07:35 <bentiss> gallo[m]: hey. Regarding MinIO, basically I need reviews for https://gitlab.freedesktop.org/freedesktop/ci-templates/-/merge_requests/158 and then tests/fixes for https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19076

07:35 <bentiss> gallo[m]: not sure what daniels was saying about the new "logging"

07:37 <daniels> bentiss: *login

07:38 <bentiss> daniels: oh, OK. But I think I sorted it out, no?

07:38 <bentiss> I mean, it seems to be working :)

07:40 <daniels> indeed you did :) his question though was looking for the exact pointers you gave so he could convert all the direct usage in Mesa CI so we can kill the old host completely

07:40 <bentiss> oh

07:42 <bentiss> gallo[m]: so there are a couple of things: 1. when the CI is relying on "ci-fairy minio", we need to remove all "ci-fairy minio login" and replace "ci-fairy minio cp" with "ci-fairy s3cp --token-file $token src dst", with one of the src or the dst being https://s3.freedesktop.org/{bucket}/{path}

07:43 <bentiss> and 2. if you are using direct upload and not ci-fairy, you should ditch the STS attribution, and rely on a single PUT with "Authorization: Bearer $TOKEN" as header

07:44 <bentiss> gallo[m]: the curl commands are given in https://gitlab.freedesktop.org/freedesktop/ci-templates/-/merge_requests/158/diffs?commit_id=7d18732a935817dc55acb6914a60989ac57b6841

07:45 <bentiss> gallo[m]: though the server only allows direct file uploads up to ~100 MB, so if you have bigger files, you need multi-part upload, as in https://gitlab.freedesktop.org/freedesktop/ci-templates/-/merge_requests/158/diffs?commit_id=2e28731d6caa16bdc20713399bdb7c52052081ce

07:46 <bentiss> gallo[m]: and no, there are no upstream documentation about this, because this is an unusual approach, that relies on upstream components, but likely not common to put them together

07:48 <bentiss> gallo[m]: last thing that comes to my mind: to be able to test this, you need a JWT token. So either you do that in effective gitlab pipelines, or you extract one from a job that will be valid for 1 hour (I do `echo $CI_JOB_JWT_V1 | base64 -w 0` to bypass the masking mechanism from gitlab that hides the token in the logs)

07:52 <bentiss> daniels: IIRC you had some docs on the MinIO approach for the projects (with how you build the proxies and what not), where was it? and do they need updating?

07:54 <bentiss> seems to be https://gitlab.freedesktop.org/freedesktop/freedesktop/-/issues/275

08:08 <daniels> bentiss: that's perfect, thankyou!

08:08 <daniels> bentiss: so that doc is out there but we didn't implement the tiering because it turned out that RO mirroring with no policy was no better than nginx ...

08:09 <daniels> the new caching is just nginx that passes on the HEAD from every req to validate access, then serves cached content if it has it

08:09 <bentiss> daniels: OK. So it should be similar with in the new requests too

08:09 <daniels> yep

08:09 <bentiss> \o/

14:07 <gallo[m]> Thank you for the detailed answer bentiss. I will take a look into it.

14:08 <bentiss> \o/ thanks!

14:13 <bentiss> gallo[m]: I have seen that you made some comments on ci-fairy. Is that all (should I start fixing your comments) or do you still have some more comments?

14:18 <jenatali> Out of curiosity, is there another gstreamer hackfest going on? Seems like all of the shared CI runners are way overloaded

14:27 <gallo[m]> bentiss I am still reviewing the multipart upload commit, but I have finished reviewing the second one https://gitlab.freedesktop.org/freedesktop/ci-templates/-/merge_requests/158/diffs?commit_id=7d18732a935817dc55acb6914a60989ac57b6841 from the MR series

14:27 <bentiss> gallo[m]: great, thanks. I'll let you finish the multipart so I do not force push while you are reviewing then

14:30 <daniels> jenatali: it's the same gstreamer hackfest, but it looks like it's just a lot of everything all at once. probably points to us being underprovisioned and needing more capacity across the board.

14:31 <daniels> (gstreamer, mesa, virglrenderer, NetworkManager, cairo)

14:38 <jenatali> 👍

14:39 <jenatali> Yeah higher capacity and greater fairness would be great

15:09 <gallo[m]> bentiss I just finished the review.

15:09 <bentiss> gallo[m]: thanks!