Dmitrii Kuvaiskii
Dmitrii Kuvaiskii
@boryspoplawski This PR is ready for review.
Looks like reasonable optimizations.
@yao-ji Is this something that you want to work on?
Thanks for the great summary of the issue! One thing is missing though, before we find the root cause: do you supply files from `/test-fs/` as inputs (`argv[1]`) to your...
Ok, I looked into this. This is a complicated problem. Basically, Gramine has the `inode` concept for each file (regardless of whether the file is opened or not). And this...
> So these metadata of the inode structure are kept in some kind of cache (dcache)? Yes. > ...and leaving it to the application to take care of the semantics?...
> I guess what we could do is retrieve data from the host more often, effectively "refreshing" the inode. That would mean higher overhead but maybe not much higher. @pwmarcz...
> our case scenario that processes (other enclaves) do not get the updated size when just trying to read the inode structure. I think that's the same scenario... If we...
A similar issue happens in Gunicorn, see https://github.com/gramineproject/examples/pull/80. Basically, Gunicorn has this logic: - the worker process is supposed to update the ctime of the shared file, - the parent...
Jenkins, test this please