On Fri, Sep 15, 2023 at 2:01 AM Daniel Gryniewicz <dang@redhat.com> wrote:

On 9/14/23 11:00, Frank Filz wrote:
>> I hope, that this is the right place for my question.
>> I want to ask if there are any caches by default. In the documentation I found
>> something concerning the MDCache. The most settings there are for the caching
>> of read access and metadata. Unless I am misunderstanding But I want to know,
>> if there is a some cache for write operation. In the end, I would like to know if
>> there is a possibility of data loss due to data cached at the Ganesha layer.
>> As a side note, we are using Ganesha together with GlusterFS
>
> Ganesha has a metadata cache but no data cache. I'm not sure if libglfs does any data caching.
>

Also, the metadata cache in Ganesha is a write-through cache, not a
write-back cache, to metadata is always consistent on the back end.

And on the data write path, the (Gluster) backend must acknowledge receiving the write before an OK response is sent to the NFS client, so there's no chance of data loss.

On the topic of data caching, I have implemented a stackable read through data caching FSAL for Ganesha. It worked as a prototype, but was never put into production. Happy to elaborate if anyone is curious.