sat, May 29, 2021 at 10:08:18PM -0400, James E Keenan wrote:
> On 5/29/21 8:05 PM, Ricardo Signes wrote:
> > I was looking at https://perl5.test-smoke.org/
> > <https://perl5.test-smoke.org/>?earlier today, and looking at some of
> > our branches, and just sort of generally looking around earlier, and I
> > came to wonder:? Is anybody still smoking the smoke-me branches?
> >
> > They were useful.? Can someone who has been paying attention the whole
> > time fill me (and us) in as to their status?
> >
> > Thanks!
> >
>
> Yes, but it's complicated.
>
> IIRC, perl5.test-smoke.org died last September. Disk failure. Many years
> of data lost. Eventually a replacement was put together, but it's not
> receiving (or, perhaps more precisely, displaying) all the reports being
> generated.

Unless it's changed, when Test::Smoke makes one attempt to send a
report by http(s), and if that fails no further attempts are made.

> And many fewer testers are sending reports. Go to TonyC's aggregator,
> http://perl.develop-help.com/. Note how the overwhelming majority of
> reports are coming from one tester: Carlos Guevara. And I believe that one
> consequence of the disk failure last September is that develop-help is only
> picking up reports transmitted via email, that is, sent to some perl mailing
> list. develop-help is *not* picking up reports that are being transmitted
> via HTTP::Tiny. test-smoke.org *is* picking up reports transmitted via
> HTTP::Tiny, but I don't know whether it is picking up reports transmitted
> via email.

My aggregator had test-smoke.org fetching disabled because it was only
failing (and sending me failure messages)

It still seems to be only failing, I tried loading both the API
endpoint and the home page a few times and got 502s each time.

I've set it to fetch daily, maybe we'll get lucky.

test-smoke.org only receives message sent via HTTP::Tiny.

> Let me put this concretely. I have never been able to figure out how to
> send reports via email, so all my reports have been transmitted via http.
> Up until September, my reports would appear on *both* develop-help.com and
> test-smoke.org. Now they *only* appear on test-smoke.org.

It used to be the default, and if it's useful, the appropriate parts from
my old smokecurrent_config are:

'mail' => '1',
'mail_type' => 'MIME::Lite',
'mserver' => 'localhost',
'from' => 'tony@develop-help.com',
'to' => 'daily-build-reports@perl.org',

Mail is going to remain the fastest way to get reports into
perl.develop-help.com - I'm subscribed to the daily-build-reports list
and a procmail rule pushes the message to perl.develop-help.com as
soon as I receive it[1][2], while I have to poll for test-smoke.org for
reports.

> Also, the "new" test-smoke.org is (I have been told) just a VM somewhere and
> is often subject to 504 timeout errors. The people who maintain that have,
> I believe, time constraints. Recommend consulting with TonyC first, then
> Tux.

perl.develop-help.com is also a VM (at Linode Sydney).

I have no control over test-smoke.org.

I've considered offering to host it (on the same Linode), but IIRC it
needed occasional restarts even when it was hosted at booking.

Tony

[1] with retries
[2] I also check nntp.perl.org daily for any missing reports.

>
>
>
> Mail is going to remain the fastest way to get reports into
> perl.develop-help.com - I'm subscribed to the daily-build-reports list
> and a procmail rule pushes the message to perl.develop-help.com as
> soon as I receive it[1][2], while I have to poll for test-smoke.org for
> reports.
>
>
FWIW, I'd really love to not have these delivered via email to
daily-build-reports. I've already got some ugly hacks in place to keep it
working. Gigabytes of things people look at once, if ever, is not a great
use of limited resources. A compromise *might* be deleting any more than
$TIMEPERIOD old from the archives.

-R

On Mon, May 31, 2021 at 01:44:47PM -0700, Robert wrote:
> >
> >
> >
> > Mail is going to remain the fastest way to get reports into
> > perl.develop-help.com - I'm subscribed to the daily-build-reports list
> > and a procmail rule pushes the message to perl.develop-help.com as
> > soon as I receive it[1][2], while I have to poll for test-smoke.org for
> > reports.
> >
> >
> FWIW, I'd really love to not have these delivered via email to
> daily-build-reports. I've already got some ugly hacks in place to keep it
> working. Gigabytes of things people look at once, if ever, is not a great
> use of limited resources. A compromise *might* be deleting any more than
> $TIMEPERIOD old from the archives.

That wouldn't cause a problem for me, unless the nntp message numbers
were reset.

I know the intent was for the smoke reports database to replace the
mailing list, but in my experience it's been a bit too unreliable for
that.

Tony

On Sat, May 29, 2021 at 08:05:56PM -0400, Ricardo Signes wrote:
> I was looking at https://perl5.test-smoke.org/ earlier today, and
> looking at some of our branches, and just sort of generally looking
> around earlier, and I came to wonder: Is anybody still smoking the
> smoke-me branches?

When I set up my sparc64 smoker with Test::Smoke I couldn't figure out
how to make it do smoke-me branches. I did ask in #perl-qa for
direction but still wasn't able to figure out how to make it do that.

> They were useful. Can someone who has been paying attention the whole
> time fill me (and us) in as to their status?

If someone tells me how, I can make my smoker do it.

My smoker has lots of fairly slow processors, so if it was possible, I
would love to be able to do a smoke where I do all the different builds
there in parallel, but (I can't recall the details at the moment) there
was some collision with a file in /tmp or someplace that meant parallel
perl tests didn't work.

l8rZ,
--
andrew - http://afresh1.com

"Programming today is a race between software engineers striving to
build bigger and better idiot-proof programs, and the Universe
trying to produce bigger and better idiots. So far, the Universe is
winning." -- Rich Cook

On 6/5/21 3:18 PM, Andrew Hewus Fresh wrote:
> On Sat, May 29, 2021 at 08:05:56PM -0400, Ricardo Signes wrote:
>> I was looking at https://perl5.test-smoke.org/ earlier today, and
>> looking at some of our branches, and just sort of generally looking
>> around earlier, and I came to wonder: Is anybody still smoking the
>> smoke-me branches?
>
> When I set up my sparc64 smoker with Test::Smoke I couldn't figure out
> how to make it do smoke-me branches.

The first entry in smokecurrent.gitbranch is what get's smoked.
Personally, I start all smoke-test runs manually, so I simply manually
set the first line to the branch I want to smoke. Others may be able to
tell you about setting up cron jobs to smoke all branches upon their
first appearance.

I did ask in #perl-qa for
> direction but still wasn't able to figure out how to make it do that.
>

There is also #smoke

>> They were useful. Can someone who has been paying attention the whole
>> time fill me (and us) in as to their status?
>
> If someone tells me how, I can make my smoker do it.
>
> My smoker has lots of fairly slow processors, so if it was possible, I
> would love to be able to do a smoke where I do all the different builds
> there in parallel, but (I can't recall the details at the moment) there
> was some collision with a file in /tmp or someplace that meant parallel
> perl tests didn't work.
>

I have never attempted parallel runs but, yes, that would be a nice
feature. The Test-Smoke codebase is large. I have never been able to
hold it in my head enough to figure out the complete workflow or to
attempt more than modest refactorings. I suspect a lot of it is
superfluous or outdated, but I can't be sure which parts.

Thank you very much.
Jim Keenan

On Sat, Jun 05, 2021 at 03:54:55PM -0400, James E Keenan wrote:
> On 6/5/21 3:18 PM, Andrew Hewus Fresh wrote:
> > On Sat, May 29, 2021 at 08:05:56PM -0400, Ricardo Signes wrote:
> > > I was looking at https://perl5.test-smoke.org/ earlier today, and
> > > looking at some of our branches, and just sort of generally looking
> > > around earlier, and I came to wonder: Is anybody still smoking the
> > > smoke-me branches?
> >
> > When I set up my sparc64 smoker with Test::Smoke I couldn't figure out
> > how to make it do smoke-me branches.
>
> The first entry in smokecurrent.gitbranch is what get's smoked. Personally,
> I start all smoke-test runs manually, so I simply manually set the first
> line to the branch I want to smoke. Others may be able to tell you about
> setting up cron jobs to smoke all branches upon their first appearance.

If my other PRs had at gotten some traction (even just rejected) I might
already have submitted one to support multiple branches in that
gitbranch file, with a syntax for matching branches with a regex that
would loop over all the remote branches and smoke any new commits on
matching branches.

https://github.com/abeltje/Test-Smoke/pulls

Is there a better place to send Test::Smoke patches or do we need a
new Test::Smoke::Simple?



> I did ask in #perl-qa for
> > direction but still wasn't able to figure out how to make it do that.
> >
>
> There is also #smoke

TIL (Today I Learned)


> > > They were useful. Can someone who has been paying attention the whole
> > > time fill me (and us) in as to their status?
> >
> > If someone tells me how, I can make my smoker do it.
> >
> > My smoker has lots of fairly slow processors, so if it was possible, I
> > would love to be able to do a smoke where I do all the different builds
> > there in parallel, but (I can't recall the details at the moment) there
> > was some collision with a file in /tmp or someplace that meant parallel
> > perl tests didn't work.
> >
>
> I have never attempted parallel runs but, yes, that would be a nice feature.
> The Test-Smoke codebase is large. I have never been able to hold it in my
> head enough to figure out the complete workflow or to attempt more than
> modest refactorings. I suspect a lot of it is superfluous or outdated, but
> I can't be sure which parts.

What Test::Smoke does is figure out sets of flags to build perl with,
then build perl with each of them and then for each build runs tests
with several different "perlio" layers. I didn't look too hard at how
it decided which layers to test, but I had a preliminary patch that ran
all those tests in parallel, but unfortunately as I said, running perl
tests in parallel fails due to some sort of conflict.

The next step would have been to build perl in parallel in separate
working directories and have even more parallelization, but since the
first step didn't work, I never made it this far.

I may have been able to solve the failures by setting $ENV{TMPDIR} or
something (I don't recall the failure like I said) but with no
likelihood of getting it merged upstream, I gave up.



l8rZ,
--
andrew - http://afresh1.com

Unix is very simple,
but it takes a genius to understand the simplicity.
-- Dennis Ritchie