Thursday, January 9, 2014

Block Change Tracking Not Used Though Enabled

During normal checks of the backups, we observed the backups are taking longer than usual. These are QA databases which are not backed up thru schedules. We use Ad hoc backups for these hosts to conserve space. Also we usually take monthly full and rest of the days L1 incremental backups.  Also the change i.e. block change tracking was implemented recently on these hosts. We double checked that earlier runs were finishing with BCT in around couple of hours which has now gone back to previous run times of 7-8 hours. Strange that why this happened??!!

Following confirms what we were observing.
 BCT is enabled and alive.
Then we checked whether datafile backup is using BCT or not.

select completion_time, datafile_blocks, blocks_read, blocks, used_change_tracking
from v$backup_datafile where file#=8 order by 1;

 




We also checked the OS process, responsible for writing changed blocks to BCT file and found it running on both nodes of cluster .

[oracle@  rman]$ ps -ef | grep ctwr

oracle    6284     1  0 Jul26 ?        00:00:00 ora_ctwr_xxxxx1

Alert log also shows that process is active as well–
CTWR started with pid=65, OS id=2653

Block change tracking service is active.

So, we have everything looking healthy but why the BCT is still not used. We researched a bit on MOS and we got the reason why it was not used.

Cause –
If L1 backups are more than 8 then you need to increase the parameter _bct_bitmaps_per_file (init.or or spfile setting) from its default of 8 to the number of incremental that one take between full backups.   Also, if you have taken more than 8 incremental backups since the last full back up, you need to take another full back up before the change will kick in, because the historical bitmaps have already been deleted.

When I checked the number of L1 taken since last full backup they were in excess of 8!!  So we decided to take full backup and after that when we checked incremental, the issue was resolved.

Lesson learned –
If there are 8 or more than 8 L1 backups since last L0, the BCT will not be used for backups.

No comments:

Post a Comment