#Hm i might have discovered something

1 messages ยท Page 1 of 1 (latest)

uncut escarp
#

got a link to the test/repro? those sounds like logs that should be making it out through the API server. ๐Ÿค” the only things that were noticeably removed were meta client.Connect related things. The run I tested also started a service from within a Zenith module, and they didn't seem to disappear, but I guess my method for testing was only checking for vertex names present in the logs, so maybe the vertex showed up but had no logs, but ... it should, since those come from the API server!

true wing
#

uhh not in ci, but if you run cd magicache; go run ./cmd/ci test -- -run "TestLayerCache" against a dev engine for dagger.io, you should see a complete absence of logs

#

you're looking for lines like {"level":"info","tenant":"test-tenant","time":"2024-01-31T15:37:24Z","caller":"/src/service.go:442","message":"took 8.324ยตs to acquire lock"} - but you won't see them from the main engine, only from v0.9.7

uncut escarp
#

thanks, taking a look!

true wing
#

๐ŸŽ‰

uncut escarp
#

oooo maybe this is a "legacy" Dagger vs Zenith thing ๐Ÿ˜ฑ

uncut escarp
#

Yeah seems like that's the case... ugh, this puts us in a tight spot. I think we'll need to either try to funnel "legacy" Dagger code down the same path somehow. For now we'll likely need to revert this PR.

cc @graceful urchin I thought I remembered an issue around "legacy Dagger <=> Zenith" but can't seem to find it, just trying to gauge whether there's a short path to victory here before tomorrow, vs. just reverting

#

there's probably a middle ground where I restore the plumbing, but just don't bother propagating the stuff added by that PR (progrock parent vertex) which introduced the cache busting issue. will be sad to see the plumbing come back, but we can probably remove it again later.

true wing
#

i think that sounds good to me (though i'm not 100% sure about the details)
fingers crossed this doesn't make the hellish timeout issue come back (from the buildkit upgrade)

uncut escarp
#

Yeah D: ๐Ÿ™

true wing
#

i don't think it will - i think i've convinced myself that the tests are now "running slower" for some reason - based on there's not a single test that just gets blocked forever, just lots of them getting stuck on simple operations for a couple minutes.

#

(this is still on my list to investigate fyi)

uncut escarp
true wing
uncut escarp
#

yep yep

true wing
#

โค๏ธ โค๏ธ awesome ๐ŸŽ‰