Dne 20. 04. 20 v 13:52 Ondrej Nosek napsal(a):
>
>
> On Mon, Apr 20, 2020 at 11:42 AM Vít Ondruch <vondr...@redhat.com
> <mailto:vondr...@redhat.com>> wrote:
>
>
>     Dne 20. 04. 20 v 1:31 Ondrej Nosek napsal(a):
>>     Thanks for answers, comment in the text follows.
>>
>>     On Tue, Apr 14, 2020 at 12:16 PM clime <cl...@fedoraproject.org
>>     <mailto:cl...@fedoraproject.org>> wrote:
>>
>>         On Tue, 14 Apr 2020 at 12:05, Vít Ondruch
>>         <vondr...@redhat.com <mailto:vondr...@redhat.com>> wrote:
>>         >
>>         >
>>         > Dne 14. 04. 20 v 0:13 Ondrej Nosek napsal(a):
>>         >
>>         > TLDR: Is $SUBJ function reasonable to implement in fedpkg?
>>         >
>>         > Hi,
>>         >
>>         > some time ago, fedpkg issue tracker got a request [1] for
>>         method, that allows direct builds. That means without sending
>>         srpms via "--srpm" argument. Currently, user's changes can be
>>         built:
>>         >
>>         >     fedpkg scratch-build --srpm
>>         >
>>         > This way, fedpkg sends srpm file to koji. Without "--srpm",
>>         fedpkg sends just (git) hash id to koji. But fedpkg needs
>>         modification to send a right hash id for user changes.
>>         Additionally, koji doesn't allow building hash ids from
>>         forked repos.
>>         >
>>         >
>>         > Even if this was possible, that would also mean that the
>>         sources have to be uploaded into look-a-side cache. Then it
>>         very much depend what is the content of the PR. If I am
>>         building Ruby nightly snapshots, I don't think it is fair to
>>         pollute look-a-side cache with them and I am afraid this is
>>         not the only case. I wish we had a way to override the
>>         look-a-side cache somehow ....
>>
>>         If I understand correctly, this would have to be done, if sources
>>         changed only, right? I.e. if there is a change just in patch
>>         or a spec
>>         file, the sources could be fetched from the main project.
>>
>>
>>     Yes, just sources (and eventually other binary files) are
>>     uploaded to lookaside cache. In the case of specfile and patches,
>>     there is no lookaside modification. Fork shares the same
>>     lookaside cache with the main project.
>>      
>>
>>
>>         Should there be a possibility to upload sources for a fork
>>         that would
>>         then be moved to the main project when the pull request is
>>         merged?
>>
>>
>>     That sounds complicated to me and maybe it is not worth the
>>     intended result. Or I didn't find the right (easier) approach.
>>     ... New sources (and binaries) in fork need to be saved somewhere.
>>      - In a parallel lookaside (for forks)
>>      - In git repo (omitting lookaside)
>>     During the merge, some trigger would move the sources to the main
>>     lookaside. This creates a new file hash(es). And it would result
>>     in another commit done by a trigger.
>
>
>     Why it should create new hash(es)? If the fork/PR contains an
>     updated sources file (and it really should), then there is no
>     reason for new commit.
>
>
> Oh. In my last post, I wanted to say, that no commit is needed if you
> are using main lookaside. Just in case you have parallel lookaside or
> other storage, you have to upload sources from your alternative
> storage to main lookaside and it would result in the extra commit.
> But I looked at code how lookaside works internally. Hash is computed
> locally and the upload path (except the host) should be the same for
> both lookasides. So I realized that during merge, only download the
> sources from parallel lookaside and upload it to the main lookaside
> would work. Without extra commit. But is it a win? This might bring
> other problems. Timeouts when some large source is processed?
> Parallel lookaside is intended to not mess the main lookaside. So I
> think the parallel lookaside should be cleaned somehow.


It should be cleaned as often as the stalled PRs? I know we don't have
infrastructure for this, but it would be logical answer to your question.


Vít

_______________________________________________
devel mailing list -- devel@lists.fedoraproject.org
To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Fedora Code of Conduct: 
https://docs.fedoraproject.org/en-US/project/code-of-conduct/
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org

Reply via email to