RMAN backup on NFS v3 takes too much time

Problem:

I was trying to test RMAN backup on Azure blob storage with NFS v3 access. I have noticed a huge delay even when backing up a small controlfile.

The same backup on disk took 7sec, but on NFS 15min. While watching IO on NFS mountpoint using nfsiostat I saw that initial 13min there was no IO at all and actual backup was taken only at the last minute.

Another interesting thing while watching netstat output was the following:

[root@rac1 dbbackup2]# netstat -na|grep 10.0.0.
tcp 0 0 10.0.0.5:875 10.0.0.16:2048 ESTABLISHED
tcp 0 1 10.0.0.5:54997 10.0.0.16:2049 SYN_SENT

You can also turn debug on before running backup in RMAN and check the output:

RMAN> RUN
{
  ALLOCATE CHANNEL disk1 DEVICE TYPE DISK FORMAT '/dataload/%U';
  debug on;
  BACKUP current controlfile;
  debug off;
}

Solution:

Disable dNFS.

$ cd $ORACLE_HOME/rdbms/lib
$ make -f ins_rdbms.mk dnfs_off

Retry RMAN backup, in my case it took only 1sec:

...
piece handle=/dbbackup/0q0rruhd_26_1_1 tag=TAG20220425T202309 comment=NONE
channel disk1: backup set complete, elapsed time: 00:00:01
Finished backup at 25-APR-22

RMAN spread a backup job between many RAC instances in parallel to increase throughput

There are two options to allocate RMAN channels on different RAC instances to increase the throughput.

I will start with the option, that assures all RAC instances get one channel. Regarding the other option, it does load balance but in a random way, so with a small number of channels, you may see that all of them are allocated in one instance. So you decide which option is better for you.

The test is done on a 2-node cluster.

  1. Configure parallelism and two channels in RMAN. Indicate a connect string, one per instance:
$ RMAN target /
RMAN> CONFIGURE DEVICE TYPE disk PARALLELISM 2;
RMAN> CONFIGURE CHANNEL 1 DEVICE TYPE DISK CONNECT 'sys/Oracle123@ORCL1 as sysdba';
RMAN> CONFIGURE CHANNEL 2 DEVICE TYPE DISK CONNECT 'sys/Oracle123@ORCL2 as sysdba';

2. Define ORCL1 and ORCL2 aliases on each database node under $ORACLE_HOME/network/admin/tnsnames.ora:

ORCL1=
(DESCRIPTION=
           (ADDRESS= (PROTOCOL=tcp) (HOST=rac1.example.com) (PORT=1522))
           (CONNECT_DATA =
              (SERVER = DEDICATED)
              (SERVICE_NAME = orcl)
           )
      )
 
ORCL2=
(DESCRIPTION=
           (ADDRESS= (PROTOCOL=tcp) (HOST=rac2.example.com) (PORT=1522))
           (CONNECT_DATA =
              (SERVER = DEDICATED)
              (SERVICE_NAME = orcl)
           )
      )

Please note, in my case 1522 is a local listener port.

3. Run backup:

RMAN> backup database;
 
Starting backup at 22-MAR-22
using target database control file instead of recovery catalog
allocated channel: ORA_DISK_1
channel ORA_DISK_1: SID=507 instance=orcl1 device type=DISK
allocated channel: ORA_DISK_2
channel ORA_DISK_2: SID=753 instance=orcl2 device type=DISK
channel ORA_DISK_2: SID=753 instance=orcl2 device type=DISK
channel ORA_DISK_1: starting full datafile backup set
channel ORA_DISK_1: specifying datafile(s) in backup set
input datafile file number=00001 name=+DATA/ORCL/DATAFILE/system.257.1098460673
input datafile file number=00004 name=+DATA/ORCL/DATAFILE/undotbs1.259.1098460743
channel ORA_DISK_1: starting piece 1 at 22-MAR-22
channel ORA_DISK_2: starting full datafile backup set
channel ORA_DISK_2: specifying datafile(s) in backup set
input datafile file number=00003 name=+DATA/ORCL/DATAFILE/sysaux.258.1098460717
input datafile file number=00005 name=+DATA/ORCL/DATAFILE/undotbs2.265.1098461311
input datafile file number=00007 name=+DATA/ORCL/DATAFILE/users.260.1098460743

As you see, two files were backed up by the 1st channel (1st instance) and the other three files by the 2nd channel (2nd instance).

Now let’s explain another possible variant:

  1. Configure one TNS string with load balance parameter:
ORCL_BALANCE=
     (DESCRIPTION=
           (TRANSPORT_CONNECT_TIMEOUT=3) (RETRY_COUNT=6)(LOAD_BALANCE=on)
           (ADDRESS= (PROTOCOL=tcp) (HOST=rac1.example.com) (PORT=1522))
           (ADDRESS= (PROTOCOL=tcp) (HOST=rac2.example.com) (PORT=1522))
           (CONNECT_DATA =
              (SERVER = DEDICATED)
              (SERVICE_NAME = orcl)
           )
      )

Both node addresses are defined and Oracle will pick each address randomly.

2. Configure parallelism and one channel with ORCL_BALANCE string:

Please note, I did this test case on the same server where I’ve already defined CHANNEL 1 and CHANNEL 2, so I had to clear them:

RMAN> CONFIGURE CHANNEL DEVICE TYPE DISK clear;
RMAN> CONFIGURE CHANNEL 1 DEVICE TYPE DISK clear;
RMAN> CONFIGURE CHANNEL 2 DEVICE TYPE DISK clear;

Define channel and parallelism:

RMAN> CONFIGURE DEVICE TYPE disk PARALLELISM 2;
RMAN> CONFIGURE CHANNEL 1 DEVICE TYPE DISK CONNECT 'sys/Oracle123@ORCL_BALANCE as sysdba';

RMAN> backup database;
 
Starting backup at 22-MAR-22
allocated channel: ORA_DISK_1
channel ORA_DISK_1: SID=752 instance=orcl1 device type=DISK
allocated channel: ORA_DISK_2
channel ORA_DISK_2: SID=138 instance=orcl1 device type=DISK

Both channels were allocated on orcl1. Try one more time, or better configure parallelism 3, CHANNEL parameter is already defined and it is permanent until changed:

RMAN>  CONFIGURE DEVICE TYPE disk PARALLELISM 3;
 
RMAN> backup database;
 
Starting backup at 22-MAR-22
allocated channel: ORA_DISK_1
channel ORA_DISK_1: SID=752 instance=orcl1 device type=DISK
allocated channel: ORA_DISK_2
channel ORA_DISK_2: SID=138 instance=orcl1 device type=DISK
allocated channel: ORA_DISK_3
channel ORA_DISK_3: SID=753 instance=orcl2 device type=DISK

As you see it really did a random choice. But two channels were allocated on the 1st node and the last one on the 2nd node.

I think here random algorithm is not a good option, but you better know which variant is appropriate in your case.

RMAN: Displaying current backup progress

To check the progress of your current RMAN backup use this script:

#This is my favorite script

select recid
 , output_device_type
 , dbsize_mbytes
 , input_bytes/1024/1024 input_mbytes
 , output_bytes/1024/1024 output_mbytes
 , (output_bytes/input_bytes*100) compression
 , (mbytes_processed/dbsize_mbytes*100) complete
 , to_char(start_time + (sysdate-start_time)/(mbytes_processed/dbsize_mbytes),'DD-MON-YYYY HH24:MI:SS') est_complete
 from v$rman_status rs
 , (select sum(bytes)/1024/1024 dbsize_mbytes from v$datafile) 
 where status='RUNNING'
 and output_device_type is not null

HOW TO CATALOG TAPE BACKUP PIECES

You can catalog TAPE backup piece only using automatic channel:

RMAN> configure channel device type 'SBT_TAPE' parms <mml parameters>

Example:

configure channel device type 'SBT_TAPE' parms='ENV=(NB_ORA_CLIENT=ClientHostName,NB_ORA_SERV=backupServerHostName)';
RMAN> catalog device type 'SBT_TAPE' backuppiece 'arch_dEYC_ub4qtfud9_s20836_p1_t903346601';

How to catalog backup located on TAPE in RMAN?

If you are using Symantec Netbackup as a backup server, the following command will be useful for you:

1. To list available files located on tape taken by some client, run the following:

# /usr/openv/netbackup/bin/bplist -C r2n1 -t 4 -l -R /

where,

-C     client name
-t     is a policy type , where 4 means Oracle
-l     shows more details info
-R    recursively
/     root directory

–output

-rw-rw---- oracle oinstall 18939648K Aug 11 04:48 /al_dMEDIA_s3941_p1_t791009326_T20120811

2.  In my case this backuppiece is taken by r2n1 client and now I am cataloging it on orcl client. That’s why I need to pass additional parameters to channel. If these clients are the same , parameters are not necessary. (If you have some problem, like orcl can’t see r2n1’s backups see my post about alternate client)

RMAN> CONFIGURE CHANNEL DEVICE TYPE 'SBT_TAPE' PARMS 'ENV=(NB_ORA_CLIENT=r2n1,NB_ORA_SERV=JUS-BACKUP01)';

RMAN> catalog device type 'SBT_TAPE' backuppiece 'al_dMEDIA_s3941_p1_t791009326_T20120811';

3. If you want to clear your configuration do the following,or skip this step.

CONFIGURE CHANNEL DEVICE TYPE 'SBT_TAPE' CLEAR;

That’s all, good luck!