Skip navigation.

DBA Blogs

Deduping 100 Gigs Worth of Files? Gimme 5 Minutes…

Pythian Group - Mon, 2014-01-13 08:57

Shortly before the Holidays, I became aware of the Dallas/Fort Worth Perl Mongers Winter Hackaton Contest. The challenge was simple enough: Here’s a directory structure filled with 100G worth of files, go and find all the duplicate files in there, as fast as you can.

You know me: can’t resist a contest. But since I knew I didn’t have heaps of time to sink into it, and because — let’s face it — the Perl community is despairingly brim-full of smart people who always beat me to the finish line, I decided to compete on slightly different terms. My main objective would be to leverage a maximum from modern Perl tools so that effort to get a working solution would be minimal. As for the performance of the solution, I decided that I’d be satisfied, and that the case for modernity and laziness would be made if it was at least in the ballpark of the other entries.

And so it began…

The Application Interface

Writing code to interact with the users is booooooring.

The hard part, the fun part, the challenging part, is to come up with the function do_it( $foo, $bar ). Once this is done, all that remains to do is to turn the handle and give a way to the user to set $foo and $bar. And to validate the incoming values for $foo and $bar. And document what are $foo and $bar. Blergh. It’s dotting the i’s, and crossing the t’s. It’s writing the police report after that blood-pumping drug cartel take-down.

Fortunately, there are a few modules that will take that tedium off yours hands. One of my favorites is MooseX-App, and for this project I used its single-command variation, MooseX::App::Simple. Its use is pretty straightforward. The main module of the application is a Moose class

package Deduper;

use strict;
use warnings;

use MooseX::App::Simple;

Attributes of the class that you want to be accessible as options of the script are defined using the option keyword. Same story for positional parameters, with the parameter keyword. Other attributes stays untouched, and won’t be visible to the user.

parameter root_dir => (
    is => 'ro',
    required => 1,
    documentation => 'path to dedupe',

option hash_size => (
    isa => 'Int',
    is => 'ro',
    default => '1024',
    trigger => sub {
        my $self = shift;
        Deduper::File->hash_size( $self->hash_size );
    documentation => 'size of the file hash',

option stats => (
    isa => 'Bool',
    is => 'ro',
    documentation => 'report statistics',

option max_files => (
    traits => [ 'Counter' ],
    isa => 'Int',
    is => 'ro',
    predicate => 'has_max_files',
    default => 0,
    documentation => 'max number of files to scan (for testing)',
    handles => {
        dec_files_to_scan => 'dec',

has start_time => (
    is => 'ro',
    isa => 'Int',
    default => sub { 0 + time },

Last required touch for the class: a run() method that will be invoked when the app is run:

sub run {
    my $self = shift;

    say join "\t", map { $_->path } @$_ for $self->all_dupes;

With that, your app is in working condition. Command-line parsing, argument validation, --help text, it’s all taken care of for you:

$ cat ./

use Deduper; 

$ ./
Required parameter 'root_dir' missing
usage: [long options...] --help

    root_dir  path to dedupe [Required]

    --hash_size           size of the file hash [Default:"1024"; Integer]
    --help -h --usage -?  Prints this usage information. [Flag]
    --max_files           max number of files to scan (for testing) [Default:
                          "0"; Integer]
    --stats               report statistics [Flag]

Traversing Directory Structures

Traversing directory structures isn’t terribly hard. But there are a few things like symlinks that you have to watch for. What we care about, really, is to have each file of the structure handed to us on a silver platter. Let’s have somebody else deal with the underlying menial work.

Here, this somebody else is Path::Iterator::Rule. With it, visiting all files of the directory structure boils down to creating an attribute

has file_iterator => (
    is => 'ro',
    lazy => 1,
    default => sub {
        my $self = shift;
        return Path::Iterator::Rule->new->file->iter_fast( 
            $self->root_dir, {
                follow_symlinks => 0,
                sorted          => 0,

and, well, using it

while( my $file = $self->file_iterator->() ) {
    # do stuff with $file...
A Couple Of No-Brainish Optimizations

Everybody love optimizations that require no more effort than to press a ‘turbo’ button. On this project, two modules providing that kind of boost were just begging to be used.

The first is MooseX::XSAccessor, which invests the accessors of the class with super-speedy XS powers. Alas, that module does not speed up lazy-defined attributes, which I used in spade. But since its use only require to drop a

use MooseX::XSAccessor;

at the top of the file, it’s not like I have much to lose anyway.

The second module is MooseX::ClassAttribute. As you probably realized from the previous code snippets, my code create an object per file. For 100G worth of files, that turns out to be lots of objects. So having configuration attributes that will all end up having the same value over and over again for each object would be quite wasteful. In that case, using an attribute which value is shared for all objects of the class makes much more sense. And, again, using that module is dead simple:

package Deduper::File;

use Moose;
use MooseX::ClassAttribute;

class_has hash_size => (
    isa => 'Int',
    is => 'rw',
    default => 1024,

sub foo {
    my $self = shift;

    # hash_size can be used as a regular attribute
    my $hs = $self->hash_size;

The cherry on top of everything is that those class attributes are totally transparent to the rest of the code. Ever find that you want per-object values after all, change ‘class_has’ for ‘has’, and you’re done.

Keeping Different Functionalities Nicely Segregated

A bonus that using a Moose-based class brought to the project was to keep different functionalities logically apart via method modifiers. For example, the run() method of an app typically juggles with the different options requested:

sub run {
    my $self = shift;

    my $n = $self->max_files || -1;

    until ( $self->finished ) {
        print $self->next_dupe;
        if ( $n > -1 ) {
            $self->finished(1) if $n == 0;
    $self->print_stats if $self->stats;

instead, once can encapsulate those different behaviors in separate functions

sub check_max_files {
    my $self = shift;

    $self->finished(1) if $self->files_to_scan == 0;

sub stats {
    # print a lot of stuff

sub run {
    my $self = shift;

    print $self->next_dupe until $self->finished;

and stitch in whatever is needed at creation time:

sub BUILD {
    my $self = shift;


        next_dupe => \&check_max_file,
    ) if $self->max_files;

    $self->meta->add_after_method_modifier( run => \&print_stats )
        if $self->stats;


If managed properly, this makes each method much smaller and much more single-minded. As a nice side-effect, the final application object will only contain the code that it requires. If the option --max_file isn’t passed, the algorithm won’t have an additional ‘if’ statement to deal with per iteration. It’s not much, but it will make the default case just a tad faster. I must say, however, that this is not a free lunch: if the option is passed, the running time will be slower than if a simple ‘if’ was used. But that trade-off can make sense, like here where I’m ready to have a slight penalty when I run my test runs, but really want to have the tires screeching for the main event.

Finally, The Core Algorithm

With all boilerplate stuff dealt with, we can focus on the core problem. After some experimenting, I ended up with a recipe that isn’t exactly ground-breaking, but seems to be efficient. The encountered files are first classified by file size. For files of the same size, we compare the first X bytes of the file (where ‘X’ is fairly small). If they share that beginning, we compare their last X bytes (in case we’re dealing with file types with the same preamble). Then, ultimately, we compare full-file digests (using Digest::xxHash which is pretty damn fast). Everything is only computed if required, and then cached so that we do the work only once.

Translated into code, it looks pretty much like

# in class Deduper

sub next_dupe {
    my $self = shift;

    return if $self->finished;

    while( my $file = $self->file_iterator->() ) {
        $file = Deduper::File->new( path => $file );
        my $orig = $self->is_dupe($file) or next;
        return $orig => $file;


sub is_dupe {
    my( $self, $file ) = @_;

    return $_ for $self->find_orig($file);

    # not a dupe? enter it in the registry


sub find_orig {
    my( $self, $file ) = @_;

    # do we have any file of the same size?
    my $candidates = $self->files->{$file->size}
        or return;

    my @c;

    # only have a sub-hash if we have more than one
    # file, so as not to compute the 'hash' (beginning of the file) 
    # needlessly
    if( ref $candidates eq 'Deduper::File' ) {
        return if $candidates->hash ne $file->hash;
        @c = ( $candidates );
    else {
        @c = @{ $candidates->{$file->hash} || return };

    # first check if any share the same inode
    my $inode = $file->inode;
    for ( @c ) {
        return $_ if $_->inode == $inode;

    # then check if dupes
    for ( @c ) {
        return $_ if $_->is_dupe($file);


sub add_file {
    my( $self, $file ) = @_;

    if( my $ref = $self->files->{$file->size} ) {
        if ( ref $ref  eq 'Deduper::File' ) {
            $ref = $self->files->{$file->size} = { $ref->hash => [ $ref ] };
        push @{$ref->{$file->hash}}, $file;
    else {
        # nothing yet, just put the sucker
        $self->files->{$file->size} = $file;


# and then in Deduper::File

sub is_dupe {
    my( $self, $other ) = @_;

    # if we are here, it's assumed the sizes are the same
    # and the beginning hashes are the same

    # different hashes?
    return $self->end_hash eq $other->end_hash
        && $self->digest eq $other->digest;
And I Give You The Pudding, Fully Cooked

The application, as submitted to the contest, can be found on GitHub.

“And how did it fare, performance-wise?”, you ask? If the contest results are to be trusted, it processed the 100G monster in a little less than 4 minutes which, I think, ain’t too shabby.

Mostly considering that, y’know, it turned out to be the best time of all submitted entries. :-)

Categories: DBA Blogs

Join Alex Carlos Mauro Scott and speak at the Great Lakes Oracle Conference 2014

Grumpy old DBA - Sat, 2014-01-11 08:34
The Great Lakes Oracle Conference is May 12-14 2014 in Cleveland at CSU.  Beautiful campus environment and state of the art student center main ballroom.  Plus well of course the chance to speak at the city where the river caught on fire right ( long story ask me over a beer ). We have 1/2 day workshops on Monday ( Alex Gorbachev / ( Carlos Sierra and Mauro Pagano ) / Scott Spendolini ) and then main conference starts Tuesday with keynotes by Steven Feuerstein and Tom Kyte. Call for Abstracts is open now here:  Please consider joining us at a rapidly growing regional conference that Carol Dacko considers "One of the best choices for any Oracle Professional to attend".   
Categories: DBA Blogs

Changing OEM12c Repository retention settings

DBASolved - Fri, 2014-01-10 16:14

Today, I’ve been working on a script to pull some sizing/metric information out of the Oracle Enterprise Manager (OEM) repository database.  As I was working through (pulling my hair out…lol) the details; I had a discussion with an good friend.  He mentioned to me that I needed to change  the retention setting on the repository for usage with the script (I was working on) and long-term historical tracking.   Although, I already knew this needed to be done, it took some poking around to find the document number in MOS.

In case you want to check out the document or documentation, you can find this information at these points of interest:
MOS Doc ID: 1405036.1

Armed with the documentation and a few more grey hairs, I took at look at what default settings were configured after building the repository.  The script in Listing 1 will display what the current/default retention settings are.

Listing 1: Script for retention settings
select table_name, partitions_retained
from em_int_partitioned_tables

After running the script in Listing 1, your output should tell you if you are configured with the default settings.  The default settings for retention are 7, 32 and 12.  Figure 1 shows you the output from the environment I’m working in.  As you can see, the retention settings have not been changed.

Figure 1: Current retention settings


These numbers may look a bit funny when you first look at them.  Before changing theses settings it is good to understand what these values mean.  The values for EM_METRIC_VALUES and EM_METRIC_VALUES_HOURLY are displayed in DAYS  The value for EM_METRIC_VALUES_DAILY is displayed in MONTH.

For tracking and historical purposes, these settings are a bit low.  Most organizations will want to have data that is retained for longer period of times; mostly due to security and compliance reasons.  Now, this brings up the topic of how to change these values.

Oracle has provided a PL/SQL API to allow these changes to be done.  The SQL statements in Listing 2 shows how this API is used for each of the retention changes.

Listing 2: PL/SQL API to change retention settings

  gc_interval_partition_mgr.set_retention(‘SYSMAN’, ‘EM_METRIC_VALUES’, 10);

  gc_interval_partition_mgr.set_retention(‘SYSMAN’, ‘EM_METRIC_VALUES_HOURLY’, 90);

  gc_interval_partition_mgr.set_retention(‘SYSMAN’, ‘EM_METRIC_VALUES_DAILY’, 36);

Changes that were made by the PL/SQL API, the retention setting for the OEM repository have been changed.  Figure 2, shows you the updated information that the repository will use.

Figure 2: Updated retention settings


With the updated settings, the OEM repository will now keep raw metric data for 10 days, hourly metric data for 90 days (3 months) and keep all the daily metric data for 36 months (3 years).  This increase in retention should help with reporting on growth of monitored targets over the long term.


twitter: @dbasolved


Filed under: OEM
Categories: DBA Blogs

The Twelve Days of NoSQL: Day Twelve: Concluding Remarks

Iggy Fernandez - Fri, 2014-01-10 10:20
Originally posted on So Many Oracle Manuals, So Little Time:
[twitter-follow screen_name='Oratweets'] Day One: Disruptive Innovation Day Two: Requirements and Assumptions Day Three: Functional Segmentation Day Four: Sharding Day Five: Replication and Eventual Consistency Day Six: The False Premise of NoSQL Day Seven: Schemaless Design Day Eight: Oracle NoSQL Database Day Nine: NoSQL Taxonomy Day…
Categories: DBA Blogs

Meet us at NRF14 booth #2233

Pythian Group - Fri, 2014-01-10 10:18
Categories: DBA Blogs

SQL Server 2012: Installing and configuring Distributed Replay

Pythian Group - Fri, 2014-01-10 09:46

Today we are going to discuss one of the powerful new functions of SQL Server 2012. The name of the feature is Distributed Replay which can be used to assess the impact of changes and upgrades or SQL Server tunings by replaying a trace captured from production SQL Server environment to a test environment.
What are the benefits?

Unlike SQL Server profiler, the Distributed Replay can run the recorded trace against as much as 16 Distributed Replay Client, all these clients are controlled by one Distributed Replay Controller. Distributed Replay components are installed separately from the regular SQL Server install. The Distributed Replay Controller can act as a replacement for Ostress, except for the ability to replay SQL and RML files.

This powerful feature will help customers/DBAs in understanding the impact of any application/configuration changes on performance for multiple servers. You can customize the trace replay by choosing appropriate mode for your environment


We will use a Distributed Replay controller and a Distributed Client on server SQLNODE1, a second Distributed Replay Client on server SQLNODE2 and a target instance with SQL Server 2008 R2 instance.

For installation we need to create two Active Directory accounts .First CORP\DR_Controller for Distributed Replay Controller and CORP\DR_Client for Distributed Replay Client. Below are the screenshots for reference.

Distributed Replay installation

We will start the installation now — Make sure that you install Management Tools too as it provides Distributed Replay administration tool. For this installation I will install Distributed Replay Controller and Distributed Replay Client, as I already have Management tools installed on my SQL Server instances.

We will start installation on SQLNODE1 as it will act as a controller as well as client.

Step #1: Select the installation type. Here we will add features to existing standalone installation, selecting the same.

Step #2: Setup will install setup support files.

Step #3: You have option to include SQL Server updates.

Step #4: Setup will install setup files.

Step #5: Setup should pass all setup support rules. You may ignore warnings and go ahead on a case-by-case basis.

Step #6: Since I already have the DB Engine installed, I will choose to Add features to existing instance of SQL Server 2012.

Step #7: Since this Server will act as a Controller and Client thus selecting both here.

Step #8: This page will review the disk space requirement.

Step #9: On this page, Provide both the accounts created for Distributed Replay Controller and Distributed Replay Client in Active directory.

Step #10: On Distributed Replay Controller page, add the Active Directory account that we have created.

Step #11: On Distributed Replay Client page, provide the controller name.

Step #12: On Error Reporting Page you can opt for sending information of Error reporting to Microsoft.

Step #13: On this page setup will check if the installation will be blocked by running rules. All green here.

Step #14: We are ready to install. You might like to .cross verify the selected configurations before proceeding.

Step #15: Installation is in progress.

Step #16: Finally the setup is successful. You can close the window now.

Windows Firewall configurations

We will configure the Windows firewall on the Distributed Replay Controller to allow inbound connection for DReplayController.exe application .

Go to Windows firewall, Inbound Rules, and add a new rule for a program:

Browse to the DReplayController.exe location:

Allow the Connection.

Specify a name to the rule.

Similarly create a rule on each Distributed Replay Client to allow Distributed Replay Clients to connect and register to the Distributed Replay controller,

Start Services and check client registrations.. First start the Controller service on the Distributed Replay Controller SQLNODE1, open a command prompt screen and enter:

NET STOP “SQL Server Distributed Replay Controller”

NET START “SQL Server Distributed Replay Controller”

You should see the result of this command in the log folder of Distributed Replay Controller under C:\Program Files (x86)\Microsoft SQL Server\110\Tools\DReplayController\log.It should be like this

2014-01-06 20:30:42:482 OPERATIONAL [Controller Service] Microsoft SQL Server Distributed Replay Controller – 11.0.2100.60.

2014-01-06 20:30:42:482 OPERATIONAL [Controller Service] © Microsoft Corporation.

2014-01-06 20:30:42:482 OPERATIONAL [Controller Service] All rights reserved.

2014-01-06 20:30:42:482 OPERATIONAL [Controller Service] Current edition is: [Enterprise Edition].

2014-01-06 20:30:42:482 OPERATIONAL [Controller Service] The number of maximum supported client is 16.

2014-01-06 20:30:42:482 OPERATIONAL [Controller Service] Windows service “Microsoft SQL Server Distributed Replay Controller” has started under service account “CORP\DR_Controller”. Process ID is 2888.

2014-01-06 20:30:42:498 OPERATIONAL [Controller Service] Time Zone: India Standard Time.

2014-01-06 20:30:42:498 OPERATIONAL [Common] Initializing dump support.

2014-01-06 20:30:42:498 OPERATIONAL [Common] Dump support is ready.

Now start Distributed Replay Client service on your Distributed Replay Clients SQLNODE1 and SQLNODE2. Verify that they are correctly synchronized with your Distributed Replay Controller.

On both servers, open a command prompt screen and run:

NET STOP “SQL Server Distributed Replay Client”

NET START “SQL Server Distributed Replay Client”

Open the latest log file in the location C:\Program Files (x86)\Microsoft SQL Server\110\Tools\DReplayClient\log and check for text Registered with controller

2014-01-06 20:30:51:000 OPERATIONAL [Client Service] Microsoft SQL Server Distributed Replay Client – 11.0.2100.60.

2014-01-06 20:30:51:000 OPERATIONAL [Client Service] © Microsoft Corporation.

2014-01-06 20:30:51:000 OPERATIONAL [Client Service] All rights reserved.

2014-01-06 20:30:51:000 OPERATIONAL [Client Service] Current edition is: [Enterprise Edition].

2014-01-06 20:30:51:015 OPERATIONAL [Common] Initializing dump support.

2014-01-06 20:30:51:015 OPERATIONAL [Common] Dump support is ready.

2014-01-06 20:30:51:015 OPERATIONAL [Client Service] Windows service “Microsoft SQL Server Distributed Replay Client” has started under service account “CORP\DR_Client”. Process ID is 2872.

2014-01-06 20:30:51:015 OPERATIONAL [Client Service] Time Zone: India Standard Time.

2014-01-06 20:30:51:015 OPERATIONAL [Client Service] Controller name is “SQLNODE1″.

2014-01-06 20:30:51:015 OPERATIONAL [Client Service] Working directory is “C:\Program Files (x86)\Microsoft SQL Server\110\Tools\DReplayClient\WorkingDir”.

2014-01-06 20:30:51:015 OPERATIONAL [Client Service] Result directory is “C:\Program Files (x86)\Microsoft SQL Server\110\Tools\DReplayClient\ResultDir”.

2014-01-06 20:30:51:015 OPERATIONAL [Client Service] Heartbeat Frequency(ms): 3000

2014-01-06 20:30:51:015 OPERATIONAL [Client Service] Heartbeats Before Timeout: 3

2014-01-06 20:30:51:078 OPERATIONAL [Client Service] Registered with controller “SQLNODE1″.

You might get below error message in the Client Log if in case you forget to add the Client Service account in the Controller Service access permission page during setup or you just change the Distributed Replay client service account .I re-produced the scenario and below is the log file.

2014-01-06 18:04:54:603 OPERATIONAL [Client Service] Microsoft SQL Server Distributed Replay Client – 11.0.2100.60.

2014-01-06 18:04:54:619 OPERATIONAL [Client Service] © Microsoft Corporation.

2014-01-06 18:04:54:619 OPERATIONAL [Client Service] All rights reserved.

2014-01-06 18:04:54:806 OPERATIONAL [Client Service] Current edition is: [Enterprise Edition].

2014-01-06 18:04:54:806 OPERATIONAL [Common] Initializing dump support.

2014-01-06 18:04:54:853 OPERATIONAL [Common] Dump support is ready.

2014-01-06 18:04:54:884 OPERATIONAL [Client Service] Windows service “Microsoft SQL Server Distributed Replay Client” has started under service account “CORP\DR_Client”. Process ID is 2800.

2014-01-06 18:04:55:196 OPERATIONAL [Client Service] Time Zone: India Standard Time.

2014-01-06 18:04:55:243 OPERATIONAL [Client Service] Controller name is “SQLNODE1″.

2014-01-06 18:04:55:243 OPERATIONAL [Client Service] Working directory is “C:\Program Files (x86)\Microsoft SQL Server\110\Tools\DReplayClient\WorkingDir”.

2014-01-06 18:04:55:258 OPERATIONAL [Client Service] Result directory is “C:\Program Files (x86)\Microsoft SQL Server\110\Tools\DReplayClient\ResultDir”.

2014-01-06 18:04:55:258 OPERATIONAL [Client Service] Heartbeat Frequency(ms): 3000

2014-01-06 18:04:55:258 OPERATIONAL [Client Service] Heartbeats Before Timeout: 3

2014-01-06 18:04:55:648 CRITICAL [Client Service] [0xC8100005 (6)] Failed to connect controller with error code 0×80070005.

The error code 0×80070005 is also described as “ACCESS DENIED.” Below are the steps to fix it.

  1. On the server where controller is installed. Goto – Start -> Run and type dcomcnfg and open Component Services.
  2. Navigate to Console Root –> Component Services –> Computers –> My Computer –> DCOM Config -> DReplayController
  3. Open the properties of DReplayController and select Security tab
  4. Edit “Launch and Activation Permissions” and grant “Distributed Replay client service account” permission for “Local Activation” and “Remote Activation”.
  5. Edit “Access Permissions” and grant “Distributed Replay client service account” permission for “Local Access” and “Remote Access”.
  6. Add “Distributed Replay client service account” domain user account within “Distributed COM Users” group.
  7. Restart controller and client services like below

NET STOP “SQL Server Distributed Replay Controller”

NET STOP “SQL Server Distributed Replay Client”

NET START “SQL Server Distributed Replay Controller”

NET START “SQL Server Distributed Replay Client”

  1. Check the Distributed Replay Client log file and see the message “Registered with controller SQLNODE1”


At this point, our Distributed Replay lab is ready to use, clients and controller are registered together. We will perform a demonstration in upcoming blog post and will understand this powerful feature.

Categories: DBA Blogs

Log Buffer #354, A Carnival of the Vanities for DBAs

Pythian Group - Fri, 2014-01-10 09:33

The new year is here with a bang, and the rythem of everything new is propagating into the database technologies, hence providing database bloggers ample material to write about. This Log Buffer Edition covers all that and more.


Do you know that DB12c sqlldr is a little smarter…finally ? Now, *.ctl file is not always necessary.

ODI – High performance data movement using Datapump.

Informatica PowerCenter 9.5.1 Hot Fix 2 is now available in the Business Intelligence (BI) 11g media pack for download and use with BI Applications 7.9.x.

This was tested on Oracle 11.2. If you create a tablespace with a datafile with autoextend on, the file’s maxsize cannot be less than its size. If it is, you get an ORA-02494

Oracle Exadata X4 (Part 2): The All Flash Database Machine?

SQL Server:

What are companies doing to optimize BYOD practices?

Stairway to SQLCLR Level 1: What is SQLCLR?

Relational database or graph database? Why not have both?

SQL Server 2014 Hybrid Cloud Scenarios: Migrating On-Premises SQL Server to Windows Azure Virtual Machines

An AlwaysOn Bug with Service Broker Transmission Queue


MySQL Enterprise Monitor 2.3.15 Is Now Available.

New MySQL web installer, Have you tried it yet?

One of the most important config for InnoDB is the innodb_buffer_pool_size.

XtraBackup Complains of Missing perl-DBD-MySQL.

High-Velocity Data—AKA Fast Data or Streaming Data—seems to be all the rage these days.

Categories: DBA Blogs

How to Install a Clustered SQL Server 2012 Instance – Step-by-step – Part 4

Pythian Group - Fri, 2014-01-10 09:32

We have reached the last article of this series. To close this series out, we will talk about Distributed Transaction Coordinator, or simply DTC. I’ll try to do some simple explanation. After that, I’ll demonstrate how to prepare the DTC for a clustered instance.

What is DTC (MS DTC)?
The MS DTC is a OS level service, which comes automatically installed and running under the Network Service account. Its role is to ensure that a distributed transaction is consistent, even with failures.
Those transactions might be initiated when a transaction is dealing with data on multiple computers via network or when the transaction is dealing with multiple processes in a single computer.

All the participants of a distributed transaction works in sync with the other participants (computers) involved in a transaction, looking for the right moment to commit or abort its work. For this reason, we need to make sure that the computers can reach each other.

Do I need to configure MS DTC on my environment?
The answer for this question is the standard for almost everything involved with SQL Server; It depends. You need to understand whether or not you will perform distributed transactions. If you have more than one instance in the same computer, you won’t need the DTC. On the other hand, if you have a two nodes cluster with two clustered instances communicating with each other, you will need the DTC – the instances could be in different nodes. Another possible scenario is when you have the database engine and SSIS installed, in this case you will need to configure the DTC.

For more information, check this link:

How to create a clustered MS DTC?

Since Windows 2008, we are allowed to have more than one instance of MS DTC in a server/cluster. So, for clustered SQL Server installations is a best practice to have a Role exclusively for the DTC and a dedicated DTC for each SQL Server Role.

As documented per Microsoft, the SQL Server follow this path to choose the MS DTC instance to use:

  • Use MS DTC installed to the local group, else
    • Use the mapped instance of MS DTC, else
      • Use the cluster’s default instance of MS DTC, else
        • Use the local machine’s installed instance of MS DTC

To configure a DTC in cluster, we will need a disk and a hostname.

To configure a Role exclusively for the DTC, follow the steps:

  1. Right-click on Roles and pick the “Configure Role” option.
    Screen Shot 2014-01-02 at 14.18.43
  2. A new window will open. Click “next”.Screen Shot 2014-01-02 at 14.18.52
  3. Choose the option “Distributed Transaction Coordinator (DTC)” from the list. Click  “Next”.Screen Shot 2014-01-02 at 14.19.11
  4. Fill the hostname in the “Name” field and the IP in the “Network” section. Click “Next”.Screen Shot 2014-01-02 at 14.20.33
  5. Pick up the disk to be used. Click “Next”.Screen Shot 2014-01-02 at 15.14.42
  6. Review the configurations and click “Next”.Screen Shot 2014-01-02 at 15.14.57
  7. The installation will run and in the last step you will see a report. Click “Finish”.Screen Shot 2014-01-02 at 15.15.11
  8. Now you will be able to see a new Role created in the cluster, with all the indicated resources.
Screen Shot 2014-01-02 at 15.16.51

To add a DTC resource into the SQL Server Role, follow the steps:

  1. Right-click the Role, go to “Add Resource”->”More Resources” -> “Distributed Transaction Coordinator”.Screen Shot 2014-01-02 at 15.30.50
  2. The resource will be created in the selected Role, now we need to configure it. Right-click the “New Distributed Transaction Coordinator” and click on “Properties”.Screen Shot 2014-01-02 at 15.31.20
  3. As referred early on this article, the DTC needs a hostname and a disk to work. On dependencies you can pick up those items as shown, and click “ok”.Screen Shot 2014-01-02 at 15.32.44
  4. Now, let’s bring it online.Screen Shot 2014-01-02 at 15.32.55

How to configure the network for distributed transactions?

Note: On clustered environments, you just need to perform the following steps one time.

  1. On “Server Manager” go to “Tools”->”Component Services” or run the command “dcomcnfg”.Screen Shot 2014-01-02 at 15.33.55
  2. Expand the tree, right-click the desired DTC and choose “Properties”.Screen Shot 2014-01-03 at 11.46.28
  3. Go to the “Security” tab and check “Network DTC Acess” as well as “Allow Inbound” and “Allow Outbound”, as shown bellow. Click Ok.Screen Shot 2014-01-03 at 11.49.02
  • Let’s briefly describe the some of the options on this window:
    • Network DTC Access“: Enable/Disable the network access.
    • Allow inbound“:  Permit a distributed transaction originated from another computer to run on the current computer.
    • Allow outbound“:  Permit a distributed transaction initiated in the current computer to run on a remote computer.
    • Enable XA transactions” and “Enable SNA LU 6.2 Transactions“: Enables/Disable those particular specifications for distributed transactions.

Troubleshooting DTC

There’s a tool called DTC Ping which can help us to verify if the DTC is working correctly on all the computers that should be involved in a transaction.

You can download this tool here:

I recommend the reading of this article, to learn hos to use this tool, as well as troubleshoot the possible errors: Troubleshooting MSDTC issues with the DTCPing tool.

Another great tool is the DTC Tester. You can simulate a distributed transaction on SQL Server:

Screen Shot 2014-01-03 at 14.15.25

To download and get more info about this tool, check this link: .

The End

This way we finish this series about how to install a clustered instance. We still have too many details to cover and I will try to create separated articles with best practices, configuration alternatives, etc.

I hope you enjoyed this series and as always, if you have any doubts, contact me!

Thank you for reading!

If you want to check the other parts of this series, here are the links:

Categories: DBA Blogs

A New Year, a New Dancecard

Pythian Group - Fri, 2014-01-10 09:29

First things first: Happy New Year y’all! Health, happiness, and all that jazz to each and everyone of youses!

So the Holidays are over — or almost over, if you are one of the few lucky souls. Which means that we have to get back into the saddle. Considering that, ahem, some of us have just fallen off the grid and into the eggnog for the last two weeks, that’s easier said than done. In order to jog those hacking muscles back into shape, I decided to revisit an idea that I outlined in the Holidays of 2012: dancecard.

To recap, when I start a new Dancer (or Catalyst application, I always have to drop in the JavaScript libraries I’m using. That’s tedious. Instead, I would like to have a little tool that invites chosen libraries to come and dance with the app. A kind of Tanzkarte, if you will.

For the 2014 edition of this dancecard, I decided to take a slightly different approach. Instead of keeping a repository of the libraries locally, I wanted to be even lazier and simply have a main registry that would point to the download location or the git repository of the libraries.

For the moment I won’t go into the details of the code. Mostly because I pretty much banged the keyboard like a semi-hangover monkey for the last 2 hours, and it reflects in the quality of the codebase. Let it just be said that the git repo is at the usual place, and that I used MooseX::App to build the app.

So, what did those few hours of hacking gave me? Well, with the config file ~/.dancecard/config.yaml, which looks like:

browser_command: firefox %s
            - underscore

I can now get a list of all libraries I have at my disposal:

$ dancecard list
    no lib installed

Of course, I can also install any of those. And… look Ma: dependencies are handled automatically!

$ dancecard install backbone
 backbone installed
 underscore installed

Something I also often do is peruse the online documentation of those libraries. Noticed those “documentation” urls in the yaml registry? It’s there so that I can do,

$ dancecard doc bacbone

and have the documentation page opened automatically in Firefox. Or, for maximum laziness, I can also do,

$ dancecard doc --installed

which will open the documentation URL of every library found in the current project.

Next Steps

If anyone shows interest in the thing, I’ll bundle the app for CPAN consumption. As for features, things that I’ll probably tackle for my own use are:

  • Tags (no software is true software without tags nowadays);
  • Virtual libraries (just as a mean to install a group of libraries in a single go);
  • Fish completion (I wuv the fish shell);
  • Cached local documentation (’cause hitting the ‘Net each time I look for a piece of documentation fills me with guilt);
  • An online main registry of libraries;
  • And whatever else I may think of.
Categories: DBA Blogs

Creating A Tool: Detailing Oracle Process CPU Consumption

Detailing an Oracle Database process's CPU consumption is amazing, a lot of fun, and can lead to some “ah ha” moments. To make this posting daily digestible, I’m breaking it up into a multi-part series. Today I’m posting the third and final part entitled, Creating A Tool. Here we go…

In the previous post I ended with a discussion about how to gather an operating system's CPU consumption by function. More specific for our purposes is gathering Oracle process (session) CPU consumption at the Oracle kernel function level.

I Want Something Like This
At this point, we are ready to take the Linux perf tool output and merge that with Oracle wait event data. The output could look something like this:

Time Component secs %
------------------------------------------------------- ---------- -------
cpu : [.] kcbgtcr 29.714 66.87
cpu : [.] kdstf000010100001km 3.716 8.36
cpu : [.] lnxsum 3.541 7.97
cpu : [?] sum of funcs consuming less than 2% of CPU ti 2.393 5.38
cpu : [.] kaf4reasrp0km 2.180 4.91
wait: latch: cache buffers chains 2.100 4.73

I suspect you're not surprised that the above output was, in fact, taken from an existing tool. The tool is called "" and is the result of a collaborative effort with myself and Frits Hoogland. You can download the free tool here. If the link doesn't work, just do an search for "fulltime" and you'll see it.

How The Tools Works
Obviously the tool collects both process CPU consumption and also Oracle wait event data. But we wanted some usage flexibility, so it's a little more interesting than that. Also, the real trick is outputting the perf data into a comma delimited file using "perf report -t,". It's the "t" option that is key here. Oracle can easily read the coma delimited file as an external table and then combine that with the wait event data collected from the v$sess_time_model view.

Here is the basic idea:

Help user identify the PID to profile
Initial setup
get oracle wait times (snap 0)
get oracle CPU time (snap 0)
start oracle kernel cpu details collection
sleep x
get oracle wait times (snap 1)
get oracle CPU time (snap 1)
stop oracle kernel cpu collection
do some cool math and combine results
display results
End Loop

The looping capability easily allows for a single collection or multiple collection. The multiple collection option displays much like the "top" tool with a default cycle duration of three seconds.

If you're running on a virtual machine, make sure to change the shell script variable PERF_SAMPLE_METHOD to '-e cpu-clock' instead of the default '-e cycles'. If you look near the top of the script, you will see a comment about this and also where to make the change.

Before you run the script, take a quick look at the script, especially the top third. There is some good information about the script, default settings, and usage details.

How To Run The Script
There are two input usages. Let's call them basic and advanced.

The BASIC USAGE helps you find the Oracle process ID, sets some defaults for you, and works on a three second cycle until you control-C. For example, let's say you don't know the OS process but you do know the Oracle SID or perhaps you know the machine name or user name. Then you're in luck because the script will display this information and then prompt you for the OS PID! Here is an example of you can expect if you simply enter, "./" and selected process 60505.

For each cycle, you would see something like this:

Notice that both Oracle and OS process details are shown in conjunction with the date and time. Plus the SQL statement the process is running at the end of the sample period! Then the total time is shown along with the two high level components, that is, CPU consumption and Oracle wait time. Finally, the details are shown. Notice the details clearly identify the time component as either CPU or wait time.

If the time is tagged CPU then the Oracle kernel function is displayed (thank you perf) along with it's inferred time (based on both perf and v$sess_time_model). If the time is tagged Oracle wait time then the wait event name is displayed along with the wait time (based on v$session_event).

It's important to understand that if we ran a standard wait event report, the output would be similar but with NO CPU information. The wait events detail would still be displayed. This tool simply adds value by incorporating the CPU consumption.

The ADVANCED USAGE gives us full control. The advanced usage requires the OS process ID, the cycle duration, and the number of cycles. The advanced option makes it simple to get a single long cycle or multiple shorter cycles. For example, if I wanted to watch process 1234 in 5 second intervals, I could do this:
$ ./ 1234 5 9999
Or suppose I wanted a single 60 second sample for process 5432. I would enter this:
$ ./ 5432 60 1
The output will look exactly like the output from basic usage. You are just not prompted or given information to help you pick the OS PID.

Here is a short video so you can watch this in action! - Report both Oracle Wait and CPU time details from OraPub on Vimeo.
Why This Is So Important
Look at the example screen shot below and ask yourself, "Is there an IO bottleneck?"

If the CPU information was not combined with the wait time details, I suspect 90% of Oracle DBAs would say there is an IO problem. They would see the top wait event db file sequential read then conclude there is an IO issue. However, as the script clearly shows, CPU consumption is much larger than Oracle wait time, providing us with a greater opportunity to reduce the total time and devise additional spot-on solutions. Also looking at the report's "total time" breakdown, it clearly shows CPU consumption is 97% of the total time. I wouldn't be surprised if there was a raging CPU bottleneck!

So Where Are We In This Quest?
Now that you have access to the script and know how to use it, you're going to start asking, "Just what is kcbgtcr anyways?!" In my next posting I'll address this.

But until then, get the script, log into your non-production yet active Linux box and give it a go! If it's not production, profile the Oracle log writer or database writer background process or perhaps an active server process!

Feeling dangerous yet? (Always!)

Here are the key resource links:

1. Presentation: Detailing Oracle CPU Consumption. Download HERE.
2. Tool: Oracle wait time and kernel function CPU monitor. Download the script HERE.

Thanks for reading,


If you enjoy my blog, I suspect you'll get a lot out of my courses; Oracle Performance Firefighting,  Advanced Oracle Performance Analysis, and my new seminar, Go Faster: Make Oracle Work For You! I teach these classes around the world multiple times each year. For the latest schedule, go to . I also offer on-site training and consulting services.

P.S. If you want me to respond to a comment or you have a question, please feel free to email me directly at craig@orapub .com. Another option is to send an email to OraPub's general email address, which is currently info@ orapub. com.

Categories: DBA Blogs

Log Buffer #353, A Carnival of the Vanities for DBAs

Pythian Group - Fri, 2014-01-10 08:48

The new fresh year is already here amidst fireworks, hard partying, resolutions, and new hopes. Log Buffer also enters into this brand new year focusing on the emerging trends while appreciating the proven technologies. Enjoy the first log buffer of this year.


Keeping up with the latest Java Runtime Environment (JRE) updates is both a priority and a challenge for most organizations.

Oracle‘s product portfolio offers a multitude of opportunities to combine cross-stack products and deliver incremental value to customers and partners.

Do you Deal with Supplier Profile Management flows?

In case you took a break over New Year’s and aren’t quite ready for “real” work, but still feel obligated to visit sites such as this one, here’s a reward for your efforts.

Using SSD for a temp tablespace on Exadata

SQL Server:

This paper discuss if it is the right approach of using Hadoop as the analytics infrastructure.

Microsoft is releasing SQL Server 2014 with a new set of In-Memory OLTP features to significantly improve the OLTP performance and reduce the processing time for servers with a large amount of memory and multi-core processors.

SQL Server Optimizer — Under the Hood series of short articles.

How to use Profiler to generate TSQL scripts that can be modified and automated to run as a server-side trace by Grant Fritchey.

Big data is now a standard part of information technology architecture for most large organizations.


SSL with MySQL does not have to be complicated

This is a HOW-TO about installing MariaDB Galera Cluster on Debian/Ubuntu.

In many parts of the MySQL world, whether you have users, developers or administrators of MySQL, the season holidays are for family, relax, prayers, travel, or they are simply a moment where you can enjoy what you like or care most about.

How to recover table structure from .frm files with MySQL Utilities

Chinese and Japanese and Korean (CJK) text usually has no spaces between words. Conventional full-text search does its tokenizing by looking for spaces. Therefore conventional full-text search will fail for CJK.

Categories: DBA Blogs

Partner Webcast – Oracle Business Analytics: Make data work for you!

Organizations today have to remain competitive in a rapidly changing economic environment with increased global competition, reduced product lifecycles, and growing volumes of data. It has become...

We share our skills to maximize your revenue!
Categories: DBA Blogs

Bug 13930580 Example

Bobby Durrett's DBA Blog - Thu, 2014-01-09 13:37

It looks like we hit 13930580 on one of our systems and I wanted to put some example output showing log file sync times going above 10 milliseconds (10,000 microseconds) after the log writer switches to polling.

Here is the output in the lgwr trace file showing the switch to polling:

*** 2013-12-27 01:50:47.374
kcrfw_update_adaptive_sync_mode: post->poll long#=1 sync#=1 
sync=55336 poll=1913
4 rw=5506 rw+=5506 ack=0 min_sleep=19134

*** 2013-12-27 01:50:47.374
Log file sync switching to polling
Current scheduling delay is 1 usec
Current approximate redo synch write rate is 0 per sec

Here is a report showing the number of log file sync waits and their average time in microseconds:

END_INTERVAL_TIME          number of waits ave microseconds
-------------------------- --------------- ----------------
26-DEC-13 PM           976539       2279.06861
26-DEC-13 PM           681212       2029.32406
26-DEC-13 PM           343932       1575.26284
26-DEC-13 PM           163911       1850.74354
26-DEC-13 PM            73151       3815.28597
26-DEC-13 PM            39304       5038.05427
26-DEC-13 PM            32122       5677.00557
26-DEC-13 PM           472349       2353.95857
26-DEC-13 PM            18679       18655.5294
27-DEC-13 AM            19618       17046.2287
27-DEC-13 AM            18983       5721.99178
27-DEC-13 AM            17724       17106.3415
27-DEC-13 AM             9088       25342.7271
27-DEC-13 AM            14365          12128.7
27-DEC-13 AM            16323       12879.8831
27-DEC-13 AM            43758       15326.7298
27-DEC-13 AM            83819       14796.2851
27-DEC-13 AM           168718       13506.4363
27-DEC-13 AM           302827       19116.9491
27-DEC-13 AM           480347       19358.6655
27-DEC-13 AM           512777       15952.2358
27-DEC-13 PM           511091       13799.5512
27-DEC-13 PM           576341       10183.4347
27-DEC-13 PM           524568       10251.1259

Notice how the average wait time goes above 10,000 microseconds consistently once the log writer switches to polling between 1 and 2 am on 12/27/2013.  I didn’t show all the output but this long log file sync wait time has continued since the switch.

Also, these long log file sync times don’t correspond to long log file parallel write times.  Here are the number and averages of log file parallel write waits for the same time frame:

END_INTERVAL_TIME          number of waits ave microseconds
-------------------------- --------------- ----------------
26-DEC-13 PM           902849       1426.66601
26-DEC-13 PM           659701       1394.87763
26-DEC-13 PM           344245       1294.92401
26-DEC-13 PM           166643       1586.64944
26-DEC-13 PM            80457       4019.29429
26-DEC-13 PM            46409       5580.67827
26-DEC-13 PM            69218       5115.20904
26-DEC-13 PM           475297       2219.80541
26-DEC-13 PM            40943        19405.052
27-DEC-13 AM            38835       18160.8073
27-DEC-13 AM            24734       6321.38425
27-DEC-13 AM            33617       11723.6698
27-DEC-13 AM            36469       17485.2614
27-DEC-13 AM            19344       6955.27042
27-DEC-13 AM            17857       4399.75718
27-DEC-13 AM            45098       4923.02763
27-DEC-13 AM            83700       3610.39713
27-DEC-13 AM           160919       2841.31507
27-DEC-13 AM           266405       3523.86855
27-DEC-13 AM           384795        3367.5075
27-DEC-13 AM           437806       2729.84248
27-DEC-13 PM           448261       2442.81012
27-DEC-13 PM           511648       1880.74418
27-DEC-13 PM           481106       1919.21158

The average I/O time – log file parallel write – is pretty low when the system is active (more than 100,000 waits per hour) – usually less than 4000 microseconds and yet log file sync is always more than 10,000 after the switch to polling. Also, the CPU on the system is consistently less than 30% used so it isn’t a system load issue.

Here are some Oracle support documents related to this issue:

Adaptive Switching Between Log Write Methods can Cause 
'log file sync' Waits (Doc ID 1462942.1)

Waits for "log file sync" with Adaptive Polling vs Post/Wait 
Choice Enabled (Doc ID 1541136.1)


Here is the script I used to get the wait output:

set linesize 32000
set pagesize 1000
set long 2000000000
set longchunksize 1000
set head off;
set verify off;
set termout off;

column u new_value us noprint;
column n new_value ns noprint;

select name n from v$database;
select user u from dual;
set sqlprompt &ns:&us>

set head on
set echo on
set termout on
set trimspool on


spool "&ns.&&WAITNAME..log"

column END_INTERVAL_TIME format a26

(after.total_waits-before.total_waits) "number of waits",
(after.total_waits-before.total_waits) "ave microseconds",
before.event_name "wait name"
where before.event_name='&&WAITNAME' and
after.event_name=before.event_name and
after.snap_id=before.snap_id+1 and
after.instance_number=1 and
before.instance_number=after.instance_number and
after.snap_id=sn.snap_id and
after.instance_number=sn.instance_number and
(after.total_waits-before.total_waits) > &&MINIMUMWAITS
order by after.snap_id;

spool off

I gave log file sync as the WAITNAME and 1 as MINIMUMWAITS for the first output.  I changed WAITNAME to log file parallel write for the second one with 1 still for MINIMUMWAITS.

It looks like there is a new feature in 11.2 that was finally turned on by default in  The work around is to set a hidden parameter to false to turn off the new feature.  Check out the Oracle support docs I listed for details.

- Bobby

p.s. I forgot to mention that when I tested on a test database with and without this new feature the log file sync times were lower with the new feature, as they should be.

With _use_adaptive_log_file_sync=TRUE (feature enabled):

Top 5 Timed Foreground Events

Event Waits Time(s) Avg wait (ms) % DB time Wait Class log file sync 639,598 3,466 5 86.74 Commit DB CPU 397 9.93 buffer exterminate 683 14 21 0.36 Other virtual circuit wait 912 12 13 0.31 Network SQL*Net message to client 1,293,196 7 0 0.17 Network

With _use_adaptive_log_file_sync=FALSE (disabled as in earlier versions of 11.2):

Top 5 Timed Foreground Events

Event Waits Time(s) Avg wait (ms) % DB time Wait Class log file sync 639,644 3,553 6 87.31 Commit DB CPU 367 9.02 buffer exterminate 1,364 28 21 0.69 Other buffer busy waits 2,679 15 6 0.37 Concurrency virtual circuit wait 903 13 15 0.32 Network

With the new feature enabled log file sync was 5 milliseconds instead of 6 without it.  So, the new feature does speed up log file sync waits when it is working normally.  But, there must be some bug condition where it degrades to greater than 10 millisecond log file syncs.

Categories: DBA Blogs

Four tough choices for workshops at GLOC 2014

Grumpy old DBA - Thu, 2014-01-09 12:18
This year at the Great Lakes Oracle Conference ( May 12-14 2014 at CSU ) we are doing 1/2 day workshops monday afternoon.  Then the conference kicks into full speed with keynotes tuesday morning and session tracks after that.

We have an incredible set of people doing workshops for our attendees.  We are working on finalizing a fourth workshop in the Essbase area but not quite ready to announce that on.

We are going to go live this weekend with the ability to register for our latest addition to the workshops.

Alex Gorbachev:
From Relational to Hadoop - Migrating your data pipelineI am very excited to announce this addition!

Will update this post with links when this workshop is online for registration. 
Categories: DBA Blogs

Quick Tip on Using Sqoop Action in Oozie

Chen Shapira - Wed, 2014-01-08 18:59

Another Oozie tip blog post.

If you try to use Sqoop action in Oozie, you know you can use the “command” format, with the entire Sqoop configuration in a single line:

<pre><workflow-app name="sample-wf" xmlns="uri:oozie:workflow:0.1">
    <action name="myfirsthivejob">
        <sqoop xmlns="uri:oozie:sqoop-action:0.2">
            <command>import  --connect jdbc:hsqldb:file:db.hsqldb --table TT --target-dir hdfs://localhost:8020/user/tucu/foo -m 1</command>
        <ok to="myotherjob"/>
        <error to="errorcleanup"/>

This is convenient, but can be difficult to read and maintain. I prefer using the “arg” syntax, with each argument in its own line:

<workflow-app name="sample-wf" xmlns="uri:oozie:workflow:0.1">
    <action name="myfirsthivejob">
        <sqoop xmlns="uri:oozie:sqoop-action:0.2">
        <ok to="myotherjob"/>
        <error to="errorcleanup"/>

As you can see, each argument here is in its own “arg” tag. Even two arguments that belong together like “–table” and “TT” go in two separate tags.
If you’ll try to put them together for readability, as I did, Sqoop will throw its entire user manual at you. It took me a while to figure out why this is an issue.

When you call Oozie from the command line, all the arguments you pass are sent as a String[] array, and the spaces separate the arguments into array elements. So if you call Sqoop with “–table TT” it will be two elements, “–table” and “TT”.
When using “arg” tags in Oozie, you are basically generating the same array in XML. Oozie will turn the XML argument list into an array and pass it to Sqoop just the way it would in the command line. Then Sqoop parses it in exactly the same way.
So every item separated with space in the command line must be in separate tags in Oozie.

Its simple and logical once you figure out why :)
If you want to dig a bit more into how Sqoop parses its arguments, it is using Apache Commons CLI with GnuParser. You can read all about it.

Categories: DBA Blogs

Delphix upgrade in twenty minutes

Bobby Durrett's DBA Blog - Wed, 2014-01-08 17:29

Just got off of a Webex with Delphix support.  They upgraded our Delphix server from version to  It took about twenty minutes.  Pretty nice compared to an Oracle database upgrade I think!

The only thing that took some time was that I had to be sure I had enough space.  It looks like Delphix needs your disk space to be no more than 85% utilized to fully function.  We ended up extending our four 1.5 terabyte luns to 1.9 TB each to give us space.  Then I cleaned up some archive log space by running a snapsync on each source database to completion.  Our system just needed a little TLC to get some space free again.

But, the upgrade itself, running the scripts and rebooting the server, took 20 minutes and was all done by Delphix support for no extra charge.  Sweet.

- Bobby

Categories: DBA Blogs

Disabling Triggers in Oracle

Pythian Group - Wed, 2014-01-08 08:10

In March 2012, I put together a blog post entitled Disabling Oracle triggers on a per-session basis, outlining a way to suspend trigger execution for the current session through a PL/SQL call. Commenter Bryan posted a comment saying he couldn’t get it working in

Unfortunately Oracle seems to have disabled this use in, and most likely 12.1 as well. Boo-Hiss! This is needed functionality for DBAs!

A new parameter: enable_goldengate_replication

I tried this on an Oracle system, and I indeed got an error:

SQL> exec sys.dbms_xstream_gg.set_foo_trigger_session_contxt(fire=>true);
BEGIN sys.dbms_xstream_gg.set_foo_trigger_session_contxt(fire=>true); END;

ERROR at line 1:
ORA-26947: Oracle GoldenGate replication is not enabled.
ORA-06512: at "SYS.DBMS_XSTREAM_GG", line 13
ORA-06512: at line 1

A quick look at oerr gives a path forward, assuming you do indeed have a GoldenGate license:

[oracle@ora11gr2b ~]$ oerr ora 26947
26947, 00000, "Oracle GoldenGate replication is not enabled."
// *Cause: The 'enable_goldengate_replication' parameter was not set to 'true'.
// *Action: Set the 'enable_goldengate_replication' parameter to 'true'
//           and retry the operation.
//          Oracle GoldenGate license is needed to use this parameter.

The Oracle reference gives a bit more info

ENABLE_GOLDENGATE_REPLICATION controls services provided by the RDBMS for Oracle GoldenGate (both capture and apply services). Set this to true to enable RDBMS services used by Oracle GoldenGate.

The RDBMS services controlled by this parameter also include (but are not limited to):

Service to suppress triggers used by GoldenGate Replicat

As do the GoldenGate 12.1.2 docs:

The database services required to support Oracle GoldenGate capture and apply must be enabled explicitly for an Oracle database. This is required for all modes of Extract and Replicat.

To enable Oracle GoldenGate, set the following database initialization parameter. All instances in Oracle RAC must have the same setting.


So here goes nothing:

SQL> alter system set enable_goldengate_replication=true;

System altered.

SQL> exec sys.dbms_xstream_gg.set_foo_trigger_session_contxt(fire=>true);
BEGIN sys.dbms_xstream_gg.set_foo_trigger_session_contxt(fire=>true); END;

ERROR at line 1:
ORA-01031: insufficient privileges
ORA-06512: at "SYS.DBMS_XSTREAM_GG", line 13
ORA-06512: at line 1

Another error: missing privileges. I checked and double-checked that the required GoldenGate privileges were indeed assigned.

Tracing and permission checks

It’s time to run a 100046 trace (SQL trace) to see what’s really going on.

SQL> alter session set events '10046 trace name context forever, level 12';

Session altered.

SQL> exec sys.dbms_xstream_gg.set_foo_trigger_session_contxt(fire=>true);
BEGIN sys.dbms_xstream_gg.set_foo_trigger_session_contxt(fire=>true); END;

ERROR at line 1:
ORA-01031: insufficient privileges
ORA-06512: at "SYS.DBMS_XSTREAM_GG", line 13
ORA-06512: at line 1

And tracefile does show some interesting information. A few of the more interesting snippets:

PARSING IN CURSOR #140324121137184 len=76 dep=0 uid=91 oct=47 lid=91 tim=1388531465245781 hv=1323338123 ad='6c1f63a0' sqlid='gvq73797f12cb'
BEGIN sys.dbms_xstream_gg.set_foo_trigger_session_contxt(fire=>true); END;
PARSING IN CURSOR #140324121064984 len=187 dep=1 uid=0 oct=3 lid=0 tim=1388531465246387 hv=2028900049 ad='6c128db8' sqlid='aa9h2ajwfx3qj'

The SQL statement is actually checking two things. The first is looking for the current username in the dba_goldengate_privileges view. This view isn’t listed in the Oracle 11.2 documentation, but it does appear in the 12c docs:

ALL_GOLDENGATE_PRIVILEGES displays details about Oracle GoldenGate privileges for the user.

Oracle GoldenGate privileges are granted using the DBMS_GOLDENGATE_AUTH package.

Related Views

DBA_GOLDENGATE_PRIVILEGES displays details about Oracle GoldenGate privileges for all users who have been granted Oracle GoldenGate privileges.

USER_GOLDENGATE_PRIVILEGES displays details about Oracle GoldenGate privileges. This view does not display the USERNAME column.

I had previously run dbms_goldengate_auth to grant privs here, so should be OK.

The second check simply verifies that the DBA role had been granted to the current user, again as recommended by the documentation. (A side note: in previous versions, I had avoided granting the overly broad DBA role to the GoldenGate user in favor of specific grants for the objects it uses. There’s no reason for the GoldenGate user to need to read and modify data objects that aren’t related to its own replication activities for example. And I would argue that it helps avoid errors such as putting the wrong schema in a map statement when permissions are restricted. But sadly it’s no longer possible in the world of

Running the query manually to verify that the grants are indeed in place:



Looks good, so that doesn’t seem to be the problem.

Tracing #2: system properties

Back to the 10046 tracefile:

PARSING IN CURSOR #140324119717656 len=45 dep=1 uid=0 oct=3 lid=0 tim=1388531465253124 hv=3393782897 ad='78ae2b40' sqlid='9p6bq1v54k13j'
select value$ from sys.props$ where name = :1
FETCH #140324119717656:c=0,e=44,p=0,cr=2,cu=0,mis=0,r=0,dep=1,og=1,plh=415205717,tim=1388531465254441

Because this SQL statement involves an ordinary select without an aggregate function, I can look at the FETCH line in the tracefile to get the number of rows returned. In this case it’s r=0, meaning no rows returned.

The query itself is looking for a system property I haven’t seen before: GG_XSTREAM_FOR_STREAMS. A Google search returns only a single result: the PDF version of the Oracle 11.2 XStream guide. Quoting:

This procedure enables XStream capabilities and performance optimizations for Oracle
Streams components.
This procedure is intended for users of Oracle Streams who want to enable XStream
capabilities and optimizations. For example, you can enable the optimizations for an
Oracle Streams replication configuration that uses capture processes and apply
processes to replicate changes between Oracle databases.
These capabilities and optimizations are enabled automatically for XStream
components, such as outbound servers, inbound servers, and capture processes that
send changes to outbound servers. It is not necessary to run this procedure for
XStream components.
When XStream capabilities are enabled, Oracle Streams components can stream ID key
LCRs and sequence LCRs. The XStream performance optimizations improve efficiency
in various areas, including:
? LCR processing
? Handling large transactions
? DML execution during apply
? Dependency computation and scheduling
? Capture process parallelism

On the surface, I don’t see what this would have to do with trigger execution, but I’m going to try enabling it as per the newly read document anyway:

SQL> exec dbms_xstream_adm.ENABLE_GG_XSTREAM_FOR_STREAMS(enable=>true);

PL/SQL procedure successfully completed.

SQL> exec sys.dbms_xstream_gg.set_foo_trigger_session_contxt(fire=>true);
BEGIN sys.dbms_xstream_gg.set_foo_trigger_session_contxt(fire=>true); END;

ERROR at line 1:
ORA-01031: insufficient privileges
ORA-06512: at "SYS.DBMS_XSTREAM_GG", line 13
ORA-06512: at line 1

No dice.

Tracing #3: process names

Onto the next SQL in the tracefile:

PARSING IN CURSOR #140324120912848 len=114 dep=1 uid=0 oct=3 lid=0 tim=1388531465255628 hv=1670585998 ad='6c2d6098' sqlid='a9mwtndjt67nf'

Now we look in v$session, to see if a session associated with the process with OS PID 2293 (which happens to be the SPID of our current shadow process) has a PROGRAM column starting with the word extract. extract is, naturally, the name of the GoldenGate executable that captures data from the source system. In a GoldenGate system, however, trigger suppression does not happen in the extract process at all, but rather the replicat process that applies changes on the target system. So I’m going to skip this check and move on to the next one in the tracefile:

PARSING IN CURSOR #140324120905624 len=169 dep=1 uid=0 oct=3 lid=0 tim=1388531465257346 hv=3013382849 ad='6c122b38' sqlid='38pkvxattt4q1'

This SQL is similar to the previous one, but instead of looking for a program called extract, it looks for one called replicat, and adds an extra check, so see if the module column either starts with OGG or is called GoldenGate. And since it’s the replicat process that does trigger disabling in GoldenGate, this check is likely to be related.

To make this check succeed, I’m going to have to change both the program and module columns in v$session for the current session. of the two, module is much easier to modify: a single call to dbms_application_info.set_module. But modifying program is less straightforward. One approach is to use Java code with Oracle’s JDBC Thin driver and setting the aptly-named v$session.program property, as explained in De Roeptoeter. But I’m hoping to stay with something I can do in SQL*Plus. If you’ve looked through a packet trace of a SQL*Net connection being established, you will know that the program name is passed by the client at the time of connection establishment, so could be modified by either modifying the network packet in transit. This is also complex to get working, as it also involves fixing checksums and the like. There’s a post on Slavik’s blog with a sample OCI C program that modifies its program information. Again more complexity thn I’d like, but it gave me an idea: if the program is populated by the name of the client-side executable, why don’t we simply copy sqlplus to a name that the dbms_xstream_gg likes better?

[oracle@ora11gr2b ~]$ cp $ORACLE_HOME/bin/sqlplus ./replicat
[oracle@ora11gr2b ~]$ ./replicat ggs

SQL*Plus: Release Production on Mon Dec 30 14:09:05 2013

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

Enter password:

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

SQL> exec dbms_application_info.set_module('OGG','');

PL/SQL procedure successfully completed.

SQL> exec sys.dbms_xstream_gg.set_foo_trigger_session_contxt(fire=>true);

PL/SQL procedure successfully completed.


Wrapping up

So it looks like you can disable triggers per-session in just like previous versions, but need to jump through a few more hoops to do it. A few conclusions to draw:

  • Oracle patchsets, while normally intended to include bugfixes, can have major changes to underlying functionality too. See Jeremy Schneider’s post on adaptive log file sync for an even more egregious example. So before applying a patchset, test thoroughly!
  • The enforcement of full DBA privileges for the GoldenGate user in Oracle requires very broad permissions to use GoldenGate, which can be a concern in security-conscious or consolidated environments.

TL;DR: Yes you can still disable triggers per-session in Oracle, but you have to have a GoldenGate license, set the enable_goldengate_replication parameter, use a program name that starts with replicat, and set your module to OGG.

Categories: DBA Blogs

#Oracle Database whoami for Multitenant

The Oracle Instructor - Wed, 2014-01-08 05:49

As an enhancement to the Oracle Database whoami for versions before 12c, this also shows the Container Name to which the session is connected:

[oracle@linuxbox ~]$ sqlplus / as sysdba

SQL*Plus: Release Production on Wed Jan 8 12:34:04 2014

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

Connected to:
Oracle Database 12c Enterprise Edition Release - 64bit Production
With the Partitioning, OLAP, Advanced Analytics and Real Application Testing options

SQL> @whoami
CDB NAME: cdb1
OS USER: oracle

PL/SQL procedure successfully completed.

SQL> connect system/oracle_4U@pdb1
SQL> @whoami
CDB NAME: cdb1
OS USER: oracle
CLIENT IP ADDRESS: 555.555.5.555

PL/SQL procedure successfully completed.

The content of whoami.sql:

set serveroutput on
dbms_output.put_line('USER: '||sys_context('userenv','session_user'));
dbms_output.put_line('SESSION ID: '||sys_context('userenv','sid'));
dbms_output.put_line('CURRENT_SCHEMA: '||sys_context('userenv','current_schema'));
dbms_output.put_line('INSTANCE NAME: '||sys_context('userenv','instance_name'));
dbms_output.put_line('CDB NAME: '||sys_context('userenv','cdb_name'));
dbms_output.put_line('CONTAINER NAME: '||sys_context('userenv','con_name'));
dbms_output.put_line('DATABASE ROLE: '||sys_context('userenv','database_role'));
dbms_output.put_line('OS USER: '||sys_context('userenv','os_user'));
dbms_output.put_line('CLIENT IP ADDRESS: '||sys_context('userenv','ip_address'));
dbms_output.put_line('SERVER HOSTNAME: '||sys_context('userenv','server_host'));
dbms_output.put_line('CLIENT HOSTNAME: '||sys_context('userenv','host'));

Shortcut to get the name of the current container is:

SQL> show con_name


You may find that useful in a multitenant environment with many Pluggable Databases within one Container Database :-)

Tagged: 12c New Features, whoami
Categories: DBA Blogs

Cool Picture of Instance/Database from 12c Concepts

Bobby Durrett's DBA Blog - Tue, 2014-01-07 18:00

I just think this is a cool picture of an Oracle 12c instance and database from Oracle’s 12c Concepts manual (not my own work):

Instance and Database from 12c Concepts

This is from the concepts manual found here: url

- Bobby

Categories: DBA Blogs

Finished reading Oracle Core

Bobby Durrett's DBA Blog - Tue, 2014-01-07 12:37

Just finished reading the book by Jonathan Lewis titled “Oracle Core: Essential Internals for DBAs and Developers“.  I think I picked it up at the Collaborate 13 conference in Denver last April but haven’t had time (or taken the time) to read it.

Reading a book like Oracle Core can be a challenge because it is pretty dense with example scripts and outputs including dumps in hex.  So, I decided to take the strategy of pushing myself to crash through the book without carefully following every example.  I may only have absorbed about 10% of the material but if I didn’t jam through it I would have gotten 0%!

I picked up Oracle Core because I had read another book by the same author titled “Cost-Based Oracle Fundamentals” which has paid for itself 100 times over in terms of helping me tune queries.  I highly recommend Cost-Based Oracle Fundamentals without reservation.  But, like Oracle Core it can be a challenge to just sit down and read it and follow every SQL example and output.  Probably it would be worth making a first pass focusing on just the English language text and skimming the examples, maybe delving into the examples of most interest.

In the case of Oracle Core I haven’t yet put it to practical use but I’m glad to have at least skimmed through it.  Now I know what’s in it and can refer back to it when needed.

Next I hope to start reading up on Oracle 12c since I plan to get my certification this year.  But, I wanted to finish Oracle Core before I moved on, even if I only read it at a high level.

- Bobby



Categories: DBA Blogs