Unique Columns For V$Sql

Hi,

Until yesterday I was saying SQL_ID and CHILD_NUMBER is enough to get a sql child cursor from V$SQL and today I know I was wrong.

Probably many times you needed the sql statement that a session is running. To do that you query V$session and find the session information and then use SQL_ID column to reach out V$Sql. if you write a query like this:

you can write any filter you want here but you will immediately realized that some session rows are multiplied. Why? Because you didn’t use “CHILD_NUMBER” while reaching v$sql. V$SQL is not main data source for unique sql statements in your database. V$SQLAREA is. So instead of v$sql, you can use v$sqlarea in this query but if you need specific information about the exact sql cursor that session is currently running then you must use v$sql.

Note: I will not go deep Parent-Child cursors here. basically, every sql statements has child cursors. they all same sql but for some reason they have different properties and your session is running a “child cursor”.

so your query should be like this:

and now you reached to exact correct sql child cursor that is running for that session. right? well, I did know so but I find out, it is not! On one of my clients system, I discovered that this query returns 2 rows from v$sql which means for same SQL_ID and CHILD_NUMBER there are 2 rows in v$sql. I couldn’t find anything on docs or google (maybe couldn’t search enough) and started to dig about this sql statements.

here is the necessary information about this sql:

almost everything is same but ADDRESS and CHILD_ADDRESS. CHILD_ADDRESS is the memory address that this cursor resides on so it must be different obviously. Only thing that create this difference is ADDRESS column. ADDRESS column in V$sql is the address of the PARENT cursor which is in V$SQLAREA view. So let’s check this view:

there is only one of them exists. So some of those child cursors in v$sql are not (I am not sure the term) “valid”. they might be used before (and they were in my example) but now they are remaining of a past sql. So which child cursor in v$sql is the sql statements that session is currently running? Now we need to use ADDRESS column too. SQL_ADDRESS column in v$session is the parent address of the sql statement. So, I should add it into my queries too:

thanks for reading. comments are welcome. wish you healthy happy days.

APPLY with status WITH ERRORS in the PDB

Yesterday, I was doing some tests about noncdb to pdb conversions and during one of these tests I got a strange error. By the way, I am doing those tests because of a side effect of noncdb to pdb. I have already created an SR about it and write a post here later.

here is the environment in my test case; I installed Linux 8.6 on Virtual Box, Oracle 19c (19.16 RU) and two databases on same home, orcl and cdb19. orcl is noncdb database and cdb19 is the container database. I created both databases via dbca at the same time and dbca completed successfully (at least I thought so). Both databases are up and running. So I started noncdb to pdb conversion via autouptrade tool (latest version 22.5). I also tried, db link and unplug/plug methods too.

So after starting autoupgrade it completed successfully but when I login into database and open new pdb, it raises some errors. in pdb_plug_in_violations, I saw “APPLY with status WITH ERRORS in the PDB” error. For some reason, datapatch didn’t run correctly after dbca operation for orcl (noncdb) database. I run datapatch for orcl database again, it did some patching and re-run autoupgrade. After that operation completed and pdb created successfully.

by the way, my main problem still exists. this migrated pdb has its own data dictionary (which shouldn’t be because data dictionary is in cdb$root). anyway if you ever encounter with this error you can try datapatch again.

wish you all healthy happy days.

How to Check Empty Space in a Securefile LOB

Hi,

PS: please check last sql statement in this post for shorter and faster solution. first sql is for showing all information about the lob.

Yesterday, while checking segment space usage data in a client’s database and some of the LOB segments were absurdly big. So, I want to check what is in it and how much free space I can gain.

To do that, I cannot use just DBMS_LOB.GetLength function because for CLOB data type, it takes much more space than original data because of charsets (especially if you use AL32UTF8 charset) so I just want to check currently empty space in lob segment. if you need that kind of data then, DBMS_SPACE is the solution. DBMS_SPACE has a SPACE_USAGE procedure which checks a segment and  return space information but starting 12c, SECUREFILE is the default lob type and for securefile you cannot use same space_usage procedure with the one you use for tables.

space_usage has overloaded versions and second and third overloaded version (order in dbms_space package from top to bottom) is for SECUREFILE lob segments.

here is a sample script I wrote to check currently empty space in securefile lobs:

Caution: I check all LOBS in the database so, you might not want to do that, just modify sql and add necessary filters.

first with function block is to call dbms_space.space_usage procedure and return data in string, on second with clause (get_info) call the function and get lob information, by the way SEGMENT_TYPE for a lob segment is “LOB SEGMENT” but dbms_space.space_usage does not accept it as segment_Type, you must send “LOB” instead, then parse result as columns and list SEGMENT_SIZE_MB which is total size of LOB segment size in megabytes
USED_SIZE_MB which is actual used size in megabytes
EXPIRED_SIZE_MB which is expired data size (for versioning) in megabytes.

to calculate empty space just subtract used size from segment size. this is approximately empty size in your lob segment. if you have a partitioned table then you will get result for each partition (like mustafa.tmp table in the example). I hope this script helps in your job.

EDIT:

I forgot to mentioned why lob segments have empty spaces. biggest reason is delete operation of course. if you delete a big portion of the table (or lob segment) then deleted space will be considered as empty space. So, if you delete on a table, should you shrink the lob segment or table? NO! that space will be used by the table again. if this is your standard operation then empty spaces in table/lob segment will be reused but let’s say you inserted many rows by a mistake or because of a bug in your code and then deleted those rows. if you have enough disk space for the database then there is nothing to worry about (mostly). That table will get many inserts in time and it will use that space eventually but if too much space is reserved in the table / lob segment and maybe you have a disk space issues then you can reclaim it (by shrinking or moving table/lob segment). So, this is not a regular thing.

EDIT 2:

sql above is aiming to show all information about lob like used space, total space, expired space etc. probably , you will just need to see empty space in lob. so you can use this sql statement. this will work faster because no regexp usage and parsing operations. I also subtracted “unexpired bytes” from total size since this space is still in use.

 

 

thanks for reading, your comments are always welcome. wish you healthy days.

catcon.pl cheat sheet

Hello,

Oracle uses a container database architecture permanently anymore, you should be familiar with “catcon.pl” perl file. Let’s say you have container database with 10 different PDBs and you need to run a script file on all of those pdbs (maybe even in root and seed too) so what you need is catcon.pl.

catcon.pl will automatically execute your script file and run it on all or specified containers. Since it is a perl file it must be called with perl binary (perl in linux, perl.exe in windows) and you can find it under $ORACLE_HOME/perl/bin directory and catcon.pl itself is under $ORACLE_HOME/rdbms/admin .

so, you have a sql file to run under the /home/oracle/scripts directory called myfile.sql. let’s see how we can execute it in all pdbs:

that is it. myfile.sql file will be run in all containers (including root and seed) with the command above. Let’s dissect the command:

$ORACLE_HOME/perl/bin/perl => PERL binary to run catcon.pl
$ORACLE_HOME/rdbms/admin/catcon.pl => catcon.pl path
-d /home/oracle/scripts => directory where script is located.
-b MyScript => Prefix for the log files so you can separate them from other scripts logs.
myfile.sql => actual script name

actual script file must be the last parameter. It is that easy. in this example ORACLE_HOME and ORACLE_SID should be set before you run the command otherwise you cannot connect to the database. This is a local connection and by default it is a SYSDBA connection. So whatever you run will be run under SYS user.

To add some more functionality:

  1. -c : list of containers that you want to execute the script:
    ….catcon.pl -c ‘SALES PDB1 TEST’ => execute the script only at SALES, PDB1 and TEST pdbs.
  2. -C : list of containers that you don’t want to execute the script.
  3. -l : log directory
  4. -s : spool the output of script files.
  5. -S : (capital S) don’t execute the script in root and seed. (Edit: this option also don’t set “_oracle_scripts” parameter as true, so if you are creating a user it won’t be Oracle Maintained user)
  6. -u : specify username/password.
  7. -z : ez-connect string
  8. -n : number of workers. if you want to execute a script for 100 pdbs then you can increase the number of workers to execute script simultaneously on different pdbs. default is 1

Can we use catcon.pl to run a script on a remote server? Yes with a combination of -u and -z

run myfile.sql script under /home/oracle/scripts directory on CDBX database and only specified pdbs (PDB1…PDB8) which is on mydbserver.mydomain server using mustafa user and create 2 worker process so script will be run on 2 pdbs at a time and write log files under /home/oracle/logs.

catcon has more parameters but I just write about the ones I used most.

thanks for reading.

Expanded controlfile section n from nnn to nnn records

Hello everyone,

I have been already write revealing information that control file stores: https://mustafakalayci.me/2020/08/11/control-file-content-and-the-limits/.

Data in control file is separated into “SECTIONS” and you can see those sections name (type) via V$CONTROLFILE_RECORD_SECTION view. for example, database separates a section for “BACKUP SET”s. every section has a record size limit . All those information can be queried via v$controlfile_record_section of course.

RECORDS_TOTAL number of records allocated for that section.
RECORDS_USED number of used records.

You might encounter an alert log message from time to time like “Expanded controlfile section 11 from 100 to 200”. this is a message that database need to store more data in control file and currently allocated space is not enough, so oracle expands the control file. in this sample message, 100 is current RECORDS_USED and RECORDS_TOTAL. since all records are filled, it expands to 200, so RECORDS_TOTAL become 200. how about the section? what is section 11? unfortunately there is no SECTION_ID but V$controlfile_record_section always return rows in specific order and you can use this select to learn section ids:

So, 11 is ARCHIVED LOG. which means to be able to store currently required (based on your deletion policy in rman) database need to increase number of records in control file.

Database needs to increase those sections when required data to be stored increase. You can see the alert log messages above when this occur.

wish you all good, healthy days.

lsnrctl or new db connection is too slow

Hello everyone,

I know this is not new but I haven’t seen this problem for a long time and it takes a while to find out it.

A team leader (who is an elder brother to me) from the company I started to my first job called me and asked about a problem on their customer. Their customer complained about too slow db connection and no problem after the connection. first things first, I wanted to check database availability but because of some PATH problems (this was a Windows server) it took some time. after fixing environment variables, I finally reached out to database via local sysdba connection and database is up and running. also, connection was fast. by the way database was an 11.2 version which is quite old.

So, I started to focus on LISTENER. I first checked the status but “lsnrctl status”  command was quite slow. then I stopped and started again but these two operations was also slow. I checked the logfile under: $ORACLE_BASE/diag/tnslsnr/<SERVER_NAME>/<LISTENER_NAME>/trace but everything seemed ok! When I tried to logon via listener on the database server (using the net service name) it took around 70 seconds.

I started a listener trace (adding tracing parameters into sqlnet.ora file and restarting the listener) and tried to make a connection again. as soon as I issue my connection command (sqlplus username/pass@db) trace file is created. this means network connection was fast but while reading the trace file, after passing the connection information, listener was waiting for around 60-70 seconds.

I understand that network has no issue, connection arrives immediately but db connections and lsnrctl commands were slow. as I said I already checked, listener trace directory because I know that if listener.log file under trace directory has a big size, this could cause a slow operations on listener but it was around few megabytes. I checked few more things and finally I realized that I never checked “alert” directory under diag ($ORACLE_BASE/diag/tnslsnr/<SERVER_NAME>/<LISTENER_NAME>/alert) and when I opened it, there were over 1900 log.xml file which is the xml version of listener.log file. I deleted them all and it worked like a charm. Listener was checking all xml files to decide to write log data in which xml file. that was the cause and we were able to solve the problem.

Always check your listener log files and keep their size minimum as much as possible.

Wish you all healthy days.

SESSION_CACHED_CURSORS Parameter and How can It Affect the Database

Hello,

Let’s talk about the SESSION_CACHED_CURSORS parameter. It might have significant affect on your system so understanding how it works and what it cause is important.

This parameter limits the number of cached cursors on the session. As you know there is 2 kind of parse SOFT and HARD. Hard parse is creating the cursor from the scratch. Checking grants, syntax and semantics, creating execution plans etc. Soft parse is using a previously created cursor. So, if Oracle has already calculated all information for a query (sql statement, cursor) then, it stores it in the shared pool and can use it over and over again. not calculating everything from scratch is a big performance increment.

Even in a soft parse, Oracle still need to search through library cache and find the cursor first. SESSION_CACHED_CURSOR allows you to cache a cursor in your sessions PGA (and still in SGA). This way if you run a query over an over, it will be found and execute even faster. Basically, SESSION_CACHED_CURSOR is kind of “Softer” parse. Starting from 11g its default value is 50 and in 21c it is the same.

You can see cached cursors via V$OPEN_CURSOR view. Let’s open a new sqlplus session and check its cached information. to check cache information I will use a separate session because I don’t want to affect the main session. here is main session:

so my main sessions SID is 1470. from monitoring session I will check cached objects in the session:

I won’t be able to put all output because it will be much bigger in a minute. simple explanation for cursor types:

OPEN is currently opened (last run) sql statement.
OPEN-PL/SQL is currently opened cursors in plsql. it could be more than one, you can open many cursor remember that.
OPEN-RECURSIVE is recursive sql statements for your actual statement (for example Oracle needs a privilege check if you have access to those objects, it runs a query to find that)
SESSION CURSOR CACHED is half of what we are looking for here. direct sql statements which has been run in this session and cached in session PGA.
PL/SQL CURSOR CACHED is the other big half of what we are looking for here. Sql/plsql statements which has been run in a plsql object which is called in this session.

there are few more types but I won’t go detail all of them right now. if session runs a sql statements directly (select, insert, update, delete, merge etc) it will be cached as SESSION CURSOR CACHED. if session calls a plsql objects, all sql statements in it will be cached as PL/SQL CURSOR CACHED. Remember, to cache an sql statement, it must be run at least 3 times, only then it will be cached. by saying “run” cursor must be opened and closed. if you just run a query 3 times, you will not see it in cache because last one is still opened. run something else and then it will be in the cache.

you can see session cached cursors count via session statistics:

“session cursor cache hits” is how many times a new statement has been found in session cursor cache. if you  run a query 100 times repeatedly, after first 3 (after 3 runs it will be cached) next 97 will be read from cache and session cursor cache hits will increase by 97. “session cursor cache count” is currently number of cached objects (all sql, plsql and dictionary object queries).

as I said by default SESSION_CACHED_CURSORS is 50 so my session can cache up to 50 different statements and if there is more? Oracle will use a Least Recently Used (LRU) algorithm to keep most used (and recently) queries. Previously issued statements will be deleted from the session cache (not from library cache).

Let’s start some tests and run 3000 different queries 3 times (so they can be cached). of course instead of calling 3000 different queries I will use a simple plsql. those queries will be cached as PL/SQL CURSOR CACHED. Also, I will create a tmp table as copy of dba_objects to do my tests:

before continue, you might wonder why I used dynamic sql. if I use a static sql as “select count(*) from t where object_id = i” then i would be a bind variable and every query I run would be same. so, we couldn’t see the difference.

From the monitoring session let’s check V$OPEN_CURSOR for main session (a little bit long output)

as you see, 50 different select statement has been cached which IDs starting from 2951 to 3000. out last 50 queries. also OPEN cursor is our initial plsql code.

so what would happen if we increase the cache? is it a good thing or a bad thing? well, it depends. it is a good thing because you will be doing “softer” parse. your session will access to repeated queries faster but since this information is stored in PGA (session specific memory area) it will increase the usage of PGA. if you don’t have enough PGA it could be a pain for your database. PGA is automatically managed by Oracle. it grows or shrinks based on session needs. So, you might end up too much pga resize operation and even cannot cache that much queries at all.

Let’s check current PGA size for our main session now, from monitoring session:

so our main (testing) session is using 3 MB pga size. now, I will increase the size of session_cached_cursors to 1000 and run previous 3000 queries again.

check the pga size now:

now it uses 8MB of PGA. Numbers could be small but think it this way; we are using 2.5 times higher PGA then previous. Also, if you have 10.000 active sessions, this could be a problem. it is a high usage. if you have enough memory then, there is nothing to be afraid of or at least you should be adjusting your memory components accordingly. there is no correct size or ratio because this is highly depend on your application coding. You must test the values because if your application uses same queries during a business job then, you can have a smaller session_cached_cursors but if not then, you must increase the size based on your memory.

how about the good side? Yes, let’s measure that. Assume we have a business process running with 400 different queries for every run and run this process 1000 times in a day. how it would affect the performance increasing or decreasing the parameter. I will start with 100.

it took 29.38 seconds to run all queries. Now increasing the parameter value to 400 (which is almost sufficient to store all queries I run):

24.63 seconds. 5 seconds shorter which means around %17 percent faster. which is not bad and actually good. think your system is running faster by %17 percent without doing a complex sql tuning or changing the hardware. if you set session_cached_cursors parameter between 100 and 400, elapsed time would be between 24 and 29 accordingly, pga usage too.

finally, you can also query how many times your session used session cached cursors :

as you see currently, 399 different queries (sql statements) has been cached in the session and those queries has been accessed 2.011.646 times.

This is a quite nice trick to me but must be used cautiously because of the increase of PGA usage.

I hope this helps to increase your databases efficiency.

Wish you all healthy days.

Marked as unrecoverable datafile in V$datafile even if they are not!

Hello,

it has been a while since I posted. I lost a close relative to me and it wasn’t easy. So, let’s continue.

I run orachk on one my development databases yesterday and result was very informative. Interestingly, I had a “FAILED” check that indicates some of the datafiles are not recoverable! This is a serious issue so I checked. Check detail indicates result of the query below shouldn’t return a row:

and there were rows returned indeed. this database is not in force logging mode because it is just the development database and customer agrees with that. for out of curiosity I checked the production database and there were some datafiles returned from the query above. that concerns me. production database is in force logging mode . When I checked unrecoverable_time column, I realized that dates are really old and this is odd to me because I get full backup every day. How could these files be unrecoverable?

I run “report unrecoverable” command on RMAN and it shows nothing (for both production and development database). That is good but why those datafiles marked as unrecoverable and still listed as that way? if you do a unrecoverable operation on a table then, datafiles that those tables are stored in will be marked as unrecoverable in V$datafile.

of course, it is turned out this is a lack of knowledge of me about unrecoverable_time column on v$datafile. here is 19c documentation:

https://docs.oracle.com/en/database/oracle/oracle-database/19/refrn/V-DATAFILE.html#GUID-7BF7955C-9705-40F4-B2F6-5D7F3A32DD30

UNRECOVERABLE_TIME DATE Timestamp of the last unrecoverable change. This column is updated only if the database is in ARCHIVELOG mode.

This column is “updated only”. so, if you did a unrecoverable operation like sqlloder direct load on a table, related datafile will be marked as unrecoverable and unrecoverable_time column will be updated on v$datafile. Even after a full backup, this won’t be updated back to null! it will just keep the last unrecoverable time information.

Better way to check unrecoverable datafiles could be this:

This also make me think why ORAchk runs the first query at all since it is not provide an exact information about unrecoverable datafiles. Maybe a minor bug or maybe still there is something that I don’t know.

Edit: I realized that I didn’t mention how to clear unrecoverable_time and unrecoverable_change# columns. it won’t be clear if you take a full (or any kind of) backup. these information is stored in control file and as I mentioned before, this column is only updated. So, you need to recreate your control file and this operation requires a shutdown.

Generate “create controlfile” statement, shutdown database, startup in nomount mode, recreate controlfile, recover database if needed and then open the database. one post step is to catalog recovery area because this new control file won’t have any recovery file information.

alter database backup controlfile to trace as ‘/tmp/my_control_file2.ctl’ noresetlogs ;
shutdown immediate;
startup nomount;
create controlfile … — use noresetlogs option for not to loose any data
alter database open;

thanks for reading. wish you all healthy, happy days.

How to Convert LONG to Clob PART 2

Hi,

I just want to write a small post today. Some time ago, I shared a way to convert long to clob in a query directly: https://mustafakalayci.me/2020/04/06/how-to-convert-a-long-column-to-clob-in-a-query/ Even if this method works, it works really slow and anything slow annoys me!

So, I want to share another method which is much much faster than Sys_dburigen function but THIS IS UNDOCUMENTED and UNSUPPORTED! It is used by internally by oracle so it works fine but you must be careful. Since, no one uses LONG column in their application (right? you don’t! please don’t!) this kind of code is handy for mostly admins because for backward compatibility data dictionary has some LONG data type columns.

What we are going to use is a LIBRARY called UTL_XML_LIB . I almost never create objects under SYS user, so I will create this code under a privileged admin user. first we must grant privileges on UTL_XML_LIB library and then create an EXTERNAL procedure in plsql.

now create external procedure:

“kuxLong2Clob” is a function which returns void (basically it is a procedure) written in C by Oracle. this procedure takes a select statement for your LONG column and rowid. procedure returns CLOB data as an out parameter. I created my_long2clob procedure but to be able to call it via SQL statement, it must be a function, so:

to test it, I won’t create a new table with a LONG column but instead I will use one of in the data dictionary (which I use these kind of codes against). Let’s use SYS.TRIGGER$ table which is the base table for DBA_TRIGGERS view. ACTION# column is a LONG data type. of course we must first grant select privilege on SYS.TRIGGER$ to mustafa user.

I suppressed the output but here it is. really fast long to clob conversion in SQL.

as a CON, it will only work against tables not VIEWs! you cannot use it as “f_my_long2cloc(rowid, ‘DBA_TRIGGERS’, ‘TRIGGER_BODY’)”. Since it requires ROWID, you must be referring underlying table and its rowid.

beyond that, if you want to get LONG as VARCHAR2, similarly you can use “kpdbLong2Varchar2” procedure in DBMS_PDB_LIB library. as UTL_XML_LIB it is very fast. only difference is, it returns VARCHAR2 and trims the overflow part. Where is it in use? Check how DBA_VIEWS shows TEXT_VC column data 😉

I hope this helps to fight against LONG columns in your codes.

Wish you all healthy, happy days.

Zero Data Loss DISASTER Database for Standard Edition 2

Hi,

As you know Data Guard is an Enterprise Edition option and only accepted method to protect your data against a “DISASTER” (DR) is data guard. How about Standard Edition 2(SE2)? Is this possible? Well I did it so, king of yes 🙂 with some risk of course.

Warining: I don’t take any responsibility for wrong settings. Wrong actions can damage your database. Please be careful.

if you google Oracle SE2 Data guard or Manual Log Shipping, you will find many blog post and documents about it. Even youtube has videos about this but almost all of them (at least all I’ve seen and read) are not “Zero Data Loss”. Steps are quite easy:

  1. you have a production database based on SE2
  2. you use another server as your DR server. (it could be on the same server if you use duplicate snapshot method)
  3. copy all your backup
  4. restore database and keep it in MOUNT mode
  5. copy your archivelog ever X minutes to DR server
  6. recover database using rman or sqlplus.

And that is it. This is a valid method. Only GAP is in this, you are transferring your archive logs but not redo logs. This means, you will certainly loose some data in case of a disaster because changes in redo logs will be lost (if you cannot reach out the server). So, what can we do? basic answer is by copying redo logs too but how? Redo logs are changing constantly. Even if you copy redo logs with in a 1 minute interval, you are still be able to lost 1 minute data.

To achieve this, you must force database to write your both redo logs and archive logs onto DR server. Not copying files in an interval, force database to write it there. Probably, you already understood what I did. to use this kind of method you should (must) have a strong, reliable, stable, fast network connection between your production and disaster servers. Because, we will send redo logs and archivelogs instantly to DR server.

I am using 19.15 databases on Oracle Linux 8 (virtual box). Same things can be done in other operating systems too. Logic will be same.

Here are the steps that I am about to do:

  1. On the DR server, create a folder to share over network via NFS.
  2. On the PROD server, Mount this shared folder on PROD server.
  3. On the PROD server, Add an archive log destination to this shared folder.
  4. On the PROD server, Add redo log members for each group to this shared folder.
  5. On the DR server, create a copy of prod database via rman (not duplicate)
  6. On the DR server, catalog archive logs, which resides on shared folder and recover database.
  7. In a disaster situation, copy redo logs to their original location with original name and recover database.

that’s it. Let me demonstrate those steps.

I created 2 virtual machines. They have both Oracle DB Software and one of them (PROD) has a database and I want to use other virtual machine as DR.

  1. Oracle home is under /u01/app/oracle/product/19c/dbhome_1
  2. I (always) use OMF (Oracle Managed Files).
  3. data files are under /u01/app/oracle/oradata
  4. recovery area is under /u01/app/oracle/fast_recovery_area
  5. SID of database is cdb
  6. PROD server ip is 192.168.56.101
  7. DR server ip is       192.168.56.104

1- On the DR server, create a folder to share over network via NFS

As the oracle user (Oracle db software owner user), create necessary directories.

this is the folder that I will share over the network and mount it (map it) on the production server.

to share over network, as root user:

2- On the PROD server, Mount this shared folder on PROD server

as oracle user, create a directory to mount shared directory on DR. /u01/app/oracle/fast_recovery_area/DR_LOGS will be used for mount point. directories in it is for OMF.

as root user:

at this point if I put a file into /u01/app/oracle/fast_recovery_area/DR_LOGS directory on PROD server, it will be written to /u01/PROD_LOGS on DR server

3- On the PROD server, Add an archive log destination to the shared folder

By default for OMF, if there is no value for log_archive_dest_n parameters, then Oracle internally use log_archive_dest_1 parameter as LOCATION=USE_DB_RECOVERY_FILE_DEST. that way, archivelogs are written under /u01/app/oracle/fast_recovery_area directory with sub directory structure of course (/u01/app/oracle/fast_recovery_area(<SID>/archivelog/YYYY_MM_DD).

but! if you assign a value to any of those parameters, then it will just use that location to create archivelogs. so I will set 2 log_archive_dest_n parameter, one location for PROD server and one for shared folder.

by setting these 2 parameters (log_archive_dest 1 and 2) we make sure that oracle will create 2 copies at the same time to 2 different location and one of them is actually out DR server. As you see there is one more parameter which is very important. log_archive_min_succeed_dest, by default is 1 and if you don’t set this as 2, in an access problem situation (network might have issues or DR server could be shutdown), Oracle will just write the first location and pass the inaccessible location. this will cause not writing necessary archive logs to DR server. if we want “zero data loss” DR db then oracle must write archive log to dr server. (also you can use MANDATORY clause).

4- On the PROD server, Add redo log members for each group to the shared folder

as the archive logs, we need to do the same things for redo log too.

as log_archive_dest_n parameter, if we set any of db_create_online_log_dest_n parameter, we need to set all three of them. 2 locations are on PROD server and 1 location is on DR server. this is not a mandatory step actually but it is a good practice. Because, we can just add redo log members (last 3 command above) to the shared location and this is enough but if one day you need to create another log group and forget to add a log member to shared folder then, you could be exposed to data loss! so set it any way.

then, add 1 member for each log group to shared folder. I use standard naming here because I need to know the groups of files by their name (we will use this later).

lastly, I create pfile to move to DR server, so we will create our DR database.

5- On the DR server, create a copy of prod database via rman

on PROD server, database “cdb” is up and running. Let’s start:

I don’t want to go all the details about standard copy backups and restore a database so, I won’t put outputs here.

On PROD server, take a backup of database via RMAN and then copy everything to DR server via scp.

from shell:

copy all backups, archivelogs etc and also pfile.

then go to DR server and create necessary directories & restore your database:

these directories will be needed for database. restore database:

at this point we have a copy prod database on DR in MOUNT mode.

somet very important thing to do

DO THIS ON DR SERVER ONLY! we are deleting any copied redo logs because during recovery mode, there must not be any redo log!

6- On the DR server, catalog archive logs, which resides on shared folder and recover database

from now on, you can create a recover.sql script and call it ever n minute from cron to recover database regularly. content of recover.sql is:

we must catalog newly arrived archivelogs first and then recover database. I want to remind that /u01/PROD_LOGS directory is the one that we shared over network and map it on to prod server.

every recovery command will end up with an ERROR:

this is an expected behavior. since we don’t apply redo logs, Oracle is looking for the next archive log file in the sequence but couldn’t find it. our recovery is now working. after archive log with sequence 18 is arrived, recover command will raise the same error above for sequence 19.

7- In a disaster situation, copy redo logs to their original location with original name and recover database

from now on, our archive and redo log files are stored on DR server too. if there is a disaster situation, you must open the database with applying all logs including redo logs. this is the only moment that we will apply redo logs.

these are the redo log locations on PROD:

/u01/app/oracle/oradata/CDB/onlinelog
/u01/app/oracle/fast_recovery_area/CDB/onlinelog
/u01/app/oracle/fast_recovery_area/DR_LOGS/CDB/onlinelog (this one is the network directory on DR server)

actual directory that redo files are written is /u01/PROD_LOGS on DR server. Those directories must be existed on DR server too (we already created them in scripts above). We must copy redo logs under /u01/PROD_LOGS to actual locations now. I deliberately named redo logs as redoN.log so that I can identify which redo file is belong to which redo group. a very simple sql script and shell script will copy redo logs to original locations:

calling put_redo_logs.sh file will run an sql script which will produce copy shell commands.

After that, you can run recover.sql one more time and this time RMAN will not raise any error and you will see “Recovery Completed” message. now you can open your database and use it.

Now, why did I delete copied redo log files via scp command on DR server? if you recover a redo log in mount mode, control file will raise some flags and all recover commands will start from the sequence of applied redo log, this will cause problems after some time because archivelogs will be deleted eventually because of retention policy. This also means, if you apply a redo log during recovery, to have a healthy copy of database, you must restore both control file & database all over again.

Pros

obviously, you have a DR db with full consistent data.

Cons

This approach contains high risk because if database cannot write archived or redo log to shared folder then production database will hang! You might have some issues about network or DR server might be restarted or shutdown. You might loose disk drives on DR server which shared folder resides on etc etc…

I will write another post about those situations and make some tests. if loosing some data for a small time then, still refreshable pluggable database would be a better option. I will also write a post about it.

if you don’t have a reliable network, this option could be catastrophic! So, try it carefully. I hope this helps in some way.

Wish you all healthy, happy days.