Seems like a regression when using dagger core container --from $PRIVATE_REGISTRY
#Engine crashes when accessing private registry
1 messages · Page 1 of 1 (latest)
@primal musk we just released 0.19.1 which should resolve your issue. Sorry about the regression.
Update: @soft sonnet tells me that 0.19.1 seems to still have the regression, even though main doesn't??
I’ll check in a few when back on the computer with 0.19.1 and report back. Thanks! (That’s odd…?)
sorry, I'm re-checking this as we speak 🙏
@primal musk found what was happening. When you build the engine locally it re-uses whatever dagger cache you had before. Since it seems like I already pulled my private testing image at some point, I wasn't triggering the issue and I incorrectly assumed it was fixed on main
@hybrid ridge is looking into it now and we're looking forward to releasing another express hotfix patch tomorrow 🙏
apologies for the noise
Thanks for the speedy diagnostics!!
fix here: https://github.com/dagger/dagger/pull/11198
@primal musk fwict this may have actually existed for ~2 months. It requires that you specifically use terminal (not with-exec) and that the private image in question is not already cached in the engine via some other op (like with-exec), so might have just not been noticed until now? Either way we'll still try to get a fix released quick.
If you want a workaround in the mean time caching the image via a with-exec before running terminal should work
GitHub
Fixes #11195
The terminal implementation was not providing a session group when creating new mounts. This meant that if an input mount was derived from a still-lazy private image, no auth was avail...
Oh, probably I never hit it as many times, when we jump on a private pulled image we most likely have it already cached through many of our local executions... We just found out from some other engineer hitting the issue...
Thanks a lot Erik for the detailed explanation!