Any change to feed that could affect a package is supposed to trigger this, and that does include deleting a package, adding a package, or a detected change in any repodata files from any associated connectors. What happens is any operation like that sets an index dirty flag for the feed, and if a request is made for it after that, it is rebuilt. There is also a 30 second delay before this can get triggered, so if you:

• upload/delete a package
• wait 30s
• get repodata/repomd.xml

It should show an updated timestamp and revision.

We are currently moving to more of a job-based system for these index rebuilds, which also will provide better visibility into their creation and allow for manually triggering them. We're starting this process with Debian feeds since those exhibit the most problems with the current system, but plan to extend it to rpm and apk as well.

Anyway, if you're not seeing it update even after the 30s delay then there must be some other issue at work, so we'll need to dig into that.

Hi @felfert,

Apologies for the delay, I was out of the office recently. We tested this one quite a bit before release and didn't see any cases of the index not getting updated. Of course, I believe you that it is happening, but it's not something I've been able to reproduce. One thing to note is that those repodata files in storage only get updated on-demand - so something would have to actually make a request for the index to get ProGet to rebuild it (this is something that we are planning to change in a separate issue, but that is how it works currently).

Of course, we'd be happy to do some testing with your packages specifically if you'd like to upload some - just open a ticket at https://my.inedo.com/tickets/new which does allow file attachments. Ignore the header that says it's for paid users only; we'll associate it with this forum post.

-Greg

We've now fixed this and several other related RPM issues as PG-3094, scheduled for release this Friday in ProGet 2025.9. Thanks again for the bug report!

Hi @carsten_1879,

Thanks for trying it out. It is new, and has not been tested as much against older repositories. That said, we've reproduced this error and it's now fixed in the latest build - buildmaster:25.0.0-ci.28 if you would like to try it again. It's caused by older repos that don't support offset deltas being transmitted during a clone or fetch. Small repos that we used for testing ADOS so far were not large enough to have deltas at all, so those worked. So far as I can tell, the repository name containing a dot doesn't have any effect.

Thanks!
-Greg

Hi Marc,

Thanks for the note! That file is actually a sort of cross-platform way of storing symbolic links. BuildMaster also includes all the binaries needed for agents on both Windows and Linux, including .net runtimes for all three. In order to save space we deduplicate identical binaries from the agents. We use a simple json file instead of proper symlinks because we use the same file on Windows/Linux and it's easier this way. All the paths in the file have historically used Windows path separators due to how we generate it, and we convert those in BuildMaster at runtime to the proper format.... or at least we were supposed to. At some point there was a regression added that caused this to not happen when updating from Linux.

Anyway, this will be fixed as BM-3981 in the next release :)

There was in issue with Docker container initialization due to some of the updated startup code in ProGet 2025 - it's now been resolved, and if you pull the proget:25.0.1 it should work.

Hope this helps!

This looks like it was caused by us packaging an old version of Inedo Hub as part of the offline installer. We'll have it fixed for the 2025.1 release which will go out this week.

Hi @layfield_8963,

We've added osx-arm64 as another target, and we'll start deploying that along with the other binaries. We'll probably push out a new release today or tomorrow.

Thanks!

Hi @layfield_8963,

Looks like this was just a simple bug preventing output from being written when stdout has been redirected. It's now fixed in pgutil 2.1.1.

Thanks for the update!
-Greg

Hi Fritz,

We appreciate the feedback on this! ProGet 2025 will indeed have a separate dedicated mount for the database, but we didn't include this in the preview. As to the other issues, yes we do have a supervisor that attempts to start postgres and perform a clean shutdown on container stop... but there are some bugs we are working through, likely made worse by the ownership issues you discovered.

These concerns will absolutely be addressed before launch.

Again, thanks for trying it out and please let us know of any other issues you find or suggestions you may have!

-Greg

Hi @dimas,

I've looked into this and it appears that restricted and multiverse in noble-backports do not actually have any packages in their indexes. It's a quirk of how these connectors are implemented in ProGet that components with no packages indexed are omitted from the output index. Is this causing a problem for you? We can look into changing this behavior, but it likely won't be a trivial fix.

-Greg

Just to let you know - the PR has been merged and we'll release AWS 3.1.2 on Friday with the fix included. Thank you!

pgutil 2.1.4 is now released which has this capability. Note that in bash you'll need to quote the argument to prevent the shell from doing the globbing.

Hi @jblaine_9526,

Good idea, we'll add this!

That's a display issue in pgutil. It's not supposed to show the build number as part of the version at all, so we'll fix that. Note that we do include the build as part of the tag, but the actual release title on GitHub is just 2.1.1, 2.0.6, etc. Thanks for reporting this!

Looks like I had misunderstood the exact problem. One related issue was already fixed in 2024.25, but the aggregation issue with one connector having a package and another not having it was not. I have confirmed that both will be fixed in today's release of 2024.26, along with that Invalid Feed Type error.

Sorry for the confusion!

Hi Scott,

This fix will be included in tomorrow's release (2024.4).

Thanks!

Unfortunately it turns out that this is more complicated than we expected. You are correct about it being normalization-related. When packages are added to ProGet, essential identifying metadata like name and version get indexed in the database, and versions are supposed to be normalized when used this way. However, it looks like this normalization is not occurring for Conda packages.

We don't want to just do the obvious fix and start normalizing them, as that could lead to duplicate packages getting added and issues with replication. So, what we propose to do is extend our feed re-index operation to detect nonnormalized versions and fix, and then start normalizing versions consistently.

Long story short - it's not going in tomorrow's release, but we have a plan to get it fixed.

Hi @kc_2466,

I can't reproduce this in ProGet 2024.25. Is it possible you are using a prerelease version? It should report ProGet 2024.25 Build 11 in the page footer.

I'm confirming because this was fixed across all feeds as a relatively late commit before the release.

Thanks!
-Greg

We've added both of these properties to the index - you should see them in this week's release of 2024.26 on Friday. Thanks for the feature request!