administrators

Hi @jw ,

Thanks for confirming that; we were able to identify the bug -- this time it ws SQL-server specific.

This is fixed via PG-3163, which we're shipping in this week's maintenance release. You'll still need to de-deduplication after however.

Tanks,
Alana

Hi @pmsensi ,

You'd need to run through a proxy like Fiddler or ProxyMan, which can capture the outbound / inbound traffic. Or if you provide us with a URL/API key we might be able to play around and attach a debugger. You could always create a free trial or something and set it up that way.

First thing that comes to mind is that your endpoint is incorrect and you may need to "play around" with the URL. Typically it's like this:

• endpoint is https://server.corp/path/my-pypi
• index file is https://server.corp/path/my-pypi/simple
• ProGet automatically appends /simple unless you specify otherwise on Advanced

The "Simple " endpoint is just a list of HTML like <a href="...">. That's what ProGet/connectors/pypi follow.

Thanks,

Steve

Hi @ayatsenko_3635, @dubrsl_1715,

Thanks for the feedback and continued discussion.

So, our Composer feed is relatively new, so we are open to exploring new ideas. One option might be to do an "import" of packages into ProGet, similar to how we handle connectors in Terraform feeds. But that's something that could also be done with a script, so we'd want to see what that looks like as a prototype.

That said, we definitely can't implement "content pointers" to Git repositories. The Git-based "package" model is not only outdated but it has several "fatal" Software Supply Chain problems.

The biggest problem, by far, is that package content is hosted by a third party (i.e. GitHub) and managed by another third party (i.e. the repository owner). At any time, a package author or the host can simply delete the repository and cause a major downstream impact - like the infamous left-pad incident in npm.

This is why other package ecosystems have adopted a "read only" package repositories and disallow deletes (except for rare cases like abuse). Once you upload a package to npmjs, nuget, rubygems, etc. -- it's permanently there, and users can always rely on that being the case.

That's simply not possible with Git-based "packages". The "content pointer" must be updated periodically updated, such as if the author decides to move the Git repo to GitLab, Gitea, etc. Now you no longer have a read-only package repository, but one that must be editable. Who edits? Are they tracked? What edits are allowed? Etc.

There are several of other issues, especially with private/organizational usage, like the fact that there's no reasonable way to QA/test packages (i.e. compared to using a pre-release/repackaging workflow) and that committing/publishing are coupled (i.e. you tag a commit to publish it). This makes governance impractical.

And that's not to mention the fact that there's no real way to "cache" or "privatize" third-party remote Git repositories. Unless, of course, you mirror them into your own Git server... which is technically challenging, especially if the author changes hosts.

We first investigated feeds years ago, but they didn't have package files at the time -- only Git repository pointers. Like Rust/Cargo (which used to be Git-based, and still technically supports Git-based packages), we anticipate this same maturity coming to Packagist/Composer as well.

So while we certainly understand that this is a workflow shift, it's a natural evolution/maturation of package management. Likewise, it took decades for the development world to shift from "file shares" and legacy source control to "Git", but that's what everyone has standardized on.

That's the world ProGet operates in, so it might be a bit "too far ahead", or we might be misreading the tea leaves - but a "package mindset" is a requirement in using ProGet. But hoepfully we can make that easier, possibly with exploring "package imports" or something like that.

Cheers,
Steve

Hi @frank-benson_4606 ,

Thanks for clarifying, that makes sense.

I'm afraid that ProGet does not "crawl" the parent artifacts for metadata; we had considered it, but it's rather challenging to do from an engineering standpoint, difficult to present crawler errors, and fairly uncommon.

Thanks,
Steve

Hi @dubrsl_1715 ,

Thanks for the clear explanation; I'm afraid your best bet here is to "take the plunge" and adopt package management best practices. A key tenet being a "self-contained archive file with a manifest" that is read-only and "cryptographically sealed" (i.e. hashed).

A "pointer to a Git commit" seems convenient at first, but there's a good reason that many ecosystems (including Composer) have moved away from it -- ultimately it's just not scalable and doesn't handle team workflows very effectively.

This will likely involve updating your composer.lock files and also the way you publish packages. In general, your process for creating a new version of a PHP packages should look something like:

1. Checkout code
2. Modify composer.json file with a version number
3. Package the file
4. Publish to ProGet

As you get more mature, you can get into pre-release versioning and repackaging.

Hope that helps,

Steve

Hi @frank-benson_4606,

I looked into this a bit closer now.

Looking at the commons-io-2.14.0.pom, there is no Licenses element specified. The pom should have that, and it'd be nice if the package authors added it; if you requested that via a pull request or issue in their github, I'm sure they would. That said, that's why it's not showing in ProGet.

This is why you see the unknown license detected, and that means you have to click "Assign License Type to Package" for ProGet to associate the package/license. I assume that you did that on 2.14.0, and selected Apache-2.0.

By default, that selection only applies to the specific version, and if you wanted it to apply to all versions of commons-io (including future ones not yet published) you'd need to click on the "Apply to all versions".

If you navigate to SCA > Licenses, and click on Apache-2.0, you can see the assignment to the package under the "Purls" tab. It would show: pkg:maven/commons-io/commons-io@2.14.0 for the version you selected.

You will need to either do this for all versions or decide if you want to add an entry to the Package Name tab (i.e. pkg:maven/commons-io/commons-io) under the Apache-2.0 license definition.

Thanks,
Steve

Hi @sneh-patel_0294 ,

What I mean is, in your browser, open multiple tabs -- one for /administration/cluster on each node in the cluster, bypassing the load balancer. All nodes should show "green" for that.

The one that shows "red" still has the wrong encryption key. Modify the encryption key, and restart the servicies, and reload the tab, and it should work fine.

Thanks,
Steve

@sneh-patel_0294 to restart the services, you can do so from the Inedo Hub or the Windows Services (look for INEDOPROGETSVC and INEDOPROGETWEBSVC). If you're still using IIS, make sure to restart the app pool as well

Hi @sneh-patel_0294 ,

This message means that the decryption keys across machines are different, which result in exactly the behavior you describe (403s, logouts):
https://docs.inedo.com/docs/installation/configuration-files

I know you mentioned you already checked, so there's likely a typo, miscopy, looking at the wrong folder, etc. Note that the folder is %PROGRAMDATA%\Inedo\SharedConfig\ProGet.config, as opposed to C:\ProgramData\Inedo\... - on some machines, the program data folder is stored in a different location.

I would also make sure to restart the service/web as well. To test, you can try loading that tpage on all nodes and you should not see "Encryption key decryption failure" when refreshing the nodes.

Hope that helps,
Steve

Hi @frank-benson_4606 ,

This appears to be a known issue that will be fixed in 2025.15, releasing this Friday, that causes certain URL-based licenses to not be detected (PG-3153).

If you're using Docker, you can try upgrading to inedo/proget:25.0.15-ci.4, which should have that fix in it.

Thanks,
Steve