You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As it currently stands, if you provide a bind mount in the mounts vector with a mount_point of /, things don't go as expected. The problem is that the child process is already sitting on /, so it doesn't see the bind mount. There doesn't seem to be a good way to get it to re-trace the lookup path, since we've already pivot_rooted.
To solve this, we should modify the JobSpec. I'm thinking we can have two types of root. The first is "layers", and the second is "bind". You shouldn't be able to specify both "layers" and "bind" for the root.
We're going to have to do some work to allow the (local) worker to run without a fuse layer fs.
In the meantime, users can piece together bind mounts that cover most of / by individually bind mounting in all of the relevant directories.
The text was updated successfully, but these errors were encountered:
As it currently stands, if you provide a bind mount in the mounts vector with a mount_point of /, things don't go as expected. The problem is that the child process is already sitting on /, so it doesn't see the bind mount. There doesn't seem to be a good way to get it to re-trace the lookup path, since we've already pivot_rooted.
To solve this, we should modify the
JobSpec
. I'm thinking we can have two types of root. The first is "layers", and the second is "bind". You shouldn't be able to specify both "layers" and "bind" for the root.We're going to have to do some work to allow the (local) worker to run without a fuse layer fs.
In the meantime, users can piece together bind mounts that cover most of / by individually bind mounting in all of the relevant directories.
The text was updated successfully, but these errors were encountered: