about quality of openSUSE-repos
Hello openSUSE! We've recently received a few complaints about https://github.com/openSUSE/openSUSE-repos, which provides an alternative handling of repository paths by *RIS. Here is a little summary of the issues and our actions. *Mixing NVIDIA subpackage flavors (MicroOS one installed on TW etc) boo#1224184 *Breakage of Leap 15.6 repo paths (by removal of tailing / in a variable) boo#1224185 (please be aware that 15.6 is still a pre-release and the change did not get to 15.5:Update) Regarding the Mixing of flavors I've quickly fixed the flavor mixing problem by adding conflicts https://build.opensuse.org/package/rdiff/Base:System/openSUSE-repos?linkrev=base&rev=36 and opened https://github.com/openSUSE/openSUSE-repos/issues/54 to track our QA efforts. Regarding the breakage of Leap repo paths I implemented a little CI that checks repository definitions by variable substitution and curl https://github.com/openSUSE/openSUSE-repos/actions This already found boo#1224217 (issue with source and debug repo in Leap armv7hl) and similar Slowroll issues that Bernhard already addressed. Many thanks to Andrii who is working on a CI that tests intel repo definitions with zypper on an available Ubuntu container https://github.com/andrii-suse/openSUSE-repos/actions This adds another level of testing and involves zypper for proper variable substitution. The last remaining step in #54 is openQA coverage where I'd like to ask for help. https://progress.opensuse.org/issues/160340 The fact that we build Slowroll/TW/MicroOS in a single project adds a space for human packaging error where people can potentially end up with mixed flavors. I suppose we could be partially testing it with Adris's zypper in the Ubuntu container approach, but I think it should also be a part of openQA for TW (and its flavors) as we have Suggests in release packages that will not be present on ubuntu. **Any help on the openQA part (poo#160340) would be highly welcome.** I only wish we'd done these steps towards increased quality sooner. As breaking repo paths ends up basically by cutting off systems from any further updates. This should not be tolerable in anything post-Alpha quality, even when it's not the default. [0] https://en.opensuse.org/openSUSE:Standards_Repository_Index_Service Relevant Reddit threads https://www.reddit.com/r/openSUSE/comments/1cdhjzg/comment/l1e6ppv/ and https://www.reddit.com/r/openSUSE/comments/1cpbfry/tw_opensusereposmicroosnv... -- Thank you for your understanding Luboš Kocman openSUSE Leap Release Manager
On 5/14/24 11:36, Lubos Kocman via openSUSE Factory wrote:
[0] https://en.opensuse.org/openSUSE:Standards_Repository_Index_Service
Hi Lubos, Does the Repository Index service support the "history" repos ( https://download.opensuse.org/history/ ) available for Tumbleweed and used by the tumblweed-cli package ? If no, are there any plans to add support for them ? Thanks! -- Regards, Joe
Lubos would you be interested in bringing this up in the Friday's Tools
Workshop? I didnt know about that repo and I would like to hear more about
it
On Tue, May 14, 2024 at 9:22 PM Joe Salmeri
On 5/14/24 11:36, Lubos Kocman via openSUSE Factory wrote:
[0] https://en.opensuse.org/openSUSE:Standards_Repository_Index_Service
Hi Lubos,
Does the Repository Index service support the "history" repos ( https://download.opensuse.org/history/ ) available for Tumbleweed and used by the tumblweed-cli package ?
If no, are there any plans to add support for them ?
Thanks!
-- Regards,
Joe
--
uid Ioannis Bonatakis
Hello Ionannis,
(please see also my inline comment to Joe)
I'd love to, however, I'll be on the train to Prague until 9:15am on
this Friday.
I can definitely do so next week.
On Thu, May 16, 2024 at 1:13 AM Ioannis Bonatakis via openSUSE Factory
Lubos would you be interested in bringing this up in the Friday's Tools Workshop? I didnt know about that repo and I would like to hear more about it
On Tue, May 14, 2024 at 9:22 PM Joe Salmeri
wrote: On 5/14/24 11:36, Lubos Kocman via openSUSE Factory wrote:
[0] https://en.opensuse.org/openSUSE:Standards_Repository_Index_Service
Hi Lubos,
Does the Repository Index service support the "history" repos ( https://download.opensuse.org/history/ ) available for Tumbleweed and used by the tumblweed-cli package ?
If no, are there any plans to add support for them ?
I'm no expert on RIS, that would be Michel Andres, but let me give it a try. I briefly went through the tool and it basically manipulates with repo files in /etc/zypp. Repos in /etc/zypp under RIS management get overridden by RIS on zypp ref. So the tumbleweed-cli as it is would not be a good interface for openSUSE-repos. https://github.com/boombatower/tumbleweed-cli/blob/master/tumbleweed In openSUSE case RIS basically takes local .xml template/repoindex in /usr/share/zypp/local/... processes the repoindex and creates equivalent repo definitions in /etc/zypp.d/$PREFIX:$repo.repo. SLES uses a service (SCC, RMT ...) which generates .xmls per activated products etc. Example : https://github.com/openSUSE/openSUSE-repos/blob/main/opensuse-leap-repoindex... The value of RIS managed repositories is that the maintainer (or service admin) has control about what repositories are available to the system, what are enabled by default etc. Thinking of your use case ... I can imagine that you could have a little service/plugin that would process user input and return the .xml I'm not aware how popular the /history repo is among users, but perhaps it would be a similar user base that could be potentially interested in slowroll. Feel free to track the request as an issue on github.com/openSUSE/openSUSE-repos Contributions are always welcome! code-
Thanks!
-- Regards,
Joe
-- uid Ioannis Bonatakis
Key fingerprint = F1CE 542D C18F 89E5 14B7 228E 7483 CFE0 C770 3696
-- Best regards Luboš Kocman openSUSE Leap Release Manager
Further issues that have led me to block those packages on my machine: • The debuginfo repo is named incorrectly, breaking the `dnf debuginfo-install blah --refresh` workflow. • Autorefresh is force-enabled, leading to slow network check on every zypper command. • HTTPS is force-enabled. Not all mirrors support HTTPS, leading to slower download. (This should be configurable, some users may want to hide their package choices from bystanders)
On 5/16/24 22:04, Bruno Pitrus wrote:
Further issues that have led me to block those packages on my machine:
• The debuginfo repo is named incorrectly, breaking the `dnf debuginfo-install blah --refresh` workflow. • Autorefresh is force-enabled, leading to slow network check on every zypper command. • HTTPS is force-enabled. Not all mirrors support HTTPS, leading to slower download. (This should be configurable, some users may want to hide their package choices from bystanders)
Regarding https, I once worked at a place where one particular update would freeze at 99%. It took me a week or so to figure out that the employer's IPS was doing deep packet inspection of all network traffic and found a false positive in one of the repos. It just blocked the offending stream, didn't even have the courtesy to reset it. https wasn't an option at this time, but it would have avoided the problem. They weren't doing TLS termination MITM proxying. Regards, Lew
We initially started with http://cdn.opensuse.org exclusively for exactly
your concerns. That changed over time. I do have a PR in progress to allow
a custom disturl prefix (e.g http://download-o-o)
Please report github issue for the debuginfo repo. RIS makes such repo
renames super easy to fix.
Dne pá 17. 5. 2024 7:51 uživatel Lew Wolfgang
Further issues that have led me to block those packages on my machine:
• The debuginfo repo is named incorrectly, breaking the `dnf debuginfo-install blah --refresh` workflow. • Autorefresh is force-enabled, leading to slow network check on every zypper command. • HTTPS is force-enabled. Not all mirrors support HTTPS, leading to slower download. (This should be configurable, some users may want to hide
On 5/16/24 22:04, Bruno Pitrus wrote: their package choices from bystanders)
Regarding https, I once worked at a place where one particular update would freeze at 99%. It took me a week or so to figure out that the employer's IPS was doing deep packet inspection of all network traffic and found a false positive in one of the repos. It just blocked the offending stream, didn't even have the courtesy to reset it. https wasn't an option at this time, but it would have avoided the problem. They weren't doing TLS termination MITM proxying.
Regards, Lew
participants (5)
-
Bruno Pitrus
-
Ioannis Bonatakis
-
Joe Salmeri
-
Lew Wolfgang
-
Lubos Kocman