On 07/07/2016 12:26 PM, Christian Schwede wrote:
> On 07.07.16 12:55, Fabien Boucher wrote:
>>> Today, the story suggests to use mirror2swift[2], but I'm afraid this
>>> tool doesn't address the repodata creation, it only mirrors
>>> the rpm packages when available (and it doesn't work with elasticsearch
>>> http mirror where rpm files are not web listed).
> 
> Yes, creating repodata was never planned for this tool - it should
> simply mirror an existing public repo on Swift.
> 
Well, that's why I started this thread, we need repodata in the cache so
that yum can use it :)

>> Does it deserves an improvement of that tool then ?
>> repodata should be browsable so I think the tool could fetch them ?
> 
> That's what the tool is doing right now; but I think what Tristan meant
> is to actually take some rpms and create a missing repodata from them.
> 
>> Also could we add an option to specify directly the link to
>> the RPMs in the config file and by exception not rely on
>> the list ?
> 
> I didn't test that yet, but I think it already supports that because one
> can define multiple repos - and if one just points to a single rpm, that
> should be fine. If not, it should be fixed.
> 
It seems like we should avoid, as much as possible, micro-managing rpm
versions, so I wouldn't recommend that use case in SF.


>> That's just ideas I haven't look into the details.
>>
>>> Moreover we also discussed about using spacewalk[3] which seems a
>>> complete solution to manage our ci instances need.
>>
>> My feeling is that we should let a SF operator setup and run by
>> himself this kind of tool. I don't see s strong need to integrate
>> that in SF. Furthermore I'm concern about the difficulty to integrate
>> it tightly (in term of configuration).
> 
> Same for me. Spacewalk is great if you already have it, but if you
> explicitly need it (or integrate it in SF) it becomes another burden to
> start using SF. Every another dependency to run SF will lower the
> adoption of SF IMO.
> 
I agree, this is listed here for completeness.

>>> Thus I wonder what to do now... From my understanding we need to rely on
>>> yum utils to properly download a repository locally, and then use
>>> createrepo to produces the required repodata. I'd like to finish the
>>> original proposal and adds the missing swift publication code.
> 
> Sounds good to me; anyways looking at the existing patch it seems to be
> one step before having a local mirror?
> 
>> So mirror2swift needs to be improve to handle that ? right ?
>> Should we refine the story ?
> 
> Could we use mirror2swift to actually upload the data created by
> Tristans patch? Or just using the swift CLI?
> 
> -- Christian
> 
Based on this morning discussion, it seems like improving mirror2swift
to support repodata fetching is a good middle ground solution.
I went ahead and implemented[0] the missing bits so that we can benefit
from both efforts:

* the sf-mirror role takes care of config repo and calling mirror2swift
* mirror2swift do the heavy lifting of fetching all rpms, including
  repodata, and push them to a swift container.

If that works for you guys, I'll update my review.

Cheers,
-Tristan

[0]: https://github.com/cschwede/mirror2swift/pull/1

Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________
Softwarefactory-dev mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/softwarefactory-dev

Reply via email to