Generating test data sets with Swingbench

Reading about new features is great, but finding data to test them out on is  sometimes troublesome.  As DBAs, we’ve created test databases countless times, but the majority of the time they are empty.

Oracle does provide sample schemas that can be installed automatically by using the database configuration assistant (DBCA) when creating a database or manually by executing scripts under the Oracle software home directory.  Consult the documentation of your particular version for more details.  While these sample schemas do provide a framework that mimics a common layout, they lack the volume of data that typical production systems have.  If your employer allows “sandboxes” to be created to test new versions and features then you can stop reading here because you are one of the fortunate ones.  However, if you do not have this opportunity but would still like to create a sizeable data set to work with, there is a utility called Swingbench that fills this need very well.

Swingbench was created by Dominic Giles ( as an Oracle database load generator, but has since become a very useful suite of tools including Data Generator, Trace Analyzer, and several monitoring tools.  Best of all, these tools are free to use.  I would recommend that you provide a small donation if you find them useful, though.

Quoting the creator, “Data Generator is a utility to populate, create, and load tables with semi random data. Users can specify sequences, random text, random numbers, database columns, and files as sources for tables.”  This sounds pretty good.

I recently downloaded Swingbench version 2.5 which includes the Data Generator utility and worked with it on a Oracle VirtualBox virtual machine running Oracle Enterprise Linux (OEL) version 6 update 6 with and Oracle version non-container database.  Here are the steps I followed to create my test data set:

  • Swingbench requires Java 8 JDK.  Download the jdk-8u111-linux-x64.rpm and install it (as the root user) with the command: “rpm -Uvh jdk-8u111-linux-x64.rpm”. Note that this file is for the 64-bit version of OEL.
  • Download the file for version 2.5 from the website mentioned previously and place it in a location where the oracle user account has read, write, and execute privileges.
  • Unzip the file.  it creates a “swingbench” directory in this location
  •  As the oracle user, set the following environment variables if they are not already set (actual values depend on your environment):
    • JAVA_HOME=/usr/bin/java
  • Go to the swingbench/bin directory.  You have the option to create the sales history schema (shwizard), the calling center schema (ccwizard), or the order entry schema (oewizard).
  • Once you execute the desired wizard file, the interface appears



  • After clicking Next, you have the option to create or drop your desired schema.


  • For the connection string, use //<servername>/<database_sid>. Make sure you can ping the host by the <servername> used and that there is a listener present providing service for the <database_sid>.  Leave the connection type as the default (jdbc thin).  Leave DBA username as ‘sys as sysdba’ and make sure the password is correct.  Press Next to connect.


  • If the connection is successful, you will see the next screen where you can choose the username, password, schema tablespace, and tablespace datafile of your data set schema.  I like to keep my data separated for easier management, so I elect to create a separate tablespace and datafile.


  • The next screen allows you to choose database options such as partitioning (range or none), compression (yes or no), tablespace type (bigfile or normal), and indexing (all, primary key, or none).  Be careful here.  Do not select options that you are not licensed for in systems that are eligible for license audits.


  • Sizing options come next.  I don’t suppose I need to tell you not to size your schema any bigger than you have space available.  What is good here is that two counters at the bottom will let you know the size of the schema tablespace and the size of the temp tablespace required based on your selection.  Click next when you are done.


  • Finally, select the level of parallelism and click finish to create your schema.


  • The wizard log popup appears and the main wizard updates to tell you the progress of the creation process


  • Once the build is over, you should see a “successful” message popup window.  Click Ok to acknowledge the message and close the window.  You then have the option to review the wizard log in the wizard log window or save it to a file.  Close both the wizard log and main wizard windows after you are done.


At this point, you have the option to generate other data sets if you want to.  Enjoy.



Sequences are not gap free!

mind the gap

Yesterday I encountered a trace file on a “poorly-performing” process with some interesting waits.  The trace was incomplete because the trace process was started once the process was identified as “poorly-performing”.  Even so, it was interesting.


call     count       cpu    elapsed       disk      query    current        rows
------- ------  -------- ---------- ---------- ---------- ----------  ----------
Parse        0      0.00       0.00          0          0          0           0
Execute      0      0.00       0.00          0          0          0           0
Fetch        0      0.00       0.00          0          0          0           0
------- ------  -------- ---------- ---------- ---------- ----------  ----------
total        0      0.00       0.00          0          0          0           0


call     count       cpu    elapsed       disk      query    current        rows
------- ------  -------- ---------- ---------- ---------- ----------  ----------
Parse        0      0.00       0.00          0          0          0           0
Execute     78      0.03       0.05          0        106        106          53
Fetch       25      0.01       0.03          0         75          0          25
------- ------  -------- ---------- ---------- ---------- ----------  ----------
total      103      0.05       0.08          0        181        106          78

Trace file: PROD2_ora_2406.trc
Trace file compatibility:
Sort options: exeela  fchela  prsela  
       1  session in tracefile.
       0  user  SQL statements in trace file.
       2  internal SQL statements in trace file.
       2  SQL statements in trace file.
       2  unique SQL statements in trace file.
  384089  lines in trace file.
    1507  elapsed seconds in trace file.


Ok, we have 1507 seconds of tracing, or slightly over 25 minutes.  Of that, only .08 seconds are accounted for between the recursive (system) and non-recursive (user) statements.  Where did the time go?  There were no event waits under the non-recursive statements section, but the wait events for the recursive statements was very telling.  Note that this is not a complete listing, I just filtered out any waits that did not add up to more than 1 second.

Elapsed times include waiting on following events:
  Event waited on                             Times   Max. Wait  Total Waited
  ----------------------------------------   Waited  ----------  ------------
  enq: SV -  contention                      289471        0.48       1210.91
  gc current grant 2-way                      22811        0.13         17.21
  db file sequential read                     57139        0.53         83.05
  gc current block 2-way                       2094        0.02          1.94

We now see that about 1211 seconds, or about 20 minutes of the trace, was spent on “enq: SV – contention”.  But what is that?  This event is sequence value (SV) contention and appears when there is a high demand for sequence values across multiple instances in a RAC environment.  I typically see “log file sync” waits that point to sequence contention issues in non-RAC environments.

When sequences are created they have a default cached value of 20.  This means that every time a sequence is accessed, the next 20 values are cached in memory for quick access.  This value needs to be much higher when the sequence is used to generate values for inserts when hundreds of thousands or millions of records are inserted in one process or routine.

I always recommend increasing the size of the sequence cache values when I see these type of situations, but I am often met with one of two very weak arguments.  One is that the client does not want to have to recreate the sequences to increase the cache value while the other is that the client wants to keep the cache small in order to avoid gaps in sequence values.  Let’s take these arguments one at a time.

First,  altering the sequence cache value dynamically has been available since Oracle database version 10g release 1.  In fact, there is a quote in the Oracle 10g release 1 SQL reference guide which states “Oracle recommends using the CACHE setting to enhance performance if you are using sequences in a Real Application Clusters environment.”  The rest of this argument is disproved by creating a test sequence and changing the cache value with a simple “alter sequence” command.

The second argument is the most exasperating for me because the client usually won’t budge from their position.  Although the client is correct when they state that a smaller cache size reduces the risk of sequence value gaps, they are mistaken when they assume that a reduced cache will ensure no sequence gaps.

If you are relying on sequences to generate primary keys for newly-inserted records, there is NO guarantee that there will be no sequence value gaps.  Even Tom Kyte is in agreement here (

The closest you can get to no gaps using a sequence in this fashion is creating the sequence with the “INCREMENT BY 1”,  “NOCACHE”, and “ORDER” options together.  There will be a MAJOR performance cost using a sequence in this fashion for generating values during large insert activity.  Even with these values set, there is still the possibility of sequence value gaps.  Here is a version 11.2.04 demonstration of that fact:

SQL> create table jontab2 (jonkey number, jonstring varchar2(20));

Table created.

SQL> create sequence jonseq start with 1 increment by 1 nomaxvalue nocycle order nocache;

Sequence created.

SQL> insert into jontab2 values(jonseq.nextval, 'String');

1 row created.

SQL> insert into jontab2 values(jonseq.nextval, 'String');

1 row created.

SQL> commit;

Commit complete.

SQL> insert into jontab2 values(jonseq.nextval, 'String');

1 row created.

SQL> insert into jontab2 values(jonseq.nextval, 'String');

1 row created.

SQL> insert into jontab2 values(jonseq.nextval, 'String');

1 row created.

SQL> insert into jontab2 values(jonseq.nextval, 'String');

1 row created.

SQL> insert into jontab2 values(jonseq.nextval, 'String');

1 row created.

SQL> insert into jontab2 values(jonseq.nextval, 'String');

1 row created.

<I crash the instance here by killing the PMON process>

SQL> insert into jontab2 values(jonseq.nextval, 'String');
insert into jontab2 values(jonseq.nextval, 'String')
ERROR at line 1:
ORA-03135: connection lost contact
Process ID: 3657
Session ID: 125 Serial number: 7

SQL> exit
Disconnected from Oracle Database 11g Enterprise Edition Release - 64bit Production
With the Partitioning, OLAP, Data Mining and Real Application Testing options

<I restart the instance and login>

[oracle@rac1 ~]$ sqlplus jontest/semperf1

SQL*Plus: Release Production on Fri Aug 4 09:10:01 2017

Copyright (c) 1982, 2013, Oracle.  All rights reserved.

Connected to:
Oracle Database 11g Enterprise Edition Release - 64bit Production
With the Partitioning, OLAP, Data Mining and Real Application Testing options

SQL> insert into jontab2 values(jonseq.nextval, 'String');

1 row created.

SQL> commit;

Commit complete.

SQL> select *
  2  from jontab2
  3  order by jonkey;

---------- --------------------
         1 String
         2 String
         9 String                        --------GAAAAAAP!!!!!!!

SQL> select jonseq.nextval
  2  from dual;



Because I did not commit some newly-inserted records before the instance crashed, they were not written to the table even though the sequence incremented and created a gap in the key values of the table.  Even if you have a batch process that commits after every record inserted, you can’t account for instance crashes.

If you STILL intend to use a sequence alone to generate a “no gap” primary key, what happens if you have to delete records from the table?  Guess what, you now have gaps in your primary key values.  What are you going to do, re-index all the remaining records to remove the gaps?  I doubt it.  The best bet is to use the sequence and one or more other values in the table, perhaps even a timestamp, as the basis for a primary key value.  This way, you can increase the sequence cache and still have solid primary keys.  You just need to get over the “no gap sequence values” idea.


Oracle Multitenant – Part 2 – Creating, connecting to, removing, and querying CDBs and PDBs

Now that we have an understanding of the multitenant architecture from part 1, we want to add some practical experience starting with the creation of CDBs and PDBs.  My test environment will be Oracle VirtualBox v.5.1.22 guest running an Oracle Enterprise Linux 7.3 guest with Oracle version software installed.

Creating CDBs and PDBs

NOTE – If you create a CDB under Oracle 12c release 1, the following is not supported:

  • Flashback Transaction Query
  • Flashback Transaction Back out
  • Database Change Notification
  • Continuous Query Notification
  • Client Side Cache
  • Heat map
  • Automatic Data Optimization
  • Oracle Streams
  • Oracle Fail Safe release 4.1.0 that ships with Oracle Database

The preferred CDB creation method is through the Database Configuration Assistant (DBCA).  The DBCA wizard will guide you through the process and automatically execute the proper catalog scripts once the CDB is created.  The DBCA can also be used to generate all the scripts you need to create the CDB manually based on choices made through the DBCA create database wizard.  The DBCA can also generate a CDB creation template that you can use to run DBCA in other environments if you copy the template over.

Once you login to the OS as the Oracle software owner and set your environment, execute $ORACLE_HOME/bin/dbca


The obligatory Oracle splash screen appears


Choose the ‘create database’ option and proceed to the next screen


We are going to choose ‘typical configuration’ here to save time.  On this screen, we can select the storage type, DB file location, character set, and designate the administrative (SYS and SYSTEM) passwords.  More importantly, we have the option to create the new database as a CDB or non-CDB database.  Since CDB is our goal here, we check the CDB box and designate the name of our inclusive pluggable database, then proceed.


The summary screen appears to confirm our choices.


If you like, you can watch the progress bar and the progress of the individual steps as the database is being built


Acknowledge the finish screen when the process completes to exit the DBCA


What if we had selected the ‘advanced configuration’ option earlier?  All the options under Typical configuration are automatically disabled.


Now we need to select the type of database (single-instance, RAC, or RAC one-node) and the characteristics (custom, general purpose, or data warehouse).  We’ll select single-instance and general purpose, then proceed.


There are some interesting choices on the next screen.  After providing a global database name and SID, you can choose a CDB or non-CDB database.  If you choose a CDB, you can opt for an empty CDB with no PDBs (except the PDB$SEED) or you can choose to create one or more PDBs at the same time.  Another option that is newly available in 12c release 2 is the option to create a separate UNDO tablespace for each PDB.


Storage options come next


Recovery options on this screen.  I chose to create a flash recovery area with the recommended size and enable archiving.


Listener selection is next.  There are no existing database listeners in the system, so I opt to create one with a default name and port number.  It is recommended to avoid using the default 1521 port for security reasons, but since this is a closed system, there should be no concerns.


At the next screen, I chose not to install Database Vault or Label Security


The next screen is multi-tabbed with options for memory allocation, sizing of processes, selection of character sets,  connection mode, and installation of sample schemas.  I choose the defaults in each section.


I opted to configure Enterprise Manager database express.  We may want to view our CDB options for backup and recovery later in a separate post.


Specify the administrative user credentials at the next screen


This screen prompts you with options to create the database, create a template of your choices for additional CDBs of this type, or the generation of scripts for manual database creation.  I opt for all three.  Why not?


Confirm your choices on the summary screen


If you like, you can watch the progress bar and the progress of the individual steps as the database is being built


Acknowledge the finish screen when the process completes to exit the DBCA


Creating a PDB

If you have a CDB and want to create a PDB or add a PDB, the DBCA is also the preferred method.

To create a new PDB, start DBCA. Once you get past the 12c splash screen, select ‘Manage pluggable databases’ on the first screen, then proceed.


This screen contains all the PDB options you have in your current configuration such as create, delete, unplug, or configure.  The delete, unplug, and configure options will only by available if there is an existing PDB.  Select ‘create a pluggable database’, then proceed.


Designate the CDB that will host your new PDB and give the admin account credentials for the designated CDB, then proceed.


Here you can designate whether to create a new PDB from the ‘seed’ or create a new PDB from an existing PDB.  We are going to use the seed since we don’t have an existing PDB.


Give a name to the new PDB and the credentials of the new administrative account for the PDB.  It is interesting to note here that the wizard is not looking for a typical SYS or SYSTEM account since these accounts already exist at the CDB level.


Storage options are next as well as opting to create a default user tablespace.  The default user tablespace will be called USERS if you choose this option.  it is interesting to note here that even though the option is not given, the new PDB will have its own UNDO tablespace by default.


Confirm your choices on the summary screen, then proceed.


Watch the progress of the PDB creation.


Acknowledge the finish screen when the process completes to exit the DBCA


Other methods to create a PDB include:

  • Using the ‘create pluggable database’ command to create the PDB while connected to the CDB.
  • Plug an existing PDB into the CDB.
  • Using Oracle Cloud Control to guide you through the creation of the PDB.
  • Cloning an existing PDB to a new PDB with, or without data.
  • Creating an empty PDB and using Oracle Data Pump to import data from one database source into a PDB.

Starting and stopping the CDB

Starting and stopping a CDB is just like starting and stopping a non-CDB database.  Use the SQL*Plus utility (or Oracle Enterprise Manager) and issue the startup and shutdown commands as usual.  All of the options still exist such as startup force, shutdown abort and shutdown immediate.

Starting and stopping the PDB

First, determine the state of the PDB before issuing start or stop commands by connecting to the CDB with an account having SYSDBA, SYSOPER, SYSBACKUP or SYSDG privileges:

SQL> select name, open_mode from v$pdbs;

 --------------------       ------------------

The PDB$SEED database is ALWAYS read only.  As the template database, it needs to maintain its integrity.  Since the PDB is already mounted, you just need to open it:

SQL> alter pluggable database JON12PDB1 open;

Pluggable database altered.

You can execute the commands directly without the ‘pluggable database’ syntax by changing to the PDB within the session first:

SQL> alter session set container=JON12PDB1;

Session altered.

SQL> shutdown
 Pluggable Database closed.

I would not recommend this method, however, since it would be easy to forget what container you are pointing to and issue a shutdown mistakenly against the CDB could make you think about updating your resume.

Other methods:

  • Use the ‘alter pluggable database’ command to start and stop the PDB when connected to the CDB with an account having SYSDBA, SYSOPER, SYSBACKUP or SYSDG privileges.
  • Using the SQL*Plus ‘startup pluggable database’ and ‘shutdown pluggable database’ commands when connected to the CDB with an account having SYSDBA, SYSOPER, SYSBACKUP or SYSDG privileges.

PDB startup options (make sure you are connected to the PDB using the ‘alter session set container’ command while in the CDB in the example above):

Open (default) – Open the PDB normally (startup pluggable database JON12PDB1 😉

Force – Force the PDB open (startup pluggable database JON12PDB1 force;)

Upgrade – Open the PDB in upgrade mode (startup pluggable database JON12PDB1 upgrade;)

Restricted – Open the PDB in restricted mode (startup pluggable database JON12PDB1 restricted;)

Read only – Open the PDB in read only mode (startup pluggable database JON12PDB1 open read only;)

For those who don’t want to have to execute separate startup commands for the PDBs after starting the CDB, there is a means by which you can set the default state per PDB after the CDB is started.

For example, after starting the CDB, set the state of your desired PDB, albeit OPEN (read/write) or OPEN READ ONLY, execute the SAVE STATE command from the CDB:

alter pluggable database <PDB_NAME> save state;

Now when the CDB starts, the respective PDB will automatically go to the saved state.  You can view the saved state of each PDB in the DBA_PDB_SAVED_STATES view.  Please note that the saved state feature for PDBs is only on Oracle software versions or higher.

To clear the saved state of the PDB, issue the DISCARD STATE command:

alter pluggable database <PDB_NAME> discard state;

This command will set the state of the particular PDB back to the MOUNTED default.  Changing the saved state of the desired PDB is a simple matter of placing the PDB in the desired state and issuing the SAVE STATE command again.

PDB open options:

Open (default) – Open the PDB in read write mode (alter pluggable database JON12PDB1 open;)

Open read only – Open the PDB in read only mode (alter pluggable database JON12PDB1 open read only;)

Open resetlogs – Open the PDB after a point-in-time-recovery (PITR) (alter pluggable database JON12PDB1 open resetlogs;)

Open restricted – Open the PDB in restricted mode (alter pluggable database JON12PDB1 open restricted;)

Force – Force shutdown and restart (alter pluggable database JON12PDB1 open force;)

Upgrade  – Open in upgrade mode (alter pluggable database JON12PDB1 open upgrade;)

PDB shutdown options (make sure you are connected to the PDB using the ‘alter session set container’ command while in the CDB in the example above):

Normal (default) – Close the PDB normally, waiting for transactions to end (shutdown)

Shutdown Immediate – Close the PDB. Rollback any pending transaction. Flush the SGA (shutdown immediate)

Abort – Shutdown the PDB without delay. Dirty buffers may be written to disk (shutdown abort)

Transactional – Shutdown the PDB after all active transactions have completed (shutdown transactional)

PDB close options:

Close (default) – Close the PDB normally (alter pluggable database JON12PDB1 close;)

Immediate – Force the PDB closed (alter pluggable database JON12PDB1 close immediate;)


Connecting to the CDB/PDB

Accessing the CDB is the same as accessing a non-CDB database and we have already discussed accessing the PDB while connected to the CDB, but what about accessing the PDB directly from outside the CDB?  As I mentioned in my last post, listener services are automatically created for the CDB and each PDB on creation.  Here you can see the JON12CDB and JON12PDB1 services when checking the listener status.


Make sure the CDB is started and the PDB is open.  You can query the v$pdbs view from the CDB to see the status of all the enclosed PDBs.  Once you are sure the PDB is open, you can add an entry to your tnsnames.ora file to access the PDB directly (The entry for the CDB should already be present):

 (ADDRESS = (PROTOCOL = TCP)(HOST = localhost)(PORT = 1521))

You should then be able to access your PDB directly:


Dropping a CDB and PDB

Dropping or removing a CDB is the same as dropping a non-CDB database.  The only difference in the case of dropping a CDB is that all PDB databases beneath it will be removed as well.

Dropping or removing a PDB is a bit more complicated when performing the task manually, but not by much.  When connected to the CDB that contains the PDB, use the ‘alter pluggable database’ command to first close the PDB, then issue the ‘drop pluggable database <PDB_name> including datafiles;’ to complete the process.  Using the DBCA to drop a PDB is even easier as it performs the closing of the PDB automatically, if necessary.

Other PDB removal options:

  • The SQL prompt, using the drop pluggable database command.
  • Using Oracle Enterprise Manager Cloud Control to remove the PDB
  • Unplugging the PDB
  • Deleting the entire CDB.


Querying the CDB and PDB

Once you are connected to the CDB or PDB, queries function the same as they have in non-CDB databases.  What is more important in this stage of our multitenant experience is knowing about multitenant-specific commands and objects.  For instance, when you are connected to a CDB, you can use the command ‘show pdbs’ to get all associated PDBs listed with their current state:

SQL> show pdbs

 ---------- ------------------------------ ---------- ----------

Remember the PDB$SEED PDB is ALWAYS open in read only mode.

The container ID (CON_ID) is an important identifier in the multitenant environment.  The ‘show con_id’ command displays the CON_ID of the current CDB or PDB you are in.  Notice the value changes when I switch from the CDB to the PDB with the ‘alter session set container’ command:

SQL> show con_id

 SQL> alter session set container=jon12pdb1;

Session altered.

SQL> show con_id


NOTE – Once you alter your session to switch from the CDB to a PDB, you cannot switch back to the CDB and must exit your session and open a new one or open a different session to access the CDB once more.

The ‘show con_name’ command is similar to the ‘show con_id’ except that the name and not the ID is displayed:

SQL> show con_name


SQL> alter session set container=jon12pdb1;

Session altered.

SQL> show con_name



If you need a list of the PDB IDs and names, query the v$pdbs view:
SQL> select con_id, name
2  from v$pdbs;

 ---------- ----------
 3 JON12PDB1

In some views, the CON_ID column has been added for multitenant identification as in the case of v$tablespace:

select a.con_id,,
 from v$tablespace a, v$pdbs b
 where a.con_id=b.con_id order by 1;

 ---------- ---------- ----------
 2 TEMP       PDB$SEED
 3 TEMP       JON12PDB1
 3 USERS      JON12PDB1

NOTE – This is a listing from the CDB.  If you are connected to a PDB, you will only see the tablespaces for that particular PDB.

In other views, some filtering has been enabled, such as dba_tables.  In this case, you will see only tables associated with the particular CDB or PDB you are connected to:

SQL> select count(*)
 2  from dba_tables;


SQL> alter session set container=jon12pdb1;

Session altered.

SQL> select count(*)
 2  from dba_tables;


The difference in the counts here is that when you are connected to the PDB, you cannot see the CDB-specific objects.  The CDB_TABLES view allows you to identify tables as belonging to a CDB or PDB:

select a.con_id,, count(*)
 from cdb_tables a, v$containers b
 where a.con_id=b.con_id
 group by a.con_id,;

 ---------- ---------- ----------
 1 CDB$ROOT     2106
 3 JON12PDB1     2105

This view lets you see the CDB and enclosed PDB tables when connected to the CDB.  However, it does not work the same when connected to the PDB:

SQL> alter session set container=jon12pdb1;

Session altered.

SQL> select a.con_id,, count(*)
 from cdb_tables a, v$containers b
 where a.con_id=b.con_id
 group by a.con_id,;
 ---------- ---------- ----------
 3 JON12PDB1     2105


Now that you are on the PDB, the view only lets you see tables associated with that particular PDB.

Here are some other admin views available:

V$CONTAINERS – Contains information on the containers in the database including the ROOT and all PDBs.

V$PDB_INCARNATION – Provides information on incarnations specific to given PDBs.

V$PDBS – Provides information on PDBs within the database

DBA_CONTAINER_DATA – Displays various container data for container data objects.

DBA_HIST_PDB_INSTANCE – Contains workload repository records for PDBs

DBA_PDB_HISTORY – Contains the history of a PDB. Only visible from within a specific PDB.

DBA_PDB_SAVED_STATES – Contains the saved state of a PDB.

DBA_PDBS Contains a list of the PDBs in the database. Only visible from the ROOT container.

PDB_ALERTS Provides information on specific PDB alerts.

PDB_PLUG_IN_VIOLATIONS Provides information on incompatibilities between the CDB and the PDB.


We’ve covered quite a bit of territory in this post.  Next up, cloning and migrating PDBs


Oracle Multitenant – Part 1 – What is multitenant?

The Oracle database multitenant architecture was introduced in version 12c release 1. This architecture consists of a container database (CDB) that contains one or more pluggable databases (PDB). The CDB acts as a metadata repository for the PDBs and a source of user accounts that are utilized across the PDBs called common users. The CDB and PDB share resources, but not data. A PDB template or “seed” database allows the quick deployment of PDBs out of the CDB. This architecture was created with consolidation in mind and according to the Oracle 12c release 1 documentation, will become the default database architecture as the non-CDB architecture is deprecated (

My current work has been in non-CDB environments, but given that non-CDB databases are being deprecated, it is in my interest to start working with CDB databases. I plan to share this work with you in a series of posts for (hopefully) mutual benefit.

The multitenant architecture is a separately licensed option from Oracle, but there is an exception. If the CDB contains a single PDB, then no additional licensing is required.   This way, you can start converting non-CDB databases to CDB without having to accrue additional licensing cost.

The PDBs can be ‘plugged’ and ‘unplugged’ between container databases within limits. Access to the PDBs within a CDB is handled via SQL*Net. Whenever a PDB is created, a service of the same name is also created to handle the PDB connections and the service is automatically registered with the database listener.   It is recommended to change the name of this service for security reasons.

While the CDB and PDB share resources such as memory and processes, each has their own SYS and SYSAUX tablespace with associated objects.   There is one common UNDO tablespace at the CDB level (one per instance for RAC) to be shared, but each PDB has its own TEMP tablespace.

There is a laundry list of features in the Oracle Documentation that this new architecture provides involving consolidation:

  • Cost reduction: By consolidating hardware and sharing database memory and files, you reduce costs for hardware, storage, availability, and labor. For example, 100 PDBs on a single server share one database instance and one set of database files, thereby requiring less hardware and fewer personnel.
  • Easier and more rapid movement of data and code: By design, you can quickly plug a PDB into a CDB, unplug the PDB from the CDB, and then plug this PDB into a different CDB. The implementation technique for plugging and unplugging is similar to the transportable tablespace technique.
  • Easier management and monitoring of the physical database: The CDB administrator can attend to one physical database (one set of files and one set of database instances) rather than split attention among dozens or hundreds of non-CDBs. Backup strategies and disaster recovery are simplified.
  • Separation of data and code: Although consolidated into a single physical database, PDBs mimic the behavior of non-CDBs. For example, if user error loses critical data, a PDB administrator can use Oracle Flashback or point-in-time recovery to retrieve the lost data without affecting other PDBs.
  • Secure separation of administrative duties: A user account is common, which means that it can connect to any container on which it has privileges, or local, which means that it is restricted to a specific PDB. A CDB administrator can use a common user account to manage the CDB. A PDB administrator uses a local account to manage an individual PDB. Because a privilege is contained within the container in which it is granted, a local user on one PDB does not have privileges on other PDBs within the same CDB.
  • Ease of performance tuning: It is easier to collect performance metrics for a single database than for multiple databases. It is easier to size one SGA than 100 SGAs.
  • Support for Oracle Database Resource Manager: In a multitenant environment, one concern is contention for system resources among the PDBs running on the same computer. Another concern is limiting resource usage for more consistent, predictable performance. To address such resource contention, usage, and monitoring issues, you can use Oracle Database Resource Manager.
  • Fewer database patches and upgrades: It is easier to apply a patch to one database than to 100 databases, and to upgrade one database than to upgrade 100 databases.

The previous benefits were from the first release of CDB in Oracle version 12c release 1. Oracle 12c release 2 added the following features, although this is not a complete list:

  • CDBs can now support “thousands” of PDBs instead of the previous maximum of 252.
  • A PDB can now have a different character set than the CDB
  • Point-in-time copies of PDBs possible for development and testing purposes
  • PDB flashback now possible by setting up restore points
  • PDBs can have their own UNDO tablespaces
  • Upgrade a CDB with one or more PDBs in a single operation


Next up in this series – Creating, connecting to, removing, and querying CDBs and PDBs


Creating Corruption for Recovery Practice

Backup and recovery should be the first item on any DBA’s list when administering a new environment.  Often the emphasis is placed on backups and that is only half of creating a successful backup and recovery solution for a production environment.  Making sure recovery is possible from a situation such as block corruption using the backups is the other and arguably the more important half.

As newer versions of Oracle more availability features and higher reliability, block corruption rarely happens. New recovery features, like table recovery in version 12c are also introduced over time.  This is a good thing, but the problem is practicing recovery methods when recovery is not necessary.  The only option most DBA’s have is recovering production database backups into a non-production  environment.  While this may be enough of a backup validation process for most, it leaves the question of knowing how to handle a situation where less than a full recovery of production data is necessary.  Documenting proven recovery steps for full as well as partial recoveries will help during times of crisis when tension is high.

Using a virtual environment, I will demonstrate how to manually corrupt a database so that recovery becomes necessary.  The actual steps to recover the database once it is damaged will not be covered in this post as it has already been documented extensively in both the Oracle documentation and other DBA posts.

My virtual environment will consist of a single-instance non-ASM non-container database version on a Oracle Enterprise Linux 7u3 platform.  I will mention additional steps to perform this process on an ASM or Windows-hosted environment as well.  In this Linux environment, we will use the utility dd to perform the actual corruption.

After ensuring my database was in archivelog mode, I created my test objects:

SQL> create tablespace jontest_tsp datafile '/u01/app/oracle/oradata/JON12C/datafile/jontest_tsp1.dbf'
size 100M autoextend on next 1M extent management local segment space management auto;
Tablespace created.

SQL> create user jontest identified by semperf1 default tablespace jontest_tsp;

User created.

SQL> grant connect, resource to jontest;

Grant succeeded.

SQL> alter user jontest quota unlimited on jontest_tsp;

User altered.

SQL> conn jontest/semperf1

SQL> create table table_dbc as select dbms_random.value(0,100) N1 from dual connect by level <= 1000;

Table created.


SQL> column segment_name format a10;

SQL> select segment_name, tablespace_name from user_segments where 
---------- ------------------------------

Now that we have our test objects in place, it is time to get a backup of our environment.

RMAN> backup database plus archivelog delete input;

Starting backup at 07-JUN-17
current log archived
using target database control file instead of recovery catalog
allocated channel: ORA_DISK_1
channel ORA_DISK_1: SID=17 device type=DISK
channel ORA_DISK_1: starting archived log backup set
channel ORA_DISK_1: specifying archived log(s) in backup set
input archived log thread=1 sequence=4 RECID=1 STAMP=946036509
input archived log thread=1 sequence=5 RECID=2 STAMP=946040464
channel ORA_DISK_1: starting piece 1 at 07-JUN-17
channel ORA_DISK_1: finished piece 1 at 07-JUN-17
piece handle=/u01/app/oracle/fast_recovery_area/JON12C/backupset/2017_06_07/o1_mf_annnn_FULL_20170607_dmjhv16k_.bkp tag=FULL_20170607 comment=NONE
channel ORA_DISK_1: backup set complete, elapsed time: 00:00:01
channel ORA_DISK_1: deleting archived log(s)
archived log file name=/u01/app/oracle/fast_recovery_area/JON12C/archivelog/2017_06_07/o1_mf_1_4_dmjczcsb_.arc RECID=1 STAMP=946036509
archived log file name=/u01/app/oracle/fast_recovery_area/JON12C/archivelog/2017_06_07/o1_mf_1_5_dmjhv02b_.arc RECID=2 STAMP=946040464
Finished backup at 07-JUN-17

Starting backup at 07-JUN-17
using channel ORA_DISK_1
channel ORA_DISK_1: starting full datafile backup set
channel ORA_DISK_1: specifying datafile(s) in backup set
input datafile file number=00001 name=/u01/app/oracle/oradata/JON12C/datafile/o1_mf_system_dmj83mqt_.dbf
input datafile file number=00003 name=/u01/app/oracle/oradata/JON12C/datafile/o1_mf_sysaux_dmj82jkj_.dbf
input datafile file number=00005 name=/u01/app/oracle/oradata/JON12C/datafile/jontest_tsp1.dbf
input datafile file number=00004 name=/u01/app/oracle/oradata/JON12C/datafile/o1_mf_undotbs1_dmj851z7_.dbf
input datafile file number=00006 name=/u01/app/oracle/oradata/JON12C/datafile/o1_mf_users_dmj850wr_.dbf
channel ORA_DISK_1: starting piece 1 at 07-JUN-17
channel ORA_DISK_1: finished piece 1 at 07-JUN-17
piece handle=/u01/app/oracle/fast_recovery_area/JON12C/backupset/2017_06_07/o1_mf_nnndf_TAG20170607T130106_dmjhv391_.bkp tag=TAG20170607T130106 comment=NONE
channel ORA_DISK_1: backup set complete, elapsed time: 00:01:25
channel ORA_DISK_1: starting full datafile backup set
channel ORA_DISK_1: specifying datafile(s) in backup set
including current control file in backup set
including current SPFILE in backup set
channel ORA_DISK_1: starting piece 1 at 07-JUN-17
channel ORA_DISK_1: finished piece 1 at 07-JUN-17
piece handle=/u01/app/oracle/fast_recovery_area/JON12C/backupset/2017_06_07/o1_mf_ncsnf_TAG20170607T130106_dmjhxrz3_.bkp tag=TAG20170607T130106 comment=NONE
channel ORA_DISK_1: backup set complete, elapsed time: 00:00:02
Finished backup at 07-JUN-17

Starting backup at 07-JUN-17
current log archived
using channel ORA_DISK_1
channel ORA_DISK_1: starting archived log backup set
channel ORA_DISK_1: specifying archived log(s) in backup set
input archived log thread=1 sequence=6 RECID=3 STAMP=946040554
channel ORA_DISK_1: starting piece 1 at 07-JUN-17
channel ORA_DISK_1: finished piece 1 at 07-JUN-17
piece handle=/u01/app/oracle/fast_recovery_area/JON12C/backupset/2017_06_07/o1_mf_annnn_FULL_20170607_dmjhxt9k_.bkp tag=FULL_20170607 comment=NONE
channel ORA_DISK_1: backup set complete, elapsed time: 00:00:01
channel ORA_DISK_1: deleting archived log(s)
archived log file name=/u01/app/oracle/fast_recovery_area/JON12C/archivelog/2017_06_07/o1_mf_1_6_dmjhxt47_.arc RECID=3 STAMP=946040554
Finished backup at 07-JUN-17


If your particular environment is using ASM for datafile storage instead of the filesystem like my environment, follow these steps to copy the datafile to the OS where you can apply the corruption process.  Make note of the backup tag identifier.

alter database datafile '+DATA/RAC/DATAFILE/jontest_tsp1.df' offline;

rman target=/

RMAN> BACKUP AS COPY DATAFILE '+DATA/RAC/DATAFILE/jontest_tsp1.df' FORMAT '/tmp/jontest_tsp_%f';

After the datafile is out of ASM or the datafile is in a filesystem, check the header block as a SYSDBA user:

SQL> select header_file, header_block from dba_segments where segment_name='TABLE_DBC';

----------- ------------
   5      130

The header of segment (table) is block 130, so if we damage the next block with dd, we will cause a datafile/table corruption.

[oracle@localhost ~]$ dd of=/u01/app/oracle/oradata/JON12C/datafile/jontest_tsp1.dbf bs=8192 conv=notrunc seek=131 <<EOF
> #corrupt_this#
0+1 records in
0+1 records out
15 bytes (15 B) copied, 8.3023e-05 s, 181 kB/s

Execute DBVERIFY to confirm that we have generated some corruption:

[oracle@localhost ~]$ dbv file=/u01/app/oracle/oradata/JON12C/datafile/jontest_tsp1.dbf blocksize=8192

DBVERIFY: Release - Production on Wed Jun 7 14:08:25 2017

Copyright (c) 1982, 2014, Oracle and/or its affiliates.  All rights reserved.

DBVERIFY - Verification starting : FILE = /u01/app/oracle/oradata/JON12C/datafile/jontest_tsp1.dbf
Page 131 is marked corrupt
Corrupt block relative dba: 0x01400083 (file 5, block 131)
Bad header found during dbv: 
Data in bad block:
 type: 35 format: 3 rdba: 0x74707572
 last change scn: 0x2373.6968745f seq: 0xa flg: 0x04
 spare1: 0x6f spare2: 0x72 spare3: 0x0
 consistency value in tail: 0x8ca90602
 check value in block header: 0xb1ea
 computed block checksum: 0x176

As an interesting side note, even though DBVERIFY noticed the corruption, if we queried v$database_block_corruption, we would not see any entries:

SQL> select * from v$database_block_corruption;

no rows selected

If we immediately query the affected table, the database will see the corruption:

SQL> select count(*)
  2  from table_dbc;
select count(*)
ERROR at line 1:
ORA-01578: ORACLE data block corrupted (file # 5, block # 131)
ORA-01110: data file 5: '/u01/app/oracle/oradata/JON12C/datafile/jontest_tsp1.dbf'

The v$database_block_corruption view now has the appropriate entries

SQL> select * from v$database_block_corruption;

---------- ---------- ---------- ------------------ --------- ----------
  5   131        1    0 CORRUPT        0

RMAN will also populate this view if it encounters corrupt blocks.  This is why it is a good practice to either review RMAN backup logfiles or query the v$database_block_corruption view at the end of a backup.

If this was an ASM datafile, restore it to the original location:

rman target /


After restore and recover the datafile must be brought online

alter database datafile '+DATA/RAC/DATAFILE/jontest_tsp1.df' online;

Now you can practice a datafile, table, or block recovery.

If we wanted to corrupt a datafile in a Windows environment, we could use an editor such as UltraEdit GUI editor, which is also available in a Linux version.  Using this editor, we can either modify the header of the file to corrupt the datafile, or scroll to the table data to corrupt just the table to practice table or block recovery.

I know this goes without saying, but just like the warning labels on hair dryers that tell you not to use them in the bathtub, YOU SHOULD NOT PERFORM THIS PROCESS AGAINST A PRODUCTION DATABASE, EVEN IF YOU CREATE A SEPARATE TABLESPACE THAT CONTAINS NO PRODUCTION DATA.  You know the reason for most warning labels is that someone probably did the very thing the label warns against doing.


ORA-16957: SQL Analyze time limit interrupt


First thing this morning, I was greeted with an email from my alert log monitor containing the error in the title.  This is one of the less cryptic error messages from an Oracle database, but what does it mean?

If you are an Oracle DBA and have been dealing with performance issues for any length of time, you are familiar with the SQL Tuning Advisor.  This tool is part of the Oracle Tuning Pack .  It takes one or more SQL statements as input, analyzes them, and displays recommendations to tune the statements with rationale and expected benefits.  These recommendations can include collecting object statistics, creating indexes, creating SQL profiles or SQL plan baselines, and even recommendations to rewrite.

The SQL Tuning Advisor has two modes – automatic and manual.  The automatic mode is referred to as the Automatic SQL Tuning Task or SQL Analyze for short.  This task when enabled runs nightly as part of the default maintenance plan.  Usually it runs happily in the background and goes mostly unnoticed.  Other times it generates an ORA-16957 error. This error means that it has not completed within its limit settings. You can view these settings like this:

COLUMN parameter_value FORMAT A30  

SELECT parameter_name, parameter_value  
FROM dba_advisor_parameters  


------------------------------ ------------------------------  

LOCAL_TIME_LIMIT               1200  
TIME_LIMIT                     7200  
SQL_LIMIT                      -1

The time_limit value is the total time in seconds that the SQL Analyze job is allowed per execution within the maintenance window.  The local_time_limit is the value in seconds that the SQL Analyze job is allowed to spend analyzing each individual SQL statement.  The sql_limit parameter is the number of SQL statements to examine per execution of the job.  The -1 number indicates no limit to the number of SQL statements the job can review within its time limit.

How does it determine which statements need to be analyzed?  From the Oracle documentation, the following quote is taken:

The automated SQL tuning task does not process the following types of SQL:

  • Ad hoc SQL statements or SQL statements that do not repeat within a week
  • Parallel queries
  • Queries that take too long to run after being SQL profiled, so that it is not practical for SQL Tuning Advisor to test execution
  • Recursive SQL

 You can run SQL Tuning Advisor on demand to tune the preceding types of SQL statements.


Can we find out what SQL statement caused the ORA-16957 error?

Get the execution_name for the long run day:

set lines 200 pages 100 
col error_message for a60 
col execution_name for a15 

select execution_name, advisor_name,to_char(execution_start,'dd-mon-yy hh:mi:ss'), to_char(execution_end,'dd-mon-yy hh:mi:ss'), status,error_message 
from dba_advisor_executions 
where task_name = 'SYS_AUTO_SQL_TUNING_TASK' 
order by execution_start desc;

Take the name of the execution you are interested in and place it into the query below to identify the ID of the SQL causing the overrun.  If a sufficient amount of time has passed, you may not get any data from this query.

SELECT sql_id, sql_text FROM dba_hist_sqltext 
WHERE sql_id IN (SELECT attr1 FROM dba_advisor_objects 
WHERE execution_name = '&execution_name' 
AND type = 'SQL' AND bitand(attr7,64) <> 0 );

 Then you can take the SQL_ID from the query above and run it through the manual mode of the SQL Tuning Advisor:

Create the tuning task:

l_sql_tune_task_id VARCHAR2(100); 
l_sql_tune_task_id := DBMS_SQLTUNE.create_tuning_task ( 
sql_id => '&SQL_ID', 
scope => DBMS_SQLTUNE.scope_comprehensive, 
time_limit => 99999,   --I use a large time out value 
task_name => 'my_tuning_task', 
description => 'Tuning task for statement blahblah.'); 
DBMS_OUTPUT.put_line('l_sql_tune_task_id: ' || l_sql_tune_task_id); 


Execute the tuning task:

DBMS_SQLTUNE.EXECUTE_TUNING_TASK( task_name => 'my_tuning_task'); 

When the task is complete, get the report:

SET LONG 5000 

Be sure to drop the tuning task before executing another.  NOTE – If the tuning task results recommend using a SQL Profile, then you MUST accept the profile before dropping the tuning task.  Otherwise, the recommended profile will be lost and you will need to run the tuning task again.

  dbms_sqltune.drop_tuning_task (task_name => 'my_tuning_task');


In the event that the ORA-16957 error is seen on a regular basis, you may need to adjust the SQL Analyze task limits or limit the number of SQL statements analyzed. Use the DBMS_SQL_AUTOTUNE.SET_TUNING_TASK_PARAMETER procedure to change these values.  Here is an example of changing the time limit to 7200 seconds (2 hours):

parameter => 'TIME_LIMIT', value => 7200); 


If you need to determine if the SQL Analyze job is running against your database, use this query:

 WHERE  CLIENT_NAME = 'sql tuning advisor';
 -------------------- --------
 sql tuning advisor   ENABLED


You can disable or enable the SQL Analyze job with these commands:

     client_name => 'sql tuning advisor'
 ,   operation   => NULL
 ,   window_name => NULL

     client_name => 'sql tuning advisor'
 ,   operation   => NULL
 ,   window_name => NULL

If you are interested in viewing the latest SQL Analyze report:


:l_report := DBMS_SQLTUNE.report_auto_tuning_task(
begin_exec => NULL,
end_exec => NULL,
type => DBMS_SQLTUNE.type_text,
level => DBMS_SQLTUNE.level_typical,
section => DBMS_SQLTUNE.section_all,
object_id => NULL,
result_limit => NULL

SET LONG 1000000
PRINT :l_report

If this task is enabled in your database, then take advantage of its analysis.  Regular viewing of its output may help identify ‘usual suspects’ that continue to run poorly.


Blast from the past – Rollback segments

Those of us that have worked with Oracle databases over the years have seen new features come along that relieved some of the more challenging aspects of database administration (i.e. things that drove us crazy).  One of the new features we were happy to see was automatic undo management in Oracle 9i release 1.  This new feature made the challenge of managing rollback segments obsolete. In the days before automatic undo management, the DBA not only had to create the rollback tablespace, but also size and manage the rollback segments it contained.

For newer DBA’s, these terms may not be familiar, so a little background may be in order.  Before the undo tablespace, there was the rollback tablespace.  This was where Oracle maintained changed, but uncommitted data for read consistency, much like the undo tablespace.  The rollback tablespace was comprised of rollback segments which needed to be manually created by the DBA.  The number and size of these segments was at the discretion of the DBA.  If the number or size of the rollback segments was insufficient for the amount of database activity, then contention for the rollback segment data blocks was a real, and common, occurrence.

Here is a typical database creation script from the Oracle 8i release 3 (8.1.7) documentation:

    LOGFILE '/u01/oracle/rbdb1/redo01.log' SIZE 1M REUSE,
            '/u01/oracle/rbdb1/redo02.log' SIZE 1M REUSE,
            '/u01/oracle/rbdb1/redo03.log' SIZE 1M REUSE,
            '/u01/oracle/rbdb1/redo04.log' SIZE 1M REUSE
    DATAFILE '/u01/oracle/rbdb1/system01.dbf' SIZE 10M REUSE 
      NEXT 10M MAXSIZE 200M 


-- Alter temporary system tablespace online before proceding

-- Create additional tablespaces ...
-- RBS: For rollback segments
-- USERs: Create user sets this as the default tablespace
-- TEMP: Create user sets this as the temporary tablespace
    DATAFILE '/u01/oracle/rbdb1/rbs01.dbf' SIZE 5M REUSE AUTOEXTEND ON
      NEXT 5M MAXSIZE 150M;
    DATAFILE '/u01/oracle/rbdb1/users01.dbf' SIZE 3M REUSE AUTOEXTEND ON
      NEXT 5M MAXSIZE 150M;
    DATAFILE '/u01/oracle/rbdb1/temp01.dbf' SIZE 2M REUSE AUTOEXTEND ON
      NEXT 5M MAXSIZE 150M;

-- Create rollback segments.  
  tablespace rbs;
  tablespace rbs;
  tablespace rbs;
  tablespace rbs;

-- Bring new rollback segments online and drop the temporary system one


As you can see, you had to create and online one system rollback segment in the system tablespace before you could create the other tablespaces including rollback with its inclusive rollback segments.  You then had to bring the rollback segments you created online.  The temporary rollback segment was taken offline and removed at the end for housekeeping.

Rollback segments still exist in the newer versions of the Oracle database, but they are automatically created and managed by internal processes.  Additional segments are automatically created as needed.

Here is the output from a 9i release 1 database.  The sys-owned rollback segment takes the place of the rb_temp rollback segment in the 8i database creation script present during the creation of the database.  The difference here is that it is created automatically as part of the database creation process.  The remaining rollback segments were created automatically when the undo tablespace was created.

SQL> select * from v$version;

Oracle9i Enterprise Edition Release - Production
PL/SQL Release - Production
CORE       Production
TNS for 32-bit Windows: Version - Production
NLSRTL Version - Production

SQL> select segment_name, owner from dba_rollback_segs;

SEGMENT_NAME                   OWNER
------------------------------ ------
SYSTEM                         SYS
_SYSSMU1$                      PUBLIC
_SYSSMU2$                      PUBLIC
_SYSSMU3$                      PUBLIC
_SYSSMU4$                      PUBLIC
_SYSSMU5$                      PUBLIC
_SYSSMU6$                      PUBLIC
_SYSSMU7$                      PUBLIC
_SYSSMU8$                      PUBLIC
_SYSSMU9$                      PUBLIC
_SYSSMU10$                     PUBLIC

Except for the segment naming convention, things look very similar in a version database

SQL> select * from v$version;

Oracle Database 11g Enterprise Edition Release - 64bit Production
PL/SQL Release - Production
CORE      Production
TNS for Linux: Version - Production
NLSRTL Version - Production

SQL> select segment_name, owner from dba_rollback_segs;

SEGMENT_NAME                   OWNER
------------------------------ ------
SYSTEM                         SYS
_SYSSMU10_1197734989$          PUBLIC
_SYSSMU9_1650507775$           PUBLIC
_SYSSMU8_517538920$            PUBLIC
_SYSSMU7_2070203016$           PUBLIC
_SYSSMU6_1263032392$           PUBLIC
_SYSSMU5_898567397$            PUBLIC
_SYSSMU4_1254879796$           PUBLIC
_SYSSMU3_1723003836$           PUBLIC
_SYSSMU2_2996391332$           PUBLIC
_SYSSMU1_3724004606$           PUBLIC

When automatic undo management was first introduced, you could choose between MANUAL or AUTO.  The manual setting was the default and I suspect this had to do with not wanting to force this new feature on anyone.  It wasn’t until 11g release 1 that the AUTO setting became the default for this parameter.  These days, I don’t know of anyone who still uses manual redo management, although it is still available as of 12c release 2.   Although there are occasional ORA-01555 (snapshot too old) errors encountered with the use of auto undo management, this is one feature that I think was worth the wait.