We have an NFS share accessible from both nodes of the RAC database (19c, on Red Hat Linux). Is it OK to run expdp with a certain parallel degree creating multiple dumpfiles on the NFS share without corrupting the dump files? An example command is:
expdp username@dbconn parallel=4 dumpfile=nfsshare_dir:dump_%u.dmp logfile=nfsshare_dir:dump.log
(Since the cluster parameter has yes value by default, there's no need to specify that.)
If this is allowed or not allowed, is this explicitly documented? Since cluster parameter is yes instead of no by default, Oracle seems to allow this behavior? (Otherwise a default no value would make more sense.) On the other hand, we're aware of documents as "IMPDP Job Failing With ORA-39065, ORA-12805 ( Doc ID 1490265.1 )" where it says "The root cause of this [error, or possibly dumpfile corruption] is that the dumpfile was not on a shared filesystem and this was a 3 node RAC environment". The document doesn't say if they used NFS, and if yes, whether the cause of the error was simply that not all 3 nodes had mounted the NFS share, or that data pump dumpfiles on NFS are not even allowed to be written to simultaneously and only a clustered file system allows it.
Note: Our NFS is the regular Linux NFS, nothing vendor specific such as NetApp ONTAP. We know there's the guideline to set up NFS for RMAN backup so we followed that guideline, but we can't find such guideline for data pump.