[Gluster-devel] State of the 4.0 World

[Shyam]

On 05/03/2016 11:50 AM, Jeff Darcy wrote:
>   * DHT2 had been making good progress for a while, but has been stalled
>     recently as its lead developer (Shyam) has been unavailable.
>     Hopefully we'll get him back soon, and progress will accelerate
>     again.

Ah! yes I have been AWOL upstream for some time, hope to board that 
train sometime soon.

>   * Stat/xattr caching (on the server) is a bit of a question mark.  On
>     the one hand, it should be pretty simple to implement.  On the other
>     hand, nobody has made even a minimal effort to do so.  Recent events
>     have also raised the issue of needing to do this for correctness
>     (especially around maintaining ctime across replicas) as well as
>     performance.  This would be a *great* opportunity for a currently
>     junior/novice Gluster contributor to make their mark.

If DHT2 makes sense and see's light of day (ominous enough ;) ), this 
may need a reevaluation based on the (well! *cough*) new backend. Just 
stating for those interested.

Also, saving calling syscalls is what is attempted here, but if that is 
the case, then gluster server should remember the inode in it's cache, 
and that cache should be larger than the local FS inode cache, for read 
operations not to hit disk. So at the end of this, it would be a syscall 
invocation optimization in my head, with increasing gluster brick side 
inode table sizes.

But, overall, yes current POSIX does abuse the syscalls pretty badly.

>   * Performance monitoring etc. (last item on list) has been dropped, for
>     lack of a well defined scope or requirements.

Hmmm... this is important. We *need to* revive this and define this and 
yes I would like to assist, as soon as I am out of my hiatus.