pink cylinders

More than Four

...there's an axis for that.

Upgrading site from BlogEngine 2.5 to 3.3

I though this was going to be a nightmare, but getting the content moved from 2.5 to 3.3 was the easiest part, using the BlogML export and import. The tricky part was adapting the Standard theme to match my previously customized one. A straight copy of the theme folder did not work.

This will do for now.

Flyway DB

For years I've used a database build script framework written in-house to recompile all schema objects and run unit tests. However, the actual DDL code to migrate a schema from one version to the next was always left as a hand-rolled script that is executed up-front before the normal build process.

A colleague introduced me to Flyway DB and I am now a fan. The introduction, documentation, and tutorial material is top-notch.

Check it out here: https://flywaydb.org/

Although it has an API for integration, I specifically like the command-line utility version:  https://flywaydb.org/getstarted/firststeps/commandline

Recommended.

Testing using SQL and common table expressions

Sometimes I forget how to do simple things, like splitting a string into sections based on a delimiter.

Here's a nice way to test your code on a range of input values, and see the interim results, in one hit:

with 
myobject as (
   -- put in a range of test values:
   select 'dbo.MyFunction' as NAME
   union
   select 'setup.ins_Contact'
   union
   select 'MyOtherFunction'
),
more2 as (
   -- calculate interesting attributes:
   select
      NAME,
      len( NAME )            as LENGTH_,
      charindex( '.', NAME ) as POSITION
   from myobject
)
-- perform the test:
select 
   NAME, POSITION, LENGTH_,
   case when POSITION=0 then 'dbo'
                        else substring( NAME, 1, POSITION-1 ) 
                        end as SCHEMA_,
   substring( NAME, POSITION+1 , (LENGTH_ - POSITION ) ) as OBJECT_
from more2;
;

iMac; Linux; GRUB; OS X: Can't we all just get along?

I'm not really a fan of OS X. I'm sure the underlying OS is fine - after all, It's a Unix system. I know this! - but the Finder; the keyboard layout; the relentless updates; but most of all the fuzzy text aggravated me.

After some success with a Bootable USB drive containing a 64-bit Linux Mint ISO, I decided to follow Clem's excellent instructions here: 

http://community.linuxmint.com/tutorial/view/1643

and install Linux Mint 17.1 on my 2013 iMac.

Now, I'm not crazy: I resized the primary OS X partition and divided up the remaining space, just as Clem recommended:

/dev/sda4    30 GB   ext4
/dev/sda5     4 GB   swap
/dev/sda6   500 GB   ext4 

After the Linux install completed, I also followed the "Fixing the boot order" instructions, about installing efibootmgr and making the EFI boot Linux first. Supposedly this was to achieve:

"The boot order should now indicate that it will run Mint first, and if that ever came to fail.. it would then run Mac OS. In other words our MacBook now boots into Grub. From there we can select Mint or press Escape and type "exit" to boot into Mac (we'll fix the Mac grub entries to make it exit without having to type anything later on in this tutorial)."

I'm not sure what I did wrong... maybe nothing. Maybe we're just expected to have a complete understanding of EFI and GRUB and bootloader configuration in general, at this point.

Anyway, bottom line: I never saw GRUB, or a menu of boot options, or anything. It just booted straight into Linux. Fortunately the Mac HD volume is visible and accessible under Linux, and I've been able to copy my files over as the need arises, from inside Linux Mint. Excellent.

But what if I want to execute a Mac OS X application natively, for some reason? Dammit, I want to dual boot this thing. The normal Option- key (Mac Boot menu) no longer works, and neither does Option-R (to get to the Mac Recovery partition).

After some research, I learned that pressing 'c' during the boot process gets us to the GRUB command line. Now I could use the ls -l command to list the partitions:

hd2,gpt1      fat efi
hd2,gpt2      hfsplus   'Mac HD'
hd2,gpt3      hfsplus   'Recovery HD'
hd2,gpt4      ext *                       29 GiB
hd2,gpt5      -                            5 GiB
hd2,gpt6      ext *                      438 GiB

but this still left me with no understand of how to boot into one of them. More research:

http://askubuntu.com/questions/16042/how-to-get-to-the-grub-menu-at-boot-time

OK, so I learn about GRUB 2 and the significance of the /etc/grub.d scripts and the /etc/default/grub file, and commenting out the GRUB_HIDDEN_TIMEOUT=0 setting by prefixing it with '#'.
 
Dedoimedo has a really nice tutorial on GRUB here:
 
That taught me how any changes to these grub config files should be followed up with:
 
$ sudo update-grub
 
After that, when I restart, I get the boot menu! Yay. From the partition list (see above) I'm pretty sure that I should be able to add a menu option for the OS X partition.
 
More research:

http://askubuntu.com/questions/179689/mac-os-x-wont-boot-from-grub-menu-in-ubuntu-precise-on-apple-mba5-2

It suggests adding the following to the /etc/grub.d/40_custom script:

   menuentry "OS X" {
insmod hfsplus
set root=(hd1,gpt2)
chainloader /System/Library/CoreServices/boot.efi
}

Long story short: that works perfectly. After sudo update-grub and a reboot, we have a menu and an option that will boot into OS X.

Thanks, Internet!

Installing VirtualBox 4.3 on Windows 7-64 - Solved

I was preparing for an Oracle online workshop on Database 12c multi-tenancy, and as part of the prep, you get to download a VM image with the lab environment. Of course, being Oracle, it is a Virtual Box vm.

Now, I like Virtual Box. It works great on my Mac at home, but for some reason I've had problems installing 4.x on my work laptop (Lenovo W520, getting on a bit now but still recommended).

The problem is Virtual Box almost completely installs - until it gets a certain point, then the progress bar runs backwards and it says "rolling back install", followed by:

"Oracle VM VirtualBox 4.3.20 Setup Wizard ended prematurely" because of "an error". This is followed up by "Installation failed! Fatal error during installation.".

No other information provided. Running as Adminstrator after a cold reboot didn't help.

Well... I really wanted to have the Workshop lab environment available, so despite the fact that last time I encountered this problem I didn't find any useful help on the web, I tried again... and this time, I found something on the forums, a post by user Tronmech:

https://forums.virtualbox.org/viewtopic.php?f=6&t=61785

I'll repeat the instructions here in case for some reason that post vanishes:

  • Go into the registry at: HKEY_LOCAL_MACHINE\SYSTEM\ControlSet001\Control\Network
  • Change "MaxNumFilters" from 8 to 20 (decimal)
  • Reboot
  • Install 4.3.x (run installer as administrator, just in case)

This time, the install was successful. Thanks, Tronmech!

After some more research, it seems that Windows 7 has a maximum hard-coded limit of 14 network filter drivers... and for some reason, this setting in the registry reduces it further to 8, by default.

So setting the MaxNumFilters key to 20 probably only lifts the artificial limit and allows 14 possible filters. Applications such as Virtual Machine managers and VPN clients need to add filters to the network stack, and increasing this limit in the registry seems to be the recommended fix.

In one KB article, I read that a Windows Service Pack could not be installed until some filters were removed!

To view currently deployed Network Filter Drivers, right-click on your connection widget in the Network Connections control panel applet, and view properties:

I believe that each of those checkboxed items is a "Network Filter Driver". I have 9 of them, and I assume that the "VirtualBox Bridged Networking" driver was the 9th one added after I lifted the limit of 8.

Fuzzy Fonts in OS X web browsers

My better half and I recently upgraded our personal computers from aging Sony Vaio laptops, to twin 2013 iMacs. We've wanted to dip our feet in the OS X waters for some time now, and this seemed to be a good opportunity.

I'd always supposed Apple Macs to be a paragon of typography perfection, so imagine my surprise when I discovered the great "fuzzy font" controversy surrounding recent OS X releases.

There's a lot been written about this elsewhere so I won't repeat it:

Anyway, the problem seems to be particularly noticeable when light text on a dark background is used. Unfortunately, this directly affects my Prodigal Sounds family of web sites. 

Thanks to a post by Tim Van Damme, I found a work-around: to add a block to the web site's CSS stylesheet:

html {
-webkit-font-smoothing: antialiased;
}

This seems to force the browser (at least, Safari and Chrome) to work as though the LCD font smoothing option was turned off, and the text is more readable for those browsing on Macs, even if they have the LCD font smoothing option enabled.

If you're on a Mac, you can see the results here at www.prodigalsounds.com.

Implementing a Proper() case function in PL/SQL

I found that I needed to convert some raw column names (i.e. THIS_IS_MY_COLUMN) into a human-readable "Proper" or "Title" case form, e.g. This Is My Column.

Scouring the web, I didn't find out I could steal, so here's my implementation:

   function Proper(
p_string varchar2
) return varchar2
is
l_string varchar2(200);
l_proper varchar2(200);
begin
l_proper := '';
-- change any underscores to spaces:
l_string := replace(lower(p_string), '_',' ' );

for i in 1..length(l_string)
loop
-- obviously the first character is UPPER:
if i =1 then
l_proper := l_proper || upper( substr( l_string, i, 1 ) );
else
-- if the character is preceded by a space, UPPER it:
if substr( l_string, i-1, 1 ) = ' ' then
l_proper := l_proper || upper( substr( l_string, i, 1 ) );
else
l_proper := l_proper || substr( l_string, i, 1 );
end if;
end if;
end loop;
return l_proper;
   end;
 

From the "Look at me, I'm an idiot" department

If this was April 1st, I'd be sure that this was satire, but as it is, I'm not sure:

Java SE - Change in Version Numbering scheme in March 2013

In case this vanishes off the web, here's the gist:

Seriously? I don't know who is to blame for this loss of sanity. Oracle, you are not so damn important that you need to re-invent this. My understanding is that there is a well-established convention for numbering incremental versions of a product, that allows for such concepts as minor feature releases and bug fix releases.

The scheme is nicely described here on Stack Overflow: How to do version numbers.

What is sad is that back in the Sun days, Java release did follow this numbering convention.

Exploring AUTHID with an eye to partitioning client data

Problem

Is it possible to design a multi-tenant architecture in an Oracle DB instance, such that different customers' data is transparently partitioned whilst keeping common code and common tables in a single location?

Possible Strategy

Consider the following architecture:

 

Here we have two schemas in an Oracle database instance:

  • U_Common contains common tables and stored procedures, packages, etc.
  • U_Cust_1 contains customer-specific tables, with private data. 
  • Client processes acting on behalf of Customer_1 are connecting to the U_Cust_1 schema. 

(You may assume we also have U_Cust_2, etc.)

Assume that all access is performed through stored procedures or similar code objects.
In order for sessions connecting to U_Cust_1 to execute specific procedures in U_Common, they need:

  • EXECUTE permissions on the objects in U_Common;
  • Either:
    o    explicit namespace prefixes, e.g. U_Common.DoSomething()
    o    a synonym object referencing the code object in U_Common 

In order to have common code objects that reference both common tables and customer tables, we need to create "template" versions of the tables in the U_Common schema:

 

However, when U_Cust_1 invokes the procedure in U_Common, the session can't "see" the customer-specific version of the table, because the code is executing in the context of U_Common. It will see the template table data (or lack of it - presumably there are no rows in the template tables).

This can be addressed by adjusting the AUTHID pragma when the code objects are defined:

Default:

  create or replace package U_Common [authid definer] is ..

Change to:

  create or replace package U_Common authid current_user is ..

According to the documentation, this AUTHID clause instructs Oracle on whether to run the code with the invoker's rights, or the owner's rights. (the default is DEFINER, or owner.)

 

In practice, this means that when U_Cust_1 executes the code object defined as AUTHID CURRENT_USER,  the customer tables in U_Cust_1 are in scope, and are accessed instead of the template tables in the U_Common schema.

Two interesting observations about executing code defined as AUTHID CURRENT_USER:

  • the code does NOT see the Common tables. Any table in U_Common is "out of scope";
  • the code can see other procedures and packages in U_Common, including those defined as AUTHID DEFINER (default);
  • when the code makes a nested call to those other procedures, the tables in U_Common are now visible!

 Here's a sequence diagram of the call stack:

 

This means that everything will work just fine, so long as the code objects are classified into "programs that access common data" and "programs that access customer data".

Concerning Views

Views don’t work so well. 

You can define a view in U_Common that joins a Common table with a Customer table, but any U_Common procedure that is authid current_user will not be able to "see" the view from the context of U_Cust_1.

You can try this:

grant SELECT on View to U_Cust_1
create synonym in U_Cust_1 for the U_Common.View

But the result set will be entirely pulled from the U_Common schema tables.

So this technique probably isn't going to be used to create a transparently customer-partitioned application architecture. But it is still worth understanding how AUTHID works.