Fedora Modular bikeshed compose report: changes

[Przemek Klosowski]

On 12/02/2017 01:15 PM, Fedora Rawhide Report sent this report, quoted 
in its entirety:
>  _______________________________________________
>  devel mailing list -- devel(a)lists.fedoraproject.org
>  To unsubscribe send an email to devel-leave(a)lists.fedoraproject.org 
Could someone associated with Fedora Modular bikeshed look into 
this---the automated reports come out empty for as many cycles as I can 
remember. Maybe if there aren't any changes, don't send anything?

[Adam Williamson]

On Mon, 2017-12-04 at 16:01 -0500, Przemek Klosowski wrote:
>  On 12/02/2017 01:15 PM, Fedora Rawhide Report sent this report,
> quoted 
>  in its entirety:
>  > _______________________________________________
>  > devel mailing list -- devel(a)lists.fedoraproject.org
>  > To unsubscribe send an email to devel-leave(a)lists.fedoraproject.org
>  
>  Could someone associated with Fedora Modular bikeshed look into 
>  this---the automated reports come out empty for as many cycles as I can 
>  remember. Maybe if there aren't any changes, don't send anything? 
It's more likely that this happens because the code that's intended to
produce reports for other composes also fires for Bikeshed composes,
but doesn't know how to analyze modular composes properly. That code is
https://pagure.io/compose-utils/blob/master/f/compose_utils/changelog.py
. The reports for F27 modular composes do work, so I suspect the
'Bikeshed' name is likely related here. (I really, really think this
"Bikeshed" name was a terrible idea - it's bad enough that we treat
'Rawhide' as a version in many circumstances, meaning we have reams of
code all over the place which has to handle "any integer plus the
string 'Rawhide'" somehow, without arbitrarily inventing *new* version
strings and making the problem worse.)
-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net
http://www.happyassassin.net

[Adam Williamson]

On Mon, 2017-12-04 at 13:19 -0800, Adam Williamson wrote:
>  On Mon, 2017-12-04 at 16:01 -0500, Przemek Klosowski wrote:
>  > On 12/02/2017 01:15 PM, Fedora Rawhide Report sent this report, quoted 
>  > in its entirety:
>  > > _______________________________________________
>  > > devel mailing list -- devel(a)lists.fedoraproject.org
>  > > To unsubscribe send an email to devel-leave(a)lists.fedoraproject.org
>  > 
>  > Could someone associated with Fedora Modular bikeshed look into 
>  > this---the automated reports come out empty for as many cycles as I can 
>  > remember. Maybe if there aren't any changes, don't send anything?
>  
>  It's more likely that this happens because the code that's intended to
>  produce reports for other composes also fires for Bikeshed composes,
>  but doesn't know how to analyze modular composes properly. That code is
>  https://pagure.io/compose-utils/blob/master/f/compose_utils/changelog.py
>  . 
Well, now I refresh my memory, it's a bit more than that. I've just
dived in and debugged this, along with puiterwijk. Long story short,
it's basically because the nightly modular compose script doesn't bail
out when the compose fails, but continues with all the steps it should
only do for successful composes - including sending out this
'changelog' mail. The nightly non-modular compose script, by
comparison, checks the return code from pungi-koji and bails out if
it's non-zero, skipping all the steps that should only happen for a
successful compose.

We'll get that fixed somehow.
-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net
http://www.happyassassin.net

[Adam Williamson]

On Mon, 2017-12-04 at 13:59 -0800, Adam Williamson wrote:
>  
>  We'll get that fixed somehow. 
https://pagure.io/pungi-fedora/pull-request/483

so, this shouldn't happen any more. :)
-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net
http://www.happyassassin.net

[Adam Williamson]

On Mon, 2017-12-04 at 16:01 -0500, Przemek Klosowski wrote:
>  On 12/02/2017 01:15 PM, Fedora Rawhide Report sent this report,
> quoted 
>  in its entirety:
>  > _______________________________________________
>  > devel mailing list -- devel(a)lists.fedoraproject.org
>  > To unsubscribe send an email to devel-leave(a)lists.fedoraproject.org
>  
>  Could someone associated with Fedora Modular bikeshed look into 
>  this---the automated reports come out empty for as many cycles as I can 
>  remember. Maybe if there aren't any changes, don't send anything? 
BTW, in case you were wondering, I also looked into the 27 modular
compose report emails - why they're often full of zeroes, and why we
seem to keep getting ones for 20171204.n.1 .

Turns out the reason we keep getting emails for 20171204.n.1 is the
same reason we were getting lots of mails for bikeshed composes: the
compose script wasn't set to quit (skipping the 'send out emails' step)
if the compose failed. And the way the script works is, it reads out
the "current" compose ID before starting the compose, then when the
compose is done, it reads the "current" compose ID again, expecting it
to be the ID of the compose that just finished. Then it runs the diff
against those two compose IDs. But if the compose *failed*, then it
won't be - the "current" compose ID will still be the same as it was.

So basically, since the last successful 27-modular compose was
20171204.n.1, every time a compose fails, the script winds up
generating and sending out the 'diff' between 20171204.n.1 (which it
sees as the "old" compose) and...20171204.n.1 (which it sees as the
"new" compose, since the actual new compose failed).

I've sent a PR to stop it doing this:
https://pagure.io/pungi-fedora/pull-request/488

Of course when it winds up diffing a compose against itself the results
are all zeroes (no difference). But it's also worth noting that
sometimes there will be a "correct" compose report for modular which is
all zeroes, because the content doesn't always change every day, as it
almost always does in regular Rawhide and Branched. So even with this
change, there will be some all-zero mails. I don't think it makes sense
to not send out mails when there are no changes, because "there was a
successful new compose, but nothing changed in it" is valid information
that may be of use to someone.
-- 
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net
http://www.happyassassin.net