VirtualBox on Fedora 15

This is what I did to install Oracle VM VirtualBox on a new Fedora 15 installation.


The documentation at the bottom of download page VirtualBox for Linux gives you the repos-file for yum on Fedora. Find the repos-file and save it to /etc/yum.repos.d/virtualbox.repo, it should look like this:

[virtualbox]
name=Fedora $releasever - $basearch - VirtualBox
baseurl=http://download.virtualbox.org/virtualbox/rpm/fedora/$releasever/$basearch
enabled=1
gpgcheck=1
gpgkey=http://download.virtualbox.org/virtualbox/debian/oracle_vbox.asc

Then you can search for VirtualBox:

[root@favela ~]# yum search VirtualBox
Loaded plugins: langpacks, presto, refresh-packagekit
=========================== N/S Matched: VirtualBox ============================
VirtualBox-4.0.x86_64 : Oracle VM VirtualBox

  Name and summary matches only, use "search all" for everything.

To install:

yum install VirtualBox-4.0.x86_64

Users need to be included in the OS group vboxusers before they can run it:

usermod -a -G vboxusers oisene

VirtualBox is now ready to run from menu.

In fact this was not what I did first time, because I didn't read the fine manual to the end. I download it from here and installed the missing packages something like this:

yum install qt qt-x11 SDL
yum install kernel-headers-$(uname -r) kernel-devel-$(uname -r)
yum install binutils gcc patch glibc-headers glibc-devel
yum install dkms
rpm -ip /home/oisene/Downloads/VirtualBox-4.0-4.0.8_71778_fedora15-1.x86_64.rpm 

Without the dkms package VirtualBox will run, but it is recommended to ensure VirtualBox host kernel modules will be update during next kernel update.

Today

I have made a few changes to my blog... think I want to blog again.

So far this year in a few words; new job at Keystep - smart move, if you try two years in the wrong company it is time to move on; met quite a few smart oracle folks at the yearly conference of Oracle User Group Norway - got elected as a board member; met more smart guys at Miracle Open World in Denmark the week after. And been thinking about moving back to Brazil quite a lot lately.

Removing orphan DataPump jobs and filtering included tables with a query

Say you define a DP job from the API (DBMS_DATAPUMP) you may end up with jobs with status NOT RUNNING until you get it right. Verify if you have such a job with

select job_name,state 
from user_datapump_jobs;

Then you may try to remove it with:

declare 
 l_h number;
begin
  l_h:=dbms_datapump.attach('YOUR_JOB');
  dbms_datapump.stop_job(l_h,immediate=>1);
  commit;
end;
/

If the query above still reports the job with the same status it can be removed by dropping the master table, according to Note 336014.1:

drop table YOUR_JOB;

If the database is using a recycle bin the table has been renamed to a name starting with 'BIN'; if you rerun the query above the job is renamed to reflect this, but if you purge the recycle bin the job will be finally removed:

purge recyclebin;

If your PL/SQL code fails during definitions before you do a detach you cannot do a second attach. Since you have no reference to the job the only solution I'm aware of is to reconnect (i.e. create another database session). Better yet is to define an exception on ORA-39001 and retrieve the error message with dbms_datapump.get_status, and dbms_datapump.detach(l_handle) in the exception handler:

declare
  invalid_arg exception;
  pragma exception_init(invalid_arg,-39001);
:
:
begin
:
:
exception when invalid_arg then
    dbms_datapump.detach(l_h);
end;

If you drop the master table that is in the defining state, you will not be able to attach to the job, but if you try to attach to such a job the job will be removed for the user_datapump_jobs table.

If you try to add a metadata filter of type NAME_EXPR and you add a subquery in the expression, make sure that any table name in the subquery is prefixed with table owner or else you may receive ORA-00942 in the logfile:

dbms_datapump.metadata_filter(handle => l_h, name => 'NAME_EXPR', value =>'IN (SELECT TABLE_NAME FROM SCOTT.CHANGE_TABLE)'  ); 

Typical error in logfile when schema is missing:

Estimate in progress using BLOCKS method...
Processing object type TABLE_EXPORT/TABLE/TABLE_DATA
ORA-39125: Worker unexpected fatal error in KUPW$WORKER.GET_TABLE_DATA_OBJECTS while calling DBMS_METADATA.FETCH_XML_CLOB []
ORA-00942: table or view does not exist
ORA-06512: at "SYS.DBMS_SYS_ERROR", line 95
ORA-06512: at "SYS.KUPW$WORKER", line 6307
----- PL/SQL Call Stack -----
  object      line  object
  handle    number  name
0x7bfb80b0     15032  package body SYS.KUPW$WORKER
:
:

Slow performance with TDE and CHAR data type on 10.2.0.5

In version 10.2.0.5 of the database there is a bug when using Transparent Data Encryption (TDE) on a column with the CHAR datatype. In the following test a table is created where the primary key is defined as CHAR(11) and then encrypted with NO SALT. A simple lookup on this PK works as expected by using the respective index on 10.2.0.4. But on 10.2.0.5 a trace on event 10053 shows that an INTERNAL_FUNCTION is wrapped around the PK-column and therefore impedes use of the index.

The following test was performed on an OVM guest created from a template downloaded from edelivery.oracle.com. It contains Oracle Server EE 10.2.0.4 running on Oracle Enterprise Linux 5, 64-bit. Configuration of TDE and wallet is not shown.

create table foo (
   id char(11) encrypt no salt primary key ,
   msg varchar2(1000)
  );

insert into  foo ( select * from (
  with generator as (
      select      rownum      rn
      from        dual
      connect by
                  rownum <= 100
  )
  select
      dbms_random.string('x',11) id,
      dbms_random.string('x',1000) msg
  from
      generator   g1,
      generator   g2
  where
    rownum <= 10000
  )
);
commit;
exec dbms_stats.gather_table_stats(user,'FOO', cascade=>true,method_opt=>'for all indexed columns size 1',estimate_percent=>100)
variable lid char(11)
exec :lid:='abc123def45';
alter session set events='10053 trace name context forever, level 1';
select msg from foo where id=:lid;

The query returns immediately since the index is used and the trace file shows the good plan (a simple explain plan will show the same):

--------------------------------------------------+-----------------------------------+
| Id  | Operation                    | Name       | Rows  | Bytes | Cost  | Time      |
--------------------------------------------------+-----------------------------------+
| 0   | SELECT STATEMENT             |            |       |       |     2 |           |
| 1   |  TABLE ACCESS BY INDEX ROWID | FOO        |     1 |   514 |     2 |  00:00:01 |
| 2   |   INDEX UNIQUE SCAN          | SYS_C007236|     1 |       |     1 |  00:00:01 |
--------------------------------------------------+-----------------------------------+
Predicate Information:
----------------------
2 - access("ID"=:LID)

Now, the database is patched to version 10.2.0.5 (patch 8202632). There are no invalid objects in the database after the upgrade. Then repeating the last query from above together with the trace on event 10053, the query takes a long time and the trace file shows this plan:

-------------------------------------+-----------------------------------+
| Id  | Operation          | Name    | Rows  | Bytes | Cost  | Time      |
-------------------------------------+-----------------------------------+
| 0   | SELECT STATEMENT   |         |       |       |   388 |           |
| 1   |  TABLE ACCESS FULL | FOO     |   100 |   50K |   388 |  00:00:05 |
-------------------------------------+-----------------------------------+
Predicate Information:
----------------------
1 - filter(INTERNAL_FUNCTION("ID")=:LID)

The INTERNAL_FUNCTION is used among other situations when oracle has to perform an implicit data conversion. This may happen when the data type of a bind variable is different from the data type used in the table.

This kind of mismatch may happen during upgrades of JDBC drivers where the mapping may change; Java had support for TimeStamp before it was introduced in the Oracle database. Later when Oracle supported TimeStamp this mapping changed, leading to this kind of errors. There are other reasons for such mismatches; a trace on event 10053 will reveal the data type of the bind variable:

*******************************************
Peeked values of the binds in SQL statement
*******************************************
kkscoacd
 Bind#0
  oacdty=96 mxl=32(11) mxlc=00 mal=00 scl=00 pre=00
  oacflg=03 fl2=1000000 frm=01 csi=31 siz=32 off=0
  kxsbbbfp=2b19acc082a0  bln=32  avl=11  flg=05
  value="abc123def45"

The part "oacdty=96" shows the type is CHAR or NCHAR. This did not change between the tests, so there has to be another reason why the INTERNAL_FUNCTION is used. Also in this case the only change is the upgrade, even the parameter compatible remained the same (10.2.0.3). The tests where performed from sqlplus on the database server.

This is bug 10296606 on MOS.

Update: It was discovered that if you use a variable of VARCHAR type the INTERNAL_FUNCTION is not applied, meaning that columns of CHAR type are converted to VARCHAR after encryption, but that USER_TAB_COLUMNS continues to report CHAR for the column.

A few notes on Oracle VM

This is an old post I kept as draft 8 months before I published it...

I've been playing around with Oracle Virtual Manager (OVM) version 2.2 lately. This is collection of a few notes, not all directly related to OVM. The project started with building the server from parts I hoped would play along (they do). The CPU is an Intel i7-930, 12GB of RAM and two Western Digital 1.5TB disks (WD15EARS-00Z5B1). Current version of OVM (2.2) uses update 3 of OEL 5. The kernel lags behind a bit and support for certain wirless card is not included. The AR5008 Wireless chip from Atheros (a Dlink card) was not supported; in order to avoid more cables in the house I created a bridge between the server and another PC nearby.

The installation of OVM went smooth, but the disks are a disgrace with a horrible I/O performance until you do the partitioning right (each partition should start on a sector divisible by 4096 bytes). Meaning I had to choose a custom layout of the partitions and file systems during installation.

The OVM manager was installed in a guest based on a OVM template downloaded from Oracle.

Configuration of OVM Manager fails because host names cannot be resolved:

The server resides on a private network (192.168.x.x) without any DNS; if you get error messages indicating connection problems to the OVM server then make sure the hostname is defined in /etc/hosts, also make sure the hostname is not resolved to localhost (127.0.0.1); which was the case on a fresh installation of OVM Server. This should be done both on the OVM Server and in the guest hosting OVM Manager. (If the initial configuration of OVM Manager fails, connect with ssh and fix /etc/hosts before restarting).

Connecting to the Manager database from afar:
The listener binds to standard port 1521 on localhost. In case you want to connect to the database over SQL*Net you need to change listener.ora and replace localhost with a public IP address/name of the Manager's host. Then you need to add a line to /etc/sysconfig/iptables allowing remote access to the listener:

-A RH-Firewall-1-INPUT -m state --state NEW -m tcp -p tcp --dport 1521 -j ACCEPT

Configuring yum:
I added the following lines to the file /etc/yum.repos.d/public-yum-el5.repo

[el5_u3_base]
name=Enterprise Linux $releasever U3 - $basearch - base
baseurl=http://public-yum.oracle.com/repo/EnterpriseLinux/EL5/3/base/$basearch/
gpgkey=http://public-yum.oracle.com/RPM-GPG-KEY-oracle-el5
gpgcheck=1
enabled=1

[ovm22_2.2.1_base]
name=OracleVM 2 - 2.2.1 - $basearch - base
baseurl=http://public-yum.oracle.com/repo/OracleVM/OVM2/2.2.1/base/$basearch/
gpgkey=http://public-yum.oracle.com/RPM-GPG-KEY-oracle-el5
gpgcheck=1
enabled=1

Then gcc and others can be installed with

yum install gcc

Tuning network
After the disk performance was improved I started to get annoyed about the slow refresh of VNC. The following command reported an increasing queue length when I interacted with a GUI through VNC:

netstat -o -t -c

This Guide was helpful.

Using physical devices for storage
I could not find any way to add a physical device to a guest in the Manager. From the documentation though it is clear how one goes about to do this. Disadvantage is that the added physical device does not show up in Manager after. Anyway, the performance gain outweighs that. I had one disk that I gave entirely to a volume group and then created a logical device with pvcreate, vgcreate and lvcreate.

To add the logical volume as a physical volume to the guest:

xm block-attach 24 phy:/dev/vg00/oradb1_asm1 xvdc 'w'

24 is the id of the guest as reported by xm list, xvdc is the device name at the guest (/dev/xvdc), and 'w' means read/write. At the guest this disk device needs to partitioned with fdisk or similar as with any other physical device.

To make the change permanent the following line (in bold) was added to the disk configuration in vm.cfg:

disk = ['file:/var/ovs/mount/3FEB9655D88E4177B8571AFCF81C5B8A/running_pool/50_oel2/System.img,xvda,w',
'phy:/dev/vg00/oel2_1,xvdb,w',
'file:/var/ovs/mount/3FEB9655D88E4177B8571AFCF81C5B8A/sharedDisk/software.img,xvdc,w!',
]

Monitoring with sar -d showed that the service time substantially lower (1/10) with the physical device, even when the physical device had almost max utilization on writing.

VNC and maximum resolution
Any changes made to the screen resolution was quietly ignored and the connection to the vnc server that is started from dom0 was unreliable. After a while I gave up and started a new vnc server from inside the guest.


Update: Using OVM for my own lab was not optimal, I installed Fedora 15 which among other things supports the wireless card without any hassle. I plan to install Virtual Box on top of it for my lab needs.

Permission denied when adding ASM disk

This is a simple one, a reminder till next time I forget this, since stuff encountered after a Google-search led me astray...

Just in case you try to add disks to an existing ASM disk group using syntax like this:

alter diskgroup data
add disk '/dev/xvdd1'
add disk '/dev/xvde1';

You may get error messages like these:

ORA-15032: not all alterations performed
ORA-15031: disk specification '/dev/xvde1' matches no disks
ORA-15025: could not open disk '/dev/xvde1'
ORA-15056: additional error message
Linux-x86_64 Error: 13: Permission denied
Additional information: 42
Additional information: -1614995373
Additional information: 1730318408
ORA-15031: disk specification '/dev/xvdd1' matches no disks
ORA-15025: could not open disk '/dev/xvdd1'
ORA-15056: additional error message
Linux-x86_64 Error: 13: Permission denied
Additional information: 42
Additional information: 87328232
Additional information: 8192

Solution: Use the ASM disk names as given when creating them with oracleasm:

alter diskgroup data
add disk 'ORCL:VOL3'
add disk 'ORCL:VOL4';

Here VOL3 and VOL4 are the names given when creating the ASM-disk with oracleasm:

oracleasm createdisk VOL3 /dev/xvdd1
oracleasm createdisk VOL4 /dev/xvde1

The initialization parameter asm_diskstring is empty as is default.

The error messages from OS (Linux in this case) may lead you to think that there is something wrong with the owner or attributes of the devices you've just created, even they have the same permissions as the devices previously allocated to ASM... Then when you against reason set the owner and group to be like 'oracle:dba' the error message from Linux does not show up, but you will still get error messages like this:

ORA-15032: not all alterations performed
ORA-15031: disk specification '/dev/xvde1' matches no disks
ORA-15014: location '/dev/xvde1' is not in the discovery set
ORA-15031: disk specification '/dev/xvdd1' matches no disks
ORA-15014: location '/dev/xvdd1' is not in the discovery set

A note on MOS with id 603210.1 is one example of this confusion, it contains this command indicating that it is OK to use the physical device name (which it may be in certain environments for what I know):

create diskgroup DATADG external redundancy disk '/dev/sda14'; 

I think this is one of several reasons why certain un*x guys used to old and mature volume managers and file systems do not easily fall in love with ASM... Once one has passed these initial obstacles I find ASM OK to work with, even for non-RAC databases. It usually shortens the path from getting a LUN allocated on the SAN or whatever until more free space is available to the tablespace in the database.

[This testing was done on a Oracle VM guest, created from an OVM template downloaded from Oracle. It comes with OEL 5 64-bit and version 10.2.0.4 of the database. Oracle calls this OS for JeOS (Just enough OS), but for some reason it does not come with the 'man' command, so for me it is JneOS.]

Error when using 11g export client on a 10g database

The only reason to read this post is if you have googled for the error "ORA-00904: "POLTYP": invalid identifier". This error occurs if you try the old export command from an 11g client against a database on version 10g or lower. The export command runs a query against a table called EXU9RLS in the SYS schema. On 11g this table was expanded with the column POLTYP and the export command (exp) expects to find this column. This should not be much of a problem since Data Pump export can be used.