cloudstack-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Will Stevens <williamstev...@gmail.com>
Subject Re: [PROPOSAL] Minimum Viable CI Integration
Date Tue, 08 Mar 2016 00:47:23 GMT
Hmm. It probably is the right place, but I think a straight up "NO" is more
likely if we go that route.

It's hard to know if I should just implement 3) before we even ask in order
to improve our likelihood of getting something we can work with.

I will think about this and see what I can come up with as a simple yet
complete solution that will enable us to test against distributed CI
environments, while still fitting into a model that the Apache infra guys
will approve.
On Mar 7, 2016 6:36 PM, "Erik Weber" <terbolous@gmail.com> wrote:

> I guess the appropriate channel would be to create a jira ticket for INFRA:
> https://issues.apache.org/jira/browse/INFRA
>
> --
> Erik
>
> On Mon, Mar 7, 2016 at 11:18 PM, Will Stevens <wstevens@cloudops.com>
> wrote:
>
> > This is kind of what I was expecting.  Do you know who I would be
> > contacting?  The permissions required are VERY minimal AND they have
> > already given the 'TravisCI' application the same permissions as we need
> > for this.
> >
> > How did we get the TravisCI application enabled and the permissions
> > accepted for that integration?
> >
> > I have been trying to thinking of ways to potentially work the system
> from
> > that side.  The main issue I have with this approach is that it is not a
> > single application that we want to give permission to.  Ideally, we would
> > give each individual/organization who is contributing a CI environment
> > their own token.  In that case, we would have to register the CI of each
> > party as their own CI integration.
> >
> > Here are some ideas I have as a 'work around' to the problem:
> >
> > 1) Register a single upr CI application with Github and have the apache
> > guys enable the integration.  This will give the application a single
> > access token.  I can then compile upr with the access token embedded into
> > the binary.  I don't like this approach and I feel we would probably be
> > violating some ToS somewhere.
> >
> > 2) Provide a web server implementation that each CI party can use to
> > register their own CI endpoint as a Github application integration.  Then
> > we have the apache guys enable each of them (which will be a harder sell
> to
> > them), but then each CI party will get their own token and will be able
> to
> > post back as themselves.  This is also nice because if someone is not a
> > good community member and misbehaves, their integration can be revoked
> > without it affecting everyone else who has a CI integration.
> >
> > 3) Provide a single web server implementation that is registered as a
> > Github Application Integration.  This implementation is then approved by
> > the apache guys for the cloudstack repo.  This web server implementation
> > (let's call it upr_server) keeps our one and only access token.  I then
> > modify the upr command line tool to take a token that is provided by the
> > upr_server when a CI party registers on the upr_server website.  The
> > upr command
> > will actually target the upr_server box when posting statuses, etc, which
> > will essentially proxy the calls on to Github using the token that was
> > approved for the integration.
> >
> > I think that 3) is probably the cleanest solution and would reduce our
> > chances of getting banned by someone for being too cheeky.  It is a whole
> > lot of trouble for nothing, but if they are going to be a stick in the
> mud
> > about this and not allow access tokens to anything other than official
> > github supported integrations, then I will have to make that work...
> >
> > Ideas?  Thoughts?  Rants?  :P
> >
> > *Will STEVENS*
> > Lead Developer
> >
> > *CloudOps* *| *Cloud Solutions Experts
> > 420 rue Guy *|* Montreal *|* Quebec *|* H3J 1S6
> > w cloudops.com *|* tw @CloudOps_
> >
> > On Mon, Mar 7, 2016 at 4:12 PM, Remi Bergsma <
> RBergsma@schubergphilis.com>
> > wrote:
> >
> > > Hi Will,
> > >
> > > This is the main problem: there’s no one except Apache Infra with
> access
> > > to the Github CloudStack repo. Even committers have to push to Apache
> > git,
> > > which is mirrored to Github. We can’t close a PR, set a label, change a
> > > title or whatever basic operation. You can ask them for a token. When I
> > (as
> > > the release manager) asked for any more permission than an anonymous
> user
> > > has it was kindly refused. I really hope you’ll have more luck but I
> > > wouldn’t count on it.
> > >
> > >
> > >
> > > Regards,
> > > Remi
> > >
> > >
> > > On 07/03/16 19:10, "williamstevens@gmail.com on behalf of Will
> Stevens"
> > <
> > > williamstevens@gmail.com on behalf of wstevens@cloudops.com> wrote:
> > >
> > > >The main thing we have to sort out with this type of integration (as
> it
> > is
> > > >today) is the distribution of access tokens with the correct
> permissions
> > > on
> > > >the apache/cloudstack github repo.  The required permissions are very
> > > >limited, but I don't know if we have access to create new tokens.  If
> we
> > > >don't then I will have to develop an application integration
> workaround
> > to
> > > >make it easier for the people with access to the apache/cloudstack
> repo
> > to
> > > >give the people running CI integrations access to update statuses
> (like
> > > the
> > > >current travis integration).
> > >
> > >
> > >
> > >
> > >
> > > >If you have questions or feedback, please don't be shy.
> > > >
> > > >*Will STEVENS*
> > > >Lead Developer
> > > >
> > > >*CloudOps* *| *Cloud Solutions Experts
> > > >420 rue Guy *|* Montreal *|* Quebec *|* H3J 1S6
> > > >w cloudops.com *|* tw @CloudOps_
> > > >
> > > >On Mon, Mar 7, 2016 at 12:36 PM, Bharat Kumar <
> > > bharat.kumar@accelerite.com>
> > > >wrote:
> > > >
> > > >> Hi guys,
> > > >>
> > > >> I am also working on the similar reporting problem, here is what i
> did
> > > >>
> > > >> link to the report https://github.com/bvbharatk/cloud-stack/pull/1
> > > >>
> > > >> I am thinking this is good enough for now, I want to start posting
> the
> > > >> results on each pr as shown in the above link.
> > > >> please give me your comments or suggestions.
> > > >>
> > > >> Thanks,
> > > >> Bharat
> > > >> On 05-Mar-2016, at 7:02 PM, Will Stevens <wstevens@cloudops.com
> > <mailto:
> > > >> wstevens@cloudops.com>> wrote:
> > > >>
> > > >> Daan
> > > >>
> > > >> Regarding the obligatory provider id.  I agree, but I am still
> trying
> > to
> > > >> figure out the details.  Creating distinct runs that have their own
> > > status
> > > >> is done by setting the 'context'.  I think we would need to have two
> > > pieces
> > > >> to this.  A provider id and an environment id.
> > > >>
> > > >> So for example.  Lets assume that my provider id is 'CloudOps' and
I
> > > have
> > > >> two different environments, one for 'KVM' and one for 'Xen' (for
> > > example).
> > > >> I would then the tool would produce the following two independent
CI
> > > >> statuses.
> > > >> 'CloudOps - KVM' : with a basic description of the environment.
> > > >> 'CloudOps - Xen' : again with a basic description of the env.
> > > >> ... and so on ...
> > > >>
> > > >> I am still sorting out the details here as well as making it easy
> for
> > > us to
> > > >> integrate this into the apache/cloudstack repo with the access we
> > > currently
> > > >> have.  Adding this is 'no biggy' for me because I am building this
> > tool
> > > as
> > > >> we speak, and trying to tailor it to our (ACS) needs, so this type
> of
> > > >> feedback is perfect as it allows me to adapt the tool before I get
> too
> > > >> deep.
> > > >>
> > > >> *Will STEVENS*
> > > >> Lead Developer
> > > >>
> > > >> *CloudOps* *| *Cloud Solutions Experts
> > > >> 420 rue Guy *|* Montreal *|* Quebec *|* H3J 1S6
> > > >> w cloudops.com<http://cloudops.com> *|* tw @CloudOps_
> > > >>
> > > >> On Sat, Mar 5, 2016 at 4:17 AM, Daan Hoogland <
> > daan.hoogland@gmail.com
> > > >> <mailto:daan.hoogland@gmail.com>>
> > > >> wrote:
> > > >>
> > > >> Will
> > > >>
> > > >> Gret work, especially the thing you are showing in link [4], I would
> > > like
> > > >> to make an enhancement request and that is a obligatory provider id.
> > > Only
> > > >> if it is no biggy for you!
> > > >> Several people may decide to do a XVM on ChildrensOS for instance
> and
> > > so we
> > > >> may be aware of an obscurity that is different. If one fails and the
> > > other
> > > >> succeeds it is easily identified.
> > > >>
> > > >> Ilya,
> > > >>
> > > >> I have been playing with go and it is a very nice language for such
> a
> > > >> simple script, though it wasn't exactly designed for it. So don't
> read
> > > my
> > > >> comment/question as an objection. But we do have
> > > >> bash,python,c#,java,javascript,xslt,sql at least. That is not
> counting
> > > the
> > > >> build system and I am sure the hyperv has some extra windows
> specific
> > > >> stuff.
> > > >> To me it is inherent to the nature of across platform orchestration
> > and
> > > >> provisioning system so it is fine. It is something to consider. On
> the
> > > >> other hand when bringing in new tools we don't make the choice
> so....
> > I
> > > am
> > > >> ranting, I guess.
> > > >>
> > > >>
> > > >>
> > > >> On Sat, Mar 5, 2016 at 7:38 AM, ilya <ilya.mailing.lists@gmail.com
> > > <mailto:
> > > >> ilya.mailing.lists@gmail.com>> wrote:
> > > >>
> > > >> I see where Daan is coming from :)  I thought this would be 4th, not
> > > >> exactly 7ths.
> > > >>
> > > >> I'm not against golang by any means (if anything - its my next "go"
> to
> > > >> language these days).
> > > >>
> > > >> Things to consider:
> > > >>
> > > >> Would notify-pr support proxy? I've been thinking on ways of
> > > >> contributing test runs, there would have to be few things i'd need
> to
> > > do.
> > > >>
> > > >> 1) massage the log content - such that no environment details are
> > > >> exposed, i can probably handle this with sed/awk..
> > > >>
> > > >> 2) i'm behind multiple firewalls with no internet access. however,
> > some
> > > >> lab environments might have a proxy, so it would be nice to have a
> > > >> support for it.
> > > >>
> > > >> Thanks
> > > >> ilya
> > > >>
> > > >>
> > > >>
> > > >> On 3/4/16 6:56 AM, Will Stevens wrote:
> > > >> Yes, I have most of it already built and will be releasing it later
> > > >> today
> > > >> or over the weekend.  The reason I chose Golang is because it can
be
> > > >> cross
> > > >> compiled to be run on any system and distributed as a single binary
> > > >> with
> > > >> no
> > > >> dependencies.  This means that no one will have to worry about
> > building
> > > >> it
> > > >> or having to change their environment at all in order to use it. 
I
> am
> > > >> trying to lower the barrier to entry and make it as easy as possible
> > > >> for
> > > >> people to contribute back their CI execution details.
> > > >>
> > > >> *Will STEVENS*
> > > >> Lead Developer
> > > >>
> > > >> *CloudOps* *| *Cloud Solutions Experts
> > > >> 420 rue Guy *|* Montreal *|* Quebec *|* H3J 1S6
> > > >> w cloudops.com<http://cloudops.com> *|* tw @CloudOps_
> > > >>
> > > >> On Fri, Mar 4, 2016 at 8:53 AM, Daan Hoogland <
> > daan.hoogland@gmail.com
> > > >> <mailto:daan.hoogland@gmail.com>
> > > >>
> > > >> wrote:
> > > >>
> > > >> Will, Do you have an implementation of notify-pr? I am asking as you
> > > >> specify it will be implemented in golang which seems odd. It is not
> > > >> amongst
> > > >> the 7 or so languages already in use.
> > > >>
> > > >> On Fri, Mar 4, 2016 at 1:54 AM, Will Stevens <
> > > >> williamstevens@gmail.com<mailto:williamstevens@gmail.com>>
> > > >> wrote:
> > > >>
> > > >> Hey Everyone,
> > > >> As I am sure most of you are aware, I have been focusing a lot on
> > > >> ways
> > > >> to
> > > >> get CI integrated back into the community.
> > > >>
> > > >> Today I build a little POC to validate some ideas and get a feel for
> > > >> a
> > > >> potential approach for getting CI integrated into the Github pull
> > > >> request
> > > >> workflow.
> > > >>
> > > >> There are multiple individuals/companies focusing on CI right now
> > > >> (which
> > > >> is a good thing), but there has not really been much discussion
> > > >> (that I
> > > >> am
> > > >> aware of) for how these different CI runs will push back results to
> > > >> the
> > > >> community.  I want to make sure that nobody's work on this topic
> goes
> > > >> to
> > > >> waste, so my goal is to provide a simple and consistent way for
> > > >> everyone
> > > >> to
> > > >> push their results back to the community.
> > > >>
> > > >> Here is the basic idea (please give feedback):
> > > >> - A simple cross platform command line tool with zero dependencies
> > > >> will
> > > >> be
> > > >> provided (and open sourced) which will handle the submission of the
> > > >> CI
> > > >> results back to the community.  It is written in Golang and is
> > > >> currently
> > > >> called 'notify_pr'.
> > > >> - At the end of a CI execution, the CI should automate the execution
> > > >> of
> > > >> this tool to handle updating the appropriate PR with the results.
> > > >>
> > > >> Configuration can be done via the command line or through an INI
> > > >> file.
> > > >> Here is an example of the configuration details.  The commit is the
> > > >> commit that the CI just executed against.
> > > >>
> > > >> token  = <your github token>
> > > >> owner  = apache
> > > >> repo   = cloudstack
> > > >> commit = c8443496d3fad78a4b1a48a0992ce545bde299e8
> > > >>
> > > >> summary_file = <a text file summary of the run>
> > > >> full_detail_dir = <a directory structure to be recursively uploaded
> > > >> to
> > > >> object store>
> > > >> full_detail_files = <a comma separated list of files to upload
to
> > > >> object
> > > >> store>
> > > >> store_api = <swift or s3>
> > > >> store_endpoint = <url endpoint>
> > > >> store_identity = <keystone identity or aws access key>
> > > >> store_secret = <keystone password or aws secret key>
> > > >>
> > > >> I have not yet implemented the object storage endpoints, but I have
> > > >> code
> > > >> to do it from a different project, so I just need to add it.  I will
> > > >> be
> > > >> able to host my CI output in a swift object store, but others may
> > > >> need
> > > >> to
> > > >> use AWS S3.  Maybe we can get sponsorship for this storage.  We will
> > > >> only
> > > >> keep the logs for a window of like a week or so on the object store
> > > >> so
> > > >> the
> > > >> storage usage will not be ever growing.
> > > >>
> > > >> Basically, the tool takes the details of the repository you are
> > > >> validating
> > > >> a Pull Request for and the commit you are building.  It also takes
> > > >> the
> > > >> files you would like to push to the pull request.  The summary file
> > > >> will
> > > >> be
> > > >> shown as text in the pull request comment and the other files will
> be
> > > >> uploaded to an object store and will be publically available for a
> > > >> period
> > > >> of time (probably about a week and then get cleaned up, details
> TBD).
> > > >> The
> > > >> files to be uploaded to object store could be either specified as
> > > >> individual files, OR a target directory and all the files in that
> > > >> directory
> > > >> will be recursively uploaded to the object store.
> > > >>
> > > >> When the tool is run, it will scan through all the open pull
> requests
> > > >> in
> > > >> the target repository and when it finds the pull request
> > > >> corresponding
> > > >> to
> > > >> the commit in question, it will post the details as a comment to
> that
> > > >> pull
> > > >> request.  This functionality is currently working (see the attached
> > > >> screenshot).  I can change the formatting and such, this is just an
> > > >> example.
> > > >>
> > > >> This is still a very rough concept that I have only worked on for
a
> > > >> day,
> > > >> but hopefully you guys agree that it is a good start towards a
> > > >> consistent
> > > >> feedback mechanism for the community to take advantage of the
> > > >> different
> > > >> distributed CI installations.
> > > >>
> > > >> Please voice your feedback and concerns.  I am sure I have not
> > > >> thought
> > > >> of
> > > >> everything and we may still want to make fundamental changes to the
> > > >> approach, but I think the concept is solid.
> > > >>
> > > >> Cheers,
> > > >>
> > > >> Will
> > > >>
> > > >>
> > > >>
> > > >>
> > > >> --
> > > >> Daan
> > > >>
> > > >>
> > > >>
> > > >>
> > > >>
> > > >>
> > > >> --
> > > >> Daan
> > > >>
> > > >>
> > > >>
> > > >>
> > > >>
> > > >> DISCLAIMER
> > > >> ==========
> > > >> This e-mail may contain privileged and confidential information
> which
> > is
> > > >> the property of Accelerite, a Persistent Systems business. It is
> > > intended
> > > >> only for the use of the individual or entity to which it is
> addressed.
> > > If
> > > >> you are not the intended recipient, you are not authorized to read,
> > > retain,
> > > >> copy, print, distribute or use this message. If you have received
> this
> > > >> communication in error, please notify the sender and delete all
> copies
> > > of
> > > >> this message. Accelerite, a Persistent Systems business does not
> > accept
> > > any
> > > >> liability for virus infected mails.
> > > >>
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message