On Sun, Mar 3, 2019 at 6:31 PM Thomas Munro wrote:
> A more obvious approach that probably moves us closer to the way
> kernel developers expect us to write programs is to call fsync()
> before close() (due to vfd pressure) if you've written.
Interesting
> The obvious
> problem with that is
Hello,
We corrected our previously held belief that it's safe to retry
fsync(). We still haven't dealt with the possibility that some
kernels can forget about write-back errors due to inode cache
pressure, if there is a time when no process has the file open. To
prevent that we need to hold dirt
