Delay when creating/deleting files on NFSv3 share

[jfchevrette]

We are trying to figure out a problem we're having with our new Active/Active cluster running ONTAP 7

We have a web server cluster in front of the storage systems. When ClientA create or delete a file on the NFSv3 share, other clients do not see the change right away. It takes about 20 seconds before they can see the new files or that files have been deleted.

It is my understanding that this delay might be caused by the NVRAM which is responsible of logging the writes and flushing them to disk after a few seconds. However, I thought that writes were also put into read cache so that clients would be able to see the changes right away and not only when the NVRAM flushes to disk.

When I update files, the changes are seen immediately by other nodes. The behaviour I'm seeing is only for newly created files or deleted files.

We are worried that this will cause issues when we go in production.

Am I missing something here?

Thanks in advance

Jean-Francois

[adamfox]

I've seen this before and it's not NVRAM on the storage that's the issue, it's metadata caching on the client.

All NFS clients do this to enhance performance by avoiding NFS ops.  If you need the changes to be seen immediately, you can disable it on the client side, but you may increase the load.

This is typicall done via mount options.  The option depends on the client OS.  One many it's actime, but you'll need to look at it.

When something is in NVRAM, it's also in read cache so we will serve the correct data if asked.  In this case, the op is never being requested from the storage because the client is caching it.