09:27 <MrCooper> eric_engestrom: right now, enabling LTO for binaries used in CI test jobs would result in MRs being "randomly" unmergeable, because enabling LTO has a chance in ] 0, 1 [ of breaking stuff due to as yet unknown factors (the particular commit alone not being sufficient to determine which side the coin will end up on)

09:28 <MrCooper> (that's for Mesa)

09:59 <eric_engestrom> MrCooper: yeah I know, we wouldn't be able to merge an MR that enabled lto for testing until we fixed these bugs

10:00 <eric_engestrom> daniels: indeed, the runner went down again, it looks like disk issues; we'll keep it offline until we fixed that, so half capacity until further notice

10:03 <eric_engestrom> DavidHeidelberg[m]: about lto taking longer (and that being an issue if tests have to wait on that): I expect it takes longer to build with lto enabled, but i don't have an idea of how much longer, do you have before/after numbers?

12:19 <karolherbst> daniels (or any other admin): mind banning dariaamanda769 real quick? I've did a report and it's a spam bot

12:19 <karolherbst> just one comment tho :D

12:21 <daniels> done

12:23 <karolherbst> thx

12:38 <DavidHeidelberg[m]> eric_engestrom: I would say 1 - 3 minutes extra, depending on how much components are linking

13:19 <eric_engestrom> DavidHeidelberg[m]: that's not that much (although more than I expected); IMO that shouldn't be a blocker

13:25 <DavidHeidelberg[m]> container -> build (curr 2-4 min; after 3-8 min) -> test (15min). There are two risky things. First is slowdown before the test gets started. Second, increased flakiness in CI.

13:28 <DavidHeidelberg[m]> On the other hand, it has one upside I should have calculated into the equation. The tests get faster. What I saw "subjectively" saw on the LTO pipeline is like ~ 10% increase in performance (I guess because we're bound a lot with CPU)

13:33 <DavidHeidelberg[m]> btw. yesterday I found out, that another distro uses Mesa with LTO, Chimera Linux.

14:24 <MrCooper> DavidHeidelberg[m]: since we can't enable LTO in build jobs which produce binaries for test jobs, enabling LTO in build test jobs shouldn't affect the total runtime of pipelines with test jobs

15:56 <robclark> huh, marge says "CI is taking to long" after less than 6min? https://gitlab.freedesktop.org/mesa/mesa/-/pipelines/781552

15:58 <daniels> it was triggered at 2:54pm and it's still running

15:59 <eric_engestrom> robclark: waiting time counts, it's not jsut running time

16:00 <robclark> hmm, maybe marge should wait longer.. the CI jobs themselves have timeouts.. as is it is just going to result in yet another CI run and more wait time

16:01 <daniels> yeah, we could set it higher, but it's also an indication of something catastrophically wrong

16:01 <daniels> in this case, it's that NetworkManager is doing a release and sort of DoSing everything with long-running tests which don't parallelise well

16:01 <robclark> bleh

16:01 <daniels> having someone actually spend the time to fix that would be awesome, as would someone being able to sit down and justify to Equinix why we need more CI resources

16:04 <MrCooper> robclark: longer timeout for Marge means more time wasted when something goes wrong like this

16:04 <MrCooper> well, I guess not exactly like this

16:07 <robclark> I think if it were a real timeout, marge would see that the pipeline failed (or at least that is my assumption)

16:08 <MrCooper> not always, e.g. Marge also hits the timeout when the pipeline never starts in the first place

16:09 <MrCooper> in that case, Marge's timeout limits the time wasted doing literally nothing

16:15 <MrCooper> siddh: hi, feel free to ask your question about dri-devel list DMARC failures anytime

16:15 <siddh> Yes I was wanting to report that

16:16 <siddh> When I send a mail to dri-devel, I get DMARC failure messages. Like: https://pastebin.com/i9kAmhEd

16:17 <siddh> I also get bounce probe messages, so probably emails from others are also failing DMARC verification. Eg.: https://pastebin.com/UUmEhEyh

16:17 <siddh> I don't think that should happen, as the mailing list doesn't seem to attach any extra trailing text. It does not happen for LKML.

16:30 <daniels> siddh: yes it's known but there's no huge plan for resolving it, partly because it requires moving to Mailman 3 which is pretty destructive

16:34 <siddh> 👍