[Gluster-users] md-cache invalidation

[Andrea Fogazzi]

Hi all,

I just subscribed ot the list; we have been testing GlusterFS from some time for a typical  workload small files: more than 90%-95% of accesses are stats, of an around 1 TB and we have millions of dirs and few kb files.


I am  doing some tests with md-cache, following instructions availble on blog post

http://blog.gluster.org/2016/10/gluster-tiering-and-small-file-performance/


Our installation is 3.8.5 upgraded from 3.7.14.


I am a bit confused if we can use invalidation mechanism of "upcalls"  with with 3.8.5; when I run "gluster volume set vol performance.cache-invalidation on", I get error that option is not available; instead I was able to enable features.cache-invalidation but I wasn't able to make md cache invalidation work correctly; in fact, with the mentioned configuration, what we find is that metadata are cached until timeout expires, but even when files are updated from one of the client other clients do not receive the updated version until the md cache expires at the timeout.


Could someone clarify  if the md cache invalidation upcall mechanism is supposed to work on 3.8.x, or instead it's only available in 3.9?


Would change the behaviour if using NFS instead of gluterfs on the client side?


Thanks in advance.


Kind regards,

andrea

--

Andrea Fogazzi
foggy at fogazzi.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20161115/883eefaf/attachment.html>