Oracle RAC RMAN备份无法写入NFS共享的问题求助

Environment Details:

• RMAN Backup Script:

  run { 
  SET nocfau; 
  ALLOCATE CHANNEL CH0 DEVICE TYPE DISK FORMAT '/nfs/datafile/%U_%T'; 
  ALLOCATE CHANNEL CH1 DEVICE TYPE DISK FORMAT '/nfs/datafile/%U_%T'; 
  ALLOCATE CHANNEL CH2 DEVICE TYPE DISK FORMAT '/nfs/datafile/%U_%T'; 
  ALLOCATE CHANNEL CH3 DEVICE TYPE DISK FORMAT '/nfs/datafile/%U_%T'; 
  BACKUP AS COPY INCREMENTAL LEVEL 0 TAG 'INCR_MERGE' DATABASE; 
  BACKUP ARCHIVELOG ALL FORMAT '/nfs/archivelogs/%U' NOT BACKED UP 1 TIMES; 
  RELEASE CHANNEL CH0; 
  RELEASE CHANNEL CH1; 
  RELEASE CHANNEL CH2; 
  RELEASE CHANNEL CH3; 
  }
  

• NFS Client Mount Options: rw,user,hard,noac,vers=4
• NFS Server Export Config: NFSv4,rw,no_root_squash,no_all_squash,secure
• Issue: Manual NFS writes work, but RMAN backup hangs with no output; 19c test throws ORA-17500: ODM err: No such file or directory

Let's break down your Oracle RAC RMAN backup issue step by step—since local backups work and basic NFS writes succeed manually, the problem is almost certainly tied to permission mapping, directory structure, or NFS tuning for RMAN's specific workload.

1. Validate UID/GID Consistency Across All Systems #

Your NFS export uses no_root_squash, but that only applies to the root user. For the oracle user, if its UID/GID doesn't match exactly between your RAC nodes (clients) and the NFS server, the server will map it to an anonymous user (usually nobody), which blocks write access even if you can copy files manually (maybe you tested as root, not oracle).

• Run id oracle on every RAC node and the NFS server. Ensure the UID (e.g., uid=54321(oracle)) and primary GID are identical everywhere.
• If there's a mismatch, adjust the oracle user's UID/GID to align (don't forget to update ownership of Oracle-related files afterward to avoid new permission issues).

2. Verify NFS Mount Paths and Directory Permissions #

Your RMAN script targets /nfs/datafile/, but the error references /home/oracle/nfs/ractest/...—double-check that these paths are actually mounted to your NFS share, and that the parent directories have the right permissions for the oracle user:

• On RAC nodes, run mount | grep nfs to confirm the NFS share is mounted at the path RMAN is using.
• Check directory permissions (run as oracle user, not root):

  su - oracle
  ls -ld /nfs/datafile/
  ls -ld /home/oracle/nfs/ractest/1178998226/full/datafile/
  touch /home/oracle/nfs/ractest/1178998226/full/datafile/test_backup_file
  

  If the touch command fails, you've confirmed a permission issue on the NFS share.

3. Address ASM Group vs Oracle User Mismatch #

You noted local data files are owned by asmadmin, but oracle isn't in that group. This shouldn't block RMAN backups to NFS—RMAN reads data files via Oracle's database processes (which have access to ASM) and writes backup files as the oracle user. The problem isn't replicating source file permissions; it's that the NFS directory doesn't allow oracle to create files.

4. Tune NFS Mount Options for RMAN Workloads #

Your noac (no attribute caching) mount option explains the frequent GETATTR requests in nfsiostat. While noac ensures consistency, it adds overhead for large RMAN backups. Adjust your options to balance reliability and performance:

• Add rsize=1048576,wsize=1048576 to use 1MB read/write blocks (optimal for large file transfers).
• Keep hard (safer for critical backups) but consider adding timeo=600 to reduce timeout-related hangs.
• Remount the share with updated options:

  umount /nfs/datafile
  mount -t nfs4 -o rw,user,hard,noac,vers=4,rsize=1048576,wsize=1048576,timeo=600 <nfs-server-ip>:/export/path /nfs/datafile
  

5. Adjust RMAN Channel Configuration #

Your 4-channel setup is reasonable, but make these tweaks to avoid common pitfalls:

• Ensure the target directory exists before running the backup—RMAN won't create nested directories automatically in some cases.
• Add MAXOPENFILES 8 to each channel to limit concurrent file opens, reducing NFS contention:

  ALLOCATE CHANNEL CH0 DEVICE TYPE DISK FORMAT '/nfs/datafile/%U_%T' MAXOPENFILES 8;
  

• Optional: Add directory validation to your script to confirm write access upfront:

  run { 
    SET nocfau; 
    ALLOCATE CHANNEL CH0 DEVICE TYPE DISK FORMAT '/nfs/datafile/%U_%T' MAXOPENFILES 8; 
    ALLOCATE CHANNEL CH1 DEVICE TYPE DISK FORMAT '/nfs/datafile/%U_%T' MAXOPENFILES 8; 
    ALLOCATE CHANNEL CH2 DEVICE TYPE DISK FORMAT '/nfs/datafile/%U_%T' MAXOPENFILES 8; 
    ALLOCATE CHANNEL CH3 DEVICE TYPE DISK FORMAT '/nfs/datafile/%U_%T' MAXOPENFILES 8; 
    -- Validate write access to backup directory
    SQL 'CREATE OR REPLACE DIRECTORY rman_backup AS ''/nfs/datafile''';
    SQL 'GRANT READ, WRITE ON DIRECTORY rman_backup TO sys';
    BACKUP AS COPY INCREMENTAL LEVEL 0 TAG 'INCR_MERGE' DATABASE; 
    BACKUP ARCHIVELOG ALL FORMAT '/nfs/archivelogs/%U' NOT BACKED UP 1 TIMES; 
    RELEASE CHANNEL CH0; 
    RELEASE CHANNEL CH1; 
    RELEASE CHANNEL CH2; 
    RELEASE CHANNEL CH3; 
  }
  

6. Check Logs for Hidden Clues #

• On the NFS server, check /var/log/messages or /var/log/syslog for entries about permission denied or UID mapping failures when the RMAN backup runs.
• On RAC nodes, check the Oracle alert log ($ORACLE_BASE/diag/rdbms/<dbname>/<instname>/trace/alert_<instname>.log) for detailed errors beyond the RMAN-03009 message.

内容的提问来源于stack exchange，提问作者AAB