Re: Snapshot Logfile
- From: Carel-Jan Engel <cjpengel.dbalert@xxxxxxxxx>
- To: Paul Drake <bdbafh@xxxxxxxxx>
- Date: Mon, 29 Aug 2005 21:31:01 +0200
Hi Paul,
Your solution is definitely a good step forward in ruling out
role-specific namings in directories/SIDs/DB-names. I prefer not to use
the %_file_name_convert parameters. As you wrote, it adds (some/a
little) complexity to the setup. I personally prefer a symmetric setup.
I can imagine though that a setup according to your suggestion works out
in some situations. IMHO, however, a symmetric setup is less
error-prone, and more intuitive to me. The fact that I do not prefer it,
doesn't mean it is wrong though! I think we are getting close to a
matter of 'taste', in stead of 'good or wrong'.
It depends on the people involved, and the situation in the company.
When both sites are managed by 1 or 2 DBA's, and maybe at a distance of
30 KM, (or less than 1 KM!), even EAST and WEST doesn't mean that much.
Talking USA, east and west coast, and staff involved that might have
never met each other is different.
The thing is, when I work in a symmetric setup, I don't have to care
about anything in directory structures. I can use simple copy commands
like 'scp -pr * otherhost:`pwd`', which will simply send everything from
this directory to the equivalennt at the otherhost. I can even share
(most of) my parameterfiles among both instances.
I don't use SPFILEs. I use 5 files in a OFA 'pfile' dir:
generic_<SID>.ora - contains all parameters equal for each role
primary_<SID>.ora - contains all primary-role specific parameters
standby_<SID>.ora - contains all standby-role specific parameters
init<SID>_p.ora - contains IFILE= lines for generic and primary
init<SID>_s.ora - contains IFILE= lines for generic and standby
The init<SID>.ora no longer exists, neither the symlink in
$ORACLE_HOME/dbs. So, it is impossible to start the database without a
pfile= option. The DBA has either to script the thing, or think what
role he wants to activate. The script-set I implement at my customers'
sites takes care of it all, of course.
The generic_<SID>.ora is identical on all sites, and can be copied to
all sites easily after a change. The primary and standby files have some
site-specific parameters, and cannot be replicated. No need to tell that
this works very well in commandline based teams extremely well, and
probably not as well at GUI-based sites.
When the server that hosts the standby hosts another database in a
primary role (to make the costs of an 'idle' server more acceptable),
one can move memory sizing parameters from the generic to the
role-specific files, using different sizing depending on the role.
Not very much a strong advice in favor of or against your suggestion,
but merely a soft opinion this time, however still hoping it helps,
Best regards,
Carel-Jan Engel
===
If you think education is expensive, try ignorance. (Derek Bok)
===
On Mon, 2005-08-29 at 13:14 -0400, Paul Drake wrote:
> On 8/28/05, Carel-Jan Engel <cjpengel.dbalert@xxxxxxxxx> wrote:
>
> Hi Sami,
>
> Yes, that should work. The Rule of thumb form the doc probably
> should read
> '(max.nr.of log file groups for each thread +1 ) * maximum
> number of threads.'
> but the result will be the same. Alas, I've no DG/RAC
> experience yet, I might be wrong here.
>
> I do not particularly like your direcotroy naming, though. You
> use a directory named 'primary'. This suggests you have a
> directory 'standby' somewhere as well. What if you have to
> perform a switchover/failover? The name of the directory will
> no longer resemble it's role by then. You will have to deal
> with a confusing setup by then, where compnents of the primary
> actually are named standby and vice versa. This will make you
> system more (human) error prone.
>
> My advice is to keep systems on both ends as symmetric as
> possible, using the same direcory structure and
> instance/database names at both ends. That implies that naming
> conventions should avoid embedding roles in the names. Compare
> it with 'meaningless keys' in data modeling, which might
> survive longer than meaningfull keys. (I don't want to get
> that discussion loose here, and if it starts I will not
> contribute)
>
> A meaningfull name of a server, directory, instance or
> database can and will become problematic. One day, it's role
> will change and the name will start confusing you ISO of
> helping you. And that is something you do not want in an
> environment that is presumed to be highly available!
>
> Best regards,
>
> Carel-Jan Engel
>
>
> Carel-Jan,
>
> Are you saying that for Dataguard you do not advocate the use of
> location-specific subdirectories for file locations, e.g.
>
> <vol>\oradata\%db_name%\%db_unique_name_1%
> h:\oracle\oradata\mydb\eastmydb\
> h:\oracle\oradata\mydb\westmydb\
>
> in conjunction with the use of:
> db_file_name_convert
> log_file_name_convert
>
> Although at first it seemed as if it was introducing additional
> complexity, it sure does get rid of the idea of primary and standby
> being hardcoded into serivce_names and directory structures.
>
> It seemed to me that if someone else would have to take over in the
> event that my current location is obliterated ... that including the
> locations in the directory structure would make things more
> intuitively obvious at crunchtime. This is not to say that
> documentation and practice runs shouldn't be performed ... but one
> never knows who might have to pull the trigger if say the east coast
> data center isn't accessible.
>
> thanks,
>
> Paul
>
Other related posts:
