On Mon, 2023-09-11 at 09:40 -0700, Kevin Fenzi wrote: > These are the things that fedfind/qa users? Do we have examples of this > data?
Okay, here is (again) a list of the stuff fedfind uses. You can see sample data for each compose in the web UI itself, PDC actually has a rather good interactive API view. 1. WHAT: https://pdc.fedoraproject.org/rest_api/v1/composes/ HOW: ?release=fedora-NN&compose_label=label ?compose_id=cid WHY: To find a compose's compose ID from its label, or a compose's label from its compose ID. These functions are used on various paths. The most important one in real life is probably when we are creating and updating validation event pages for candidate composes, between the time the compose exists and the time it is synced to alt.fp.o - during this time wikitcms may need to look the compose up by label, but for fedfind to actually find it, cid_from_label has to work, because it can only 'natively' find a compose by label if it's been synced to alt.fp.o; if the compose has not yet been synced, the only way fedfind can find the compose is to use cid_from_label then locate it on kojipkgs under its compose ID. WORKAROUND: it would be difficult to work around this if we did not have the ability. For wikitcms we could try 'embedding' the compose ID in the event pages somehow. For some other less important uses there is probably no workaround. 2. WHAT: https://pdc.fedoraproject.org/rest_api/v1/compose-images/ HOW: compose-images/(composeid) WHY: To provide image metadata for nightly composes that have been garbage-collected, and releases in the mirror system after they have been split across directories and their metadata stripped WORKAROUND: if there is no store of metadata by compose ID then fedfind just can't do this any more. It's not critical functionality to anything important I'm aware of, though - it's just a nice feature that can be used to e.g. run long-term analysis of image size changes across multiple releases, stuff like that. If this goes away I will just drop this feature from fedfind and you will no longer be able to find the real metadata for these composes (it would provide synthesized metadata for mirrored releases, but not garbage-collected nightlies). Note, retrieving the original metadata for mirrored releases depends on the cid_from_label feature (see 1). 3. WHAT: https://pdc.fedoraproject.org/rest_api/v1/releases/ HOW: ?version=39&name=Fedora (for e.g.) WHY: To find the compose that was "previous" to any given compose. The first result for this PDC query for any given release is a dict with a handy 'compose_set' key whose value is a list of all the composes for that release, in reverse order by date. So we can easily find the compose that came 'before' any given compose. This is, I *think*, only used by the 'check-compose' script which sends out those "compose check report" emails. WORKAROUND: fedfind has a hideous alternative implementation of this which involves just blindly trying possible decrements and seeing if they exist. e.g. if you ask for the compose previous to Fedora-39- 20230911.n.1, it will try 20230911.n.0, if that doesn't exist, it will try 20230910.n.5 (we start counting backwards from 5 because, hey, gotta start somewhere, and doing more than 5 composes on one day is unlikely), then 20230910.n.4, and so on, until it hits something that exists. So, if we can't get a nice data set like this any more...we'll just have to use that. Bleh. We might also just drop check-compose and kill the feature, I suppose. I don't pay as much attention to those reports as I used to. 4. WHAT: https://pdc.fedoraproject.org/rest_api/v1/rpms/ HOW: ?compose=cid&arch=src&name=^package1$&name=^package2$.... WHY: To get the NEVR (name, epoch, version, release) for a given set of packages from a compose. This can be used by relvalconsumer (the thing that creates the wiki validation events) to decide whether any "important" packages changed between the last tested compose and the compose it's currently deciding whether to create a validation event for: if it's been more than 3 days but less than 14 days since the last event, it will create a new event *if* any important package's version changed. WORKAROUND: Looking back on the history of this, I don't think anything's actually using it right now. There is an ugly alternative version of this one, too, which scrapes dl.fedoraproject.org's HTML output (told you it was ugly!). I initially replaced that version with the PDC one for some compose types, but then ran into problems with that, and ultimately decided to provide both side-by-side; I think relvalconsumer is the only thing that actually uses this feature, and it currently uses the ugly HTML scraping version, not the PDC version. I probably meant to try and make it use the PDC version but never got around to it. The ugly version is working okay in practice, so I suppose the workaround here would just be to drop the pretty PDC version and rely on HTML scraping forever (sob). -- Adam Williamson (he/him/his) Fedora QA Fedora Chat: @adamwill:fedora.im | Mastodon: @ad...@fosstodon.org https://www.happyassassin.net _______________________________________________ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/ List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org Do not reply to spam, report it: https://pagure.io/fedora-infrastructure/new_issue
