Strange DBA script switchlog

Question

Hi guys,
I'm a system admin and one of my production system, that running oracle, is a bit strange for me and I would like your input. I'm not an oracle expert but my limited knowledge show me that it is not setup correctly. This is a 200GB Database
1. All redo log are in 1 file system /u02, only 8 log file, no grouping... Is it the right setup ?! No multiplexiing
I will, at least, do a 2 redo grouping into 2 different file system.
2. They defined 3 control files, all in the /u01 (where is the dbf file) directory (there is only 1 x 200GB oradata file system)
I will try to put the 3 control file in different location, 1 in the archive log, 1 in the each fs
3. My RPO for that database (Recovery Point Objective) is 60 minutes. So DBA use a script, in a crontab, to force a switch log every 60 minutes and they have a complex script to ensure the archivelog is close (fssync in OS) before sending it over WAN to a DR box using scp.
Now on this one... I have a huge problem and I don't understand. First, I will switch the log every 15 to 20 minutes and let oracle do it (not the cron). They put the checkpoint_timeout to 0. I will replace scp by rsync and sync the archivelog directory over wan every 5 mintues and put another cleanup script in cron to remove any archive older than 7 days.
Now, btw, I looking in the script that force Oracle to switch the log every 60 minutes and here is the command that do it... It look pretty strange to me
1. get the log sequence number : (select sequence# from v$log where status in ('ACTIVE','CURRENT');)
2. do a switch log file and archive all : (alter system switch logfile; alter system archive log all;)
3. Sleep 10 second ?! because it takes 10 second for Oracle to dump in archive log ?!
4. For each seqNo found in step 1:
4.1 select archived,sequence# from v$log where sequence# in ( ${seqNo} ); > file1
4.2 select sequence#,first_time from v$log_history where sequence# in ( ${seqNo} ); > file2
4.3 check if, for each (and there is only one BTW), is ARCHIVED = YES in v$log and present in v$log_history
4.4.Check, using fuser, if the file is written in UNIX : fuser ${archlogpath}/*${1}*${seqNo}*arc | cut -d: -f2 to be sure the file is close
4.5 scp the file over wan
Do I'm crazy or it's completely overkill ?!
Thanks

Answer 1

834010 wrote:
Hi guys,
I'm a system admin and one of my production system, that running oracle, is a bit strange for me and I would like your input. I'm not an oracle expert but my limited knowledge show me that it is not setup correctly. This is a 200GB Database
1. All redo log are in 1 file system /u02, only 8 log file, no grouping... Is it the right setup ?! No multiplexiing
I will, at least, do a 2 redo grouping into 2 different file system.
2. They defined 3 control files, all in the /u01 (where is the dbf file) directory (there is only 1 x 200GB oradata file system)
I will try to put the 3 control file in different location, 1 in the archive log, 1 in the each fs
3. My RPO for that database (Recovery Point Objective) is 60 minutes. So DBA use a script, in a crontab, to force a switch log every 60 minutes and they have a complex script to ensure the archivelog is close (fssync in OS) before sending it over WAN to a DR box using scp.
Now on this one... I have a huge problem and I don't understand. First, I will switch the log every 15 to 20 minutes and let oracle do it (not the cron). They put the checkpoint_timeout to 0. I will replace scp by rsync and sync the archivelog directory over wan every 5 mintues and put another cleanup script in cron to remove any archive older than 7 days.
Now, btw, I looking in the script that force Oracle to switch the log every 60 minutes and here is the command that do it... It look pretty strange to me
1. get the log sequence number : (select sequence# from v$log where status in ('ACTIVE','CURRENT');)
2. do a switch log file and archive all : (alter system switch logfile; alter system archive log all;)
3. Sleep 10 second ?! because it takes 10 second for Oracle to dump in archive log ?!
4. For each seqNo found in step 1:
4.1 select archived,sequence# from v$log where sequence# in ( ${seqNo} ); > file1
4.2 select sequence#,first_time from v$log_history where sequence# in ( ${seqNo} ); > file2
4.3 check if, for each (and there is only one BTW), is ARCHIVED = YES in v$log and present in v$log_history
4.4.Check, using fuser, if the file is written in UNIX : fuser ${archlogpath}/*${1}*${seqNo}*arc | cut -d: -f2 to be sure the file is close
4.5 scp the file over wan
Do I'm crazy or it's completely overkill ?!
ThanksI think you have a better understanding of oracle than the dBA who configured this!
On redo, yes your online redo logs should be multiplexed, with each member on a different mount point.
Likewise, your multiplexed control files should be on different mount points.
The script you describe sounds pretty bizzare. Does anyone have an explanation on why the archivelogs are being manually scp'd over the lan? Is that for simple backup purposes or is someone trying to re-invent DataGuard?