[Fuego] About ftc requests

Tim Bird tbird20d at gmail.com
Sat Jun 16 10:21:02 UTC 2018


On Wed, Jun 13, 2018 at 1:04 AM, Daniel Sangorrin
<daniel.sangorrin at toshiba.co.jp> wrote:
>> -----Original Message-----
>> From: Tim.Bird at sony.com [mailto:Tim.Bird at sony.com]
>> > -----Original Message-----
>> > From: Daniel Sangorrin
>> > Hi Tim,
>> >
>> > There are several functions in ftc associated to what I think is the prototype
>> > of a fuego web app that you showed us during an ELC BoF session.
>> > - put-request
>> > - put-test
>> > - list-requests
>> > - run-request
>> > - package-run
>> > - put-run
>> >
>> > Are there any news or roadmap for this effort?
>> Unfortunately, it's been on my backburner.  The main reason it got stalled
>> was that I wanted to add security for all these transfers, but that turns it
>> into a much bigger project to get this working.
>>
>> What we have now is actually very close to sufficient, for an insecure
>> (unencrypted, unauthenticated) transfer of materials.
>>
>> There's a bug right now on the server with handling json with Unicode
>> (non-ASCII) characters, but it shouldn't be too hard to fix that.
>
> Is this code public already?
No.  I need to do that.  This is based on my tbwiki software, which has
heretofore been unpublished. But I should do that, and add it to the
Fuego project.

> I think the standard method for authentication is to provide a token in the headers. Of course, this should be passed over https when run remotely.
> KernelCI and Squad already have support for it.
Yes, but the way kernelCI is doing it is quite insecure, if I recall correctly.
I was going to use a similar approach, but using different cryptography.

>
>> The end goal is to automate some of this, but with very little effort
>> (IMHO), we do things like the following workflow manually:
>>
>> 1) create a test
>> 2) put the test on the server (ftc put-test)
>> 3) add a request to run the test on another person's board (ftc put-request)
>> we would need a list of host:boards that were available,
>> but to start with we could just know each other's host and board names.
>> 4) notify the person to run the test (e.g. by e-mail or slack).  They would do:
>> 5) ftc run-request
>> 6) ftc put-run (maybe make it an option to do this automatically for runs based on
>> remote requests)
>> 7) the original user can download the run (we don't have ftc get-run yet)
>> 8) the original user can analyze the results, looking at the logs, collateral material,
>> etc.
>
> I think we should separate visualization from request management and run them as two separated "micro" services. This way we can substitute the implementation of each micro service and we can install only the visualization service locally.
I agree with this.

>
> When working with a Visualization service (kernelci, squad, how about "vision" for fuego's?) a user would use functions such as.
> ftc put-run <-- with ability to notify by e-mail in case it was a requested run
> ftc get-run
I'm not sure I understand the 'get-run'.  do kernelci and squad allow
you to download data?  I thought they were upload-only?

>
> When working with the Request management a user would use functions such as:
> ftc put-board <-- announce that people can test on my board
> ftc rm-board <-- remove the board from the fuego network
> ftc list-boards <-- add a --remote flag to list also remote boards?
> ftc put-request <-- puts the request. I think ftc put-test is not strictly needed because we can pass the test's git repo url or standard fuego test name on the request. ftc put-test/get-test would be convenient for uploading tarballs though to reduce storage needs.
I agree that is the test is in the fuego-core repository, no put-test
should be needed.
But having put-test allows someone to make an ad-hoc or
under-development test, and use
it in the test network (put-test, then put-request), without having to
have it in the fuego-core repo.

One use case I want to support is:
 - user finds bug
 - user notifies upstream developer
 - upstream developer needs more data, or wants to try something, so
writes a test
 - upstream developer does: put-test, put-request
 - user does: run-request, put-run
 - upstream developer does: get-run
 - upstream developer uses run data to find the bug, possibly
repeating if more data is needed,or
 the test needs to be refined.

> ftc get-request <-- get next request (can be used for polling also)
> ftc run-request <-- downloads (and clears) request and executes ftc run-test

Yes.

> [Note] how about using push/pull instead of put/get?
I prefer get/put.

> [Note] A lab would run ftc get-request, ftc run-request, ftc put-run automatically using a simple polling script.
Agreed.

>> This would allow anyone in the Fuego "network" to run a test on another person's
>> board.
>
> About "network": Squad supports the concept of teams/groups and group user permissions. Having something similar would be nice.
> We also need to be flexible in regards of how we schedule the requests.
> For example, we should be able to put requests like:
> "run this on a bbb board, I don't care which one"
> "run this on a board with a cortex-a9, I don't care which one"
> "run this on a board with architecture powerpc, I don't care which one"
>
Yes.  And boards should list their attributes, and maybe include lots
of info, including their kernel config
(with put-board).  So we could also do:
"run this on a board with a synposis USB phy"
"run this on a board with a CAN bus"
"run this on a board with an nvidia GPU"

and
"run this on all boards with ethernet chipset X"
and pick up data from all the runs.

and also describe things external to the board itself, that require
special hardware setups:
"run this in a lab that has audio test (capture through external
cable) capability"
"run this in a lab that has off-board power measurement capability"


>> For example, Wang could update his busybox test, and make requests for the tests
>> to
>> run on boards in my lab, so he can test how the test works on more than just his
>> board.
>> I could test new tests on boards in multiple labs (yours, Liu's, Wang's, Khiem's,
>> Dhinakar's),
>> which would help a lot with test development and quality.
>>
>> One thing I'd like to discuss at the Fuego Jamboree is whether someone is willing to
>> be the test subject for this, and actually try it out soon after the meeting (the
>> following
>> week).
>
> I volunteer.

Great!  Let's talk.


 -- Tim Bird
Senior Staff Software Engineer, Sony Corporation
Architecture Group Chair, Core Embedded Linux Project, Linux Foundation


More information about the Fuego mailing list