[Babase] Papio Information

Ryan Hardy rh87 at duke.edu
Wed Jun 10 14:53:52 EDT 2009


I just wanted to send out a final bit of information regarding papio.   
Some of this came up from discussions Karl and I had, and some was  
related to some smaller clean up we were doing.

First, Karl had pointed out that there were two "local" backup scripts  
running that were doing the same job: dumping the database to another  
partition within the server.  After a review of the scripts, I feel  
that Karl's modified script is considerably better.  Thus, I've  
removed the older backup script from /etc/cron.daily.

Second, I've removed the temporary backup of the full database files  
from the /disk partition, as the transition from last week seems to be  
fine.  I do still have a full backup residing on the old disk (that  
was once part of the failed pair that started all this); I'll leave  
the data on that disk in place until we determine a need for the disk.

Third, I wanted to briefly go over the external backup procedures of  
papio.  Currently, the only area of the filesystem of papio that is  
backed up is "/srv/disk" and all the contents within.  This contains  
the database backup, the user and group directories, the usr_local  
directory (which houses /usr/local) and what looks like the  
DocumentRoot for the web sites.  This data is backed up fully once per  
month (in papio's case, on the third Sunday of the month).  "Diffs"  
occur every Sunday except the Sunday on which the "Full" performed.   
This is basically any data that has changed since the last "Full".   
Every other day, "Incremental" backups are done, which is any data  
that has changed since the last "Diff".  "Diff" and "Full" backups are  
sent to tape, while "Incremental" backups are sent to disk on our  
backup server.

We currently aim to keep 6 months of "Full" backups, 3 months of  
"Diff" backups, and 1 month of "Incremental" backups.  One thing worth  
mentioning is that we are only able to restore single files (by name)  
for a period of about 60 days; after that, we just have to fully  
restore the job.  I regularly monitor the execution of the jobs and re- 
run them if any problems occur.  We also periodically test both the  
disk-based and tape-based restores, though I haven't been here long  
enough to tell you how often that occurs.  I did go ahead and restore  
a file from May's "Full" backup this afternoon, just to verify.  I'm  
afraid I can't speak to the motives behind any of these decisions, as  
I wasn't around for their discussion leading to this implementation.

Please let me know if you have any questions.

-Ryan


More information about the Babase mailing list