Add coverage for improved countme system age #1501
Conversation
Make sure *no* countme flag is present.
Also adapt the HTTP matching patterns in the countme feature file accordingly. This will be used in the next commit, no functional change right now.
Shuffle some steps around, add blank lines and also emphasize that we're dealing with *calendar* weeks/months (i.e. aligned with the calendar, not with the first countme hit). No functional change.
We'll need this step in a "when" context in the next commit. No functional change.
|
I think we also need to add the machine-id code into dnf5. |
|
Yup, porting this to dnf5 was the intention, indeed. I made the patch in dnf4 just because that's what I was most familiar with. It's probably best to first let the dnf4 patch go through the usual review process and only then create a dnf5 version of it, to keep things more manageable. I can help with the latter too (or just do it myself, the related code is basically the same so it shouldn't be too hard). |
Can we make sure that the change only lands in rawhide and f41+? That way, I'll know where/when the change happened and can take that into account in interpreting the data. |
Yep, I'll sync about that with our colleague doing releases. |
|
@jan-kolarik is dnf4 shipped in f41/rawhide at all? I'm thinking if it makes sense to make this change in dnf4 and if we shouldn't just skip dnf4 entirely. |
Ehm, good point 😄 Well, basically it is not, as the default is already the dnf command from dnf5 package. So this would only affect users explicitly using |
|
OK, therefore the dnf4 patch still makes sense to pursue, I suppose. I wonder, however, if we shouldn't apply the patch conditionally (in both the libdnf4 and libdnf5 specs) only on F41+ then? Having a dnf config option to choose whether the age reported should be absolute (based on the machine-id timestamp) or relative (based on the first countme flag sent for the given repo) would be another option. It would make it easier/cleaner to control on which distros this is enabled, although it might as well be overkill, so I'm not sure. WDYT? |
|
@dmnks Sorry, I somehow missed your comment.
I think so. If we then port it to dnf5, the work should be easier, and we'll cover all dnf clients in Rawhide, as some may use dnf5 and others dnf4 in F41.
We usually try to keep the latest Fedora and upstream in sync, so I'd prefer to apply the revert patch for F40 and F39. If no one else takes it, I'd like to review your PR in the coming days. Thanks! |
No worries.
Ack!
Ack, whatever works for you best 👍
Sounds like a plan, thank you! If/when you get around to it, feel free to ping me on Slack if you need any help. |
I've created the dnf5 counterpart for this: rpm-software-management/dnf5#1525. |
|
/packit test rpm-software-management/libdnf#1662 |
|
@dmnks Oh, I just noticed that now - this needs to be targeted against |
|
Oh... Thanks. I should open a new PR against |
Probably yes, I don't know if you can switch the target branch of the existing PR. |
|
Ported: #1512 |
Adapt the countme feature to the libdnf fix for issue rpm-software-management#1611, namely: - Turn the main scenario into a scenario outline to capture the various machine-id(5) configurations (see the table). - Add an upgrade scenario (from F39 to F40) that verifies that system age is now independent of $releasever on systems with a machine-id file. - Use NO_FAKE_STAT=1 in faketime invocations so that filesystem timestamps are *not* reported relative to the target time (this would break our custom machine-id timestamps we set here), see faketime(1) for details. - Add a new MachineId class to encapsulate the machine-id file, similar to OSRelease. - Mark the touched scenarios as destructive (due to them overriding the machine-id file).
There is a difference between `makecache` command from dnf5 and old dnf4. dnf4 always fetches new metalink/mirrorlist and ignores expiration of the repo, this is likely a long-standing bug. On the other hand dnf5 doesn't contant the remote if the cache is fresh. Since the countme tests rely on the dnf4 behavior ensure it by using `--refresh` for dnf5.
According to OS-RELEASE(5) man page values with characters outside of A–Z, a–z, 0–9 such as spaces or dots must be enclosed in double/single quotes. While handling of the `os-release` file is still work in progress in dnf5 make the step more robust by wrapping all values in quotes. Old dnf4 was able to handle even non-conforming `os-release` file.
|
I have added a couple more commits to enable the tests for dnf5. |
This updates the countme feature file in accordance with rpm-software-management/libdnf#1662.