Closed royguo closed 3 years ago
This is newly added in the experimental branch to support aio. If I/O is issued through pwrite instead of aio, it is not needed to do this.
This is done in the experimental branch to ensure we won't exceed more than one outstanding write. We could increase the number of buffers / queue depth if needed, but having async buffer flushes(one background write) improved things considerably so its probably not motivated.
@royguo : Can we close this?
https://github.com/westerndigitalcorporation/zenfs/blob/4f577cd4d4bf4b40b65ecd8b8680f37dc7b7780c/fs/zbd_zenfs.cc#L151-L160
Why do we need to Sync() here in a synchronize
Append()
method? Isn't it true that a successfulpwrite
guarantees data persistency?