resultsdb roadmap

[Róman Joost]

Hi,

I've already tried to ask on #fedora-qa, but I think the mailing list is
the better medium to discuss this. Thanks to jskladen for answering.

We've been looking into Taskotron and Resultsdb for a while and would
like to deploy an instance running RPMGrill[1] and become contributors.

I assume, running rpmgrill in Taskotron could mean, that each rpmgrill
test result translates to an entry in Resultsdb as an outcome (PASSED,
FAILED). Yet the question for the developer as to why it failed is
important.

I have mainly three main questions around Taskatron and specifically
Resultsdb:

1. Ed Santiago is currently running RPMGrill here:

    http://rpmgrill-fc20.edsantiago.com:5000/recent

  which in terms of presenting results to the user provides a different
  experience. The results are grouped by test and provide more insight
  as to why a test failed. Test results in Resultsdb currently seem to
  have only one outcome and technical details are left on the build
  master.

  How does a representation like RPMGrill translate into Resultsdb? Are
  there plans/ideas on how to provide a more diversified result
  representation in Resultsdb (e.g. error reason, hint on how to solve
  an error).

2. Are there plans/ideas about implementing a (fulltext-) search over
  results?

3. Are there plans/ideas on implementing a waiver mechanism?

[1] - https://git.fedorahosted.org/git/rpmgrill.git

PS: Yes - I'm aware that there is already a bitbucket repository of
rpmgrill for Taskotron, but have only had a quick glimpse at it.

Thanks and Kind Regards,
-- 
Róman Joost
Software Engineer, HSS - Software Engineering and Development (Brisbane)
email: rjoost(a)redhat.com | tz: UTC+10
irc: #hss #rpmdiff

[Kamil Paral]

>  Hi,
>  
>  I've already tried to ask on #fedora-qa, but I think the mailing list is
>  the better medium to discuss this. Thanks to jskladen for answering.
>  
>  We've been looking into Taskotron and Resultsdb for a while and would
>  like to deploy an instance running RPMGrill[1] and become contributors.
>  
>  I assume, running rpmgrill in Taskotron could mean, that each rpmgrill
>  test result translates to an entry in Resultsdb as an outcome (PASSED,
>  FAILED).  
If you are interested, we support more outcome keywords, like INFO or NEEDS_INSPECTION:
https://docs.qadevel.cloud.fedoraproject.org/libtaskotron/latest/library....

>  Yet the question for the developer as to why it failed is
>  important.
>  
>  I have mainly three main questions around Taskatron and specifically
>  Resultsdb:
>  
>  1. Ed Santiago is currently running RPMGrill here:
>  
>      http://rpmgrill-fc20.edsantiago.com:5000/recent
>  
>    which in terms of presenting results to the user provides a different
>    experience. The results are grouped by test and provide more insight
>    as to why a test failed. Test results in Resultsdb currently seem to
>    have only one outcome and technical details are left on the build
>    master.
>  
>    How does a representation like RPMGrill translate into Resultsdb? Are
>    there plans/ideas on how to provide a more diversified result
>    representation in Resultsdb (e.g. error reason, hint on how to solve
>    an error). 
At the moment, each result contains a link to the task log, which you can inspect, e.g.:
https://taskotron.fedoraproject.org/resultsdb/results?job_id=45746
points to
https://taskotron.fedoraproject.org/taskmaster//builders/x86_64/builds/36...

That log contains both taskotron execution info and above statements and task output. If
you know an undocumented trick, you can also strip down the path a bit arrive to the
buildbot job info page:
https://taskotron.fedoraproject.org/taskmaster//builders/x86_64/builds/36...
and there you have access to taskotron.log, which contains full debug info, which we use
for debugging libtaskotron issues. This is usually not needed by package maintainers, so
we don't advertise this too much yet.

I agree it would be great if taskotron could execute rpmgrill, and then upload the results
to your special application and nicely display them. Or alternatively generate the html
report locally. Unfortunately we don't have a support for storing and displaying
artifacts (e.g. a html page) at the moment -- but we plan to do it in the future.

If you want to have something working right now, I see several ways to go:

1. You could print out a text representation of the rpmgrill results into the log. You
could also post the results into your web applications, and then include a hyperlink in
the log in order to point people to a better visualization.

2. I guess we could change the "Log →" hyperlink in resultsdb to point to your
web application directly, rather than into our buildbot log. However, I see two concerns
with that, one is that accessing the task log itself would be non-trivial, and second that
displaying any results at all would be relying on your web app availability. If it went
down, nobody could see any results at all.

In the future, generating the html result locally and then showing it from resultsdb
interface is probably the way to go.

>  
>  2. Are there plans/ideas about implementing a (fulltext-) search over
>    results? 
I'll let others to respond to this, I have no idea.

>  
>  3. Are there plans/ideas on implementing a waiver mechanism? 
We will definitely need this, for most of the checks. But we haven't started yet
designing that feature, so it won't be available any time soon.


BTW, I talked to Miroslav Vadkerti on DevConf about rpmgrill. We should meet up again in
the following weeks and try to cook some actionable plan for rpmgrill in Taskotron.

Kamil

[Róman Joost]

Dear Kamil,

On Thu, Feb 12, 2015 at 03:41:38AM -0500, Kamil Paral wrote:
>  > [...]
>  > I assume, running rpmgrill in Taskotron could mean, that each rpmgrill
>  > test result translates to an entry in Resultsdb as an outcome (PASSED,
>  > FAILED). 
>  
>  If you are interested, we support more outcome keywords, like INFO or NEEDS_INSPECTION:
>
> https://docs.qadevel.cloud.fedoraproject.org/libtaskotron/latest/library....
Great. Thanks for the pointer.

>  > Yet the question for the developer as to why it failed is
>  > important.
>  > 
>  > I have mainly three main questions around Taskatron and specifically
>  > Resultsdb:
>  > 
>  > 1. Ed Santiago is currently running RPMGrill here:
>  > 
>  >     http://rpmgrill-fc20.edsantiago.com:5000/recent
>  > 
>  >   which in terms of presenting results to the user provides a different
>  >   experience. The results are grouped by test and provide more insight
>  >   as to why a test failed. Test results in Resultsdb currently seem to
>  >   have only one outcome and technical details are left on the build
>  >   master.
>  > 
>  >   How does a representation like RPMGrill translate into Resultsdb? Are
>  >   there plans/ideas on how to provide a more diversified result
>  >   representation in Resultsdb (e.g. error reason, hint on how to solve
>  >   an error).
>  
>  At the moment, each result contains a link to the task log, which you can inspect, e.g.:
>  https://taskotron.fedoraproject.org/resultsdb/results?job_id=45746
>  points to
>
> https://taskotron.fedoraproject.org/taskmaster//builders/x86_64/builds/36...
>  
>  That log contains both taskotron execution info and above statements and task output. If
> you know an undocumented trick, you can also strip down the path a bit arrive to the
> buildbot job info page:
>  https://taskotron.fedoraproject.org/taskmaster//builders/x86_64/builds/36...
>  and there you have access to taskotron.log, which contains full debug info, which we use
> for debugging libtaskotron issues. This is usually not needed by package maintainers, so
> we don't advertise this too much yet. 
Thanks.

>  I agree it would be great if taskotron could execute rpmgrill, and
>  then upload the results to your special application and nicely display
>  them. Or alternatively generate the html report locally. Unfortunately
>  we don't have a support for storing and displaying artifacts (e.g. a
>  html page) at the moment -- but we plan to do it in the future.
>  
>  If you want to have something working right now, I see several ways to
>  go:
>  
>  1. You could print out a text representation of the rpmgrill results
>  into the log. You could also post the results into your web
>  applications, and then include a hyperlink in the log in order to
>  point people to a better visualization.
>  
>  2. I guess we could change the "Log →" hyperlink in resultsdb to point
>  to your web application directly, rather than into our buildbot log.
>  However, I see two concerns with that, one is that accessing the task
>  log itself would be non-trivial, and second that displaying any
>  results at all would be relying on your web app availability. If it
>  went down, nobody could see any results at all.
>  
>  In the future, generating the html result locally and then showing it
>  from resultsdb interface is probably the way to go. 
That's were we would like to help to make this happen. Before discussing
any technical details, my first e-mail was aimed to figure out if this
is a desirable feature or not. If that's something which looks like it
is part of where resultsdb is going, I could start a technical
discussion in another thread?

>  > 3. Are there plans/ideas on implementing a waiver mechanism?
>  
>  We will definitely need this, for most of the checks. But we haven't
>  started yet designing that feature, so it won't be available any time
>  soon. 
Thanks for sharing.

>  
>  BTW, I talked to Miroslav Vadkerti on DevConf about rpmgrill. We
>  should meet up again in the following weeks and try to cook some
>  actionable plan for rpmgrill in Taskotron. 
Wow great!

Would it make sense to create a roadmap document or is there one
already?

Kind Regards,
-- 
Róman Joost
Software Engineer, HSS - Software Engineering and Development (Brisbane)
email: rjoost(a)redhat.com | tz: UTC+10
irc: #hss #rpmdiff

[Tim Flink]

On Mon, 9 Feb 2015 15:48:45 +1000
Róman Joost <rjoost(a)redhat.com&gt; wrote:

>  Hi,
>  
>  I've already tried to ask on #fedora-qa, but I think the mailing list
>  is the better medium to discuss this. Thanks to jskladen for
>  answering.
>  
>  We've been looking into Taskotron and Resultsdb for a while and would
>  like to deploy an instance running RPMGrill[1] and become
>  contributors.
>  
>  I assume, running rpmgrill in Taskotron could mean, that each rpmgrill
>  test result translates to an entry in Resultsdb as an outcome (PASSED,
>  FAILED). Yet the question for the developer as to why it failed is
>  important. 
I don't think that desire is unique to rpmgrill and something that we
decided to leave out of the initial Taskotron system.

My initial thought is that that a discrete state (PASS, FAIL,
NEEDS_INSPECTION etc.) a one-line-or-so summary and link to output
from rpmgrill would work well. Are there other things that you'd like
to see in terms of output?

>  I have mainly three main questions around Taskatron and specifically
>  Resultsdb:
>  
>  1. Ed Santiago is currently running RPMGrill here:
>  
>      http://rpmgrill-fc20.edsantiago.com:5000/recent
>  
>    which in terms of presenting results to the user provides a
>  different experience. The results are grouped by test and provide
>  more insight as to why a test failed. Test results in Resultsdb
>  currently seem to have only one outcome and technical details are
>  left on the build master. 
This was less of a desired final outcome for us than it was the easiest
way to get started.

I think that our current method of having gigantic text-only log files
is far from optimal (and something that we did years ago in AutoQA
before making the logs more easily accessible) but we haven't improved
the log accessibility/readability in taskotron yet.

>    How does a representation like RPMGrill translate into Resultsdb?
>  Are there plans/ideas on how to provide a more diversified result
>    representation in Resultsdb (e.g. error reason, hint on how to solve
>    an error). 
I think that having rpmgrill output static html would likely be the
best approach here unless you're planning to continue supporting the
webapp that you linked to (and from previous conversations, I'm under
the impression that was never intended to be a long-term thing).

>  2. Are there plans/ideas about implementing a (fulltext-) search
> over
>    results? 
Nothing written down, no but that's been on my wishlist since before
Taskotron really got started. Unless there's more demand for the
fulltext indexing than I think there is, we'd probably start with
something less complicated like indexing the summary statements.

I'm open to other ideas, though. Just concerned about what fulltext
indexing of all logs might entail.

>  3. Are there plans/ideas on implementing a waiver mechanism? 
Yeah, it'll have to happen before we start gating any
rawhide/branched/stable builds or updates for Fedora. We don't have a
design yet but it'll happen at some point.


>  [1] - https://git.fedorahosted.org/git/rpmgrill.git
>  
>  PS: Yes - I'm aware that there is already a bitbucket repository of
>  rpmgrill for Taskotron, but have only had a quick glimpse at it. 
Someone was planning to work on getting rpmgrill working in taskotron
and that bitbucket repo was meant to hold the task that's executed in
taskotron.

From our end, that is being tracked as:

https://phab.qadevel.cloud.fedoraproject.org/T281

Let us know if you have any other questions or concerns.

Tim
>  Thanks and Kind Regards,

[Róman Joost]

Dear Tim,

On Thu, Feb 12, 2015 at 10:31:23AM +0100, Tim Flink wrote:
>  > [...]
>  > I assume, running rpmgrill in Taskotron could mean, that each rpmgrill
>  > test result translates to an entry in Resultsdb as an outcome (PASSED,
>  > FAILED). Yet the question for the developer as to why it failed is
>  > important.
>  
>  I don't think that desire is unique to rpmgrill and something that we
>  decided to leave out of the initial Taskotron system.
>  
>  My initial thought is that that a discrete state (PASS, FAIL,
>  NEEDS_INSPECTION etc.) a one-line-or-so summary and link to output
>  from rpmgrill would work well. Are there other things that you'd like
>  to see in terms of output? 
From what Kamil pointed out, there is no state missing.

>  > 1. Ed Santiago is currently running RPMGrill here:
>  > 
>  >     http://rpmgrill-fc20.edsantiago.com:5000/recent
>  > 
>  >   which in terms of presenting results to the user provides a
>  > different experience. The results are grouped by test and provide
>  > more insight as to why a test failed. Test results in Resultsdb
>  > currently seem to have only one outcome and technical details are
>  > left on the build master.
>  
>  This was less of a desired final outcome for us than it was the easiest
>  way to get started.
>  
>  I think that our current method of having gigantic text-only log files
>  is far from optimal (and something that we did years ago in AutoQA
>  before making the logs more easily accessible) but we haven't improved
>  the log accessibility/readability in taskotron yet. 
Good to know. I guess this is something to discuss and work on?


>  >   How does a representation like RPMGrill translate into
> Resultsdb?
>  > Are there plans/ideas on how to provide a more diversified result
>  >   representation in Resultsdb (e.g. error reason, hint on how to solve
>  >   an error).
>  
>  I think that having rpmgrill output static html would likely be the
>  best approach here unless you're planning to continue supporting the
>  webapp that you linked to (and from previous conversations, I'm under
>  the impression that was never intended to be a long-term thing). 
If we'd have a way to turn the TAP output into details retrievable from
resultsdb, it'll be better than rpmgrill printing HTML?


>  > 2. Are there plans/ideas about implementing a (fulltext-) search
> over
>  >   results?
>  
>  Nothing written down, no but that's been on my wishlist since before
>  Taskotron really got started. Unless there's more demand for the
>  fulltext indexing than I think there is, we'd probably start with
>  something less complicated like indexing the summary statements.
>  
>  I'm open to other ideas, though. Just concerned about what fulltext
>  indexing of all logs might entail. 
True. We've had quite big logs if some big packages contain a lot of
regressions. Good to know that it is on the wish list.

>  > 3. Are there plans/ideas on implementing a waiver mechanism?
>  
>  Yeah, it'll have to happen before we start gating any
>  rawhide/branched/stable builds or updates for Fedora. We don't have a
>  design yet but it'll happen at some point. 
Ok. This also seems like something we might be able to contribute to.
Perhaps better to discuss this in another thread?

>  > [1] - https://git.fedorahosted.org/git/rpmgrill.git
>  > 
>  > PS: Yes - I'm aware that there is already a bitbucket repository of
>  > rpmgrill for Taskotron, but have only had a quick glimpse at it.
>  
>  Someone was planning to work on getting rpmgrill working in taskotron
>  and that bitbucket repo was meant to hold the task that's executed in
>  taskotron.
>  
>  From our end, that is being tracked as:
>  
>  https://phab.qadevel.cloud.fedoraproject.org/T281
>  
>  Let us know if you have any other questions or concerns. 
Thanks. I'll have a look at the tasks. I don't want to make any
promises, but I hope we'll find some time to get the rpmgrill tasks in a
'done' state.

Thanks and Regards,
-- 
Róman Joost
Software Engineer, HSS - Software Engineering and Development (Brisbane)
email: rjoost(a)redhat.com | tz: UTC+10
irc: #hss #rpmdiff

[Róman Joost]

Hi Tim,

just a small ping :)

On Mon, Feb 16, 2015 at 09:17:50AM +1000, Róman Joost wrote:
>  > [...]
>  > This was less of a desired final outcome for us than it was the easiest
>  > way to get started.
>  > 
>  > I think that our current method of having gigantic text-only log files
>  > is far from optimal (and something that we did years ago in AutoQA
>  > before making the logs more easily accessible) but we haven't improved
>  > the log accessibility/readability in taskotron yet.
>  
>  Good to know. I guess this is something to discuss and work on? 
>  > > 3. Are there plans/ideas on implementing a waiver
> mechanism?
>  > 
>  > Yeah, it'll have to happen before we start gating any
>  > rawhide/branched/stable builds or updates for Fedora. We don't have a
>  > design yet but it'll happen at some point.
>   
If this thread has reached a dead end, I think the best way forward is
to come up with an implementation proposal for either of the two items:
keeping more test informatin in resultsdb or a waiver mechanism.

Kind Regards,
-- 
Róman Joost
Software Engineer, HSS - Software Engineering and Development (Brisbane)
email: rjoost(a)redhat.com | tz: UTC+10
irc: #hss #rpmdiff