Friday, December 6, 2013

A great moment with Anna Calvi yesterday evening

I was at the Astrolabe hall in Orleans, France yesterday evening to listen Anna Calvi performance, and it was really great. With her band she played most of her first album tracks, and some of her new one :
This small slender blonde woman really rocks, and her voice is as astounding live than on CD...

Wednesday, June 12, 2013

Windows 7 free backup/restore solutions, happy end !

In my previous posts I described how I shrinked my system partition, bought a pretty 500 GB 2.5inch SATA hard drive, but then was disappointed to discover that the first backup/restore free tool I selected (redoBackup) couldn't restore to a smaller drive than the original one.

I then gave a try to the second backup/restore tool in my shortlist, Macrium Reflect Free.

Partitionning the new drive

Before starting the backup/restore stuff, I first created as many partitions as needed on my brand new drive. After shrinking the system partition on my actual drive, it was divided in 5 parts :
  1. a 18GB Packard Bell Recovery partition
  2. a 100MB "system reserved" partition
  3. the 168GB C: partition
  4. 407GB unallocated (freed by the C "shrinking")
  5. 803GB D: partition (user data)
So I used the very handy EaseUS Partition Master Free Edition to create same size #1 to #3 partitions on my new drive. I incidentally created a fourth partition with all the remaining space (about 278GB).

Creating an image of the actual drive

Installing Macrium tool was a breeze. Once started, it was very easy to use too.

I just had a few clicks on its visual UI to select which partitions I wanted to dump from my genuine drive, select a destination folder on the fourth partition of my new drive (connected through an USB2 adapter).
Then I simply asked it to start the partition snapshot, while Windows7 was running !
Less than 2 hours later I was stunned to see the image generated, its size being only 76GB.
During the processing, a progress bar showed clearly the pace.

Restoring the image on the new drive

Okay you seem smart Macrium, let's see now if you can restore this image on my smaller drive, huh ?

Once again, only a few clicks were required to select which partitions to restore from the image, and where it had to be written on the new drive. Once again less than 2 jours later it claimed the job was done. No complaints, no problems.

The final test

It was time to open the desktop and put my hands in its guts.
2 screws to remove; hopefully the drive connecting plugs are very easy to catch; just had to unplug both power supply and SATA wire from the genuine drive, plug them onto the new drive and switch on the power.

It booted without even noticing the drive has changed.
It just took a bit more longer to show up, I guess because Macrium avoids dumping pagefile and hiberfil Win7 huge files when generating the image, so boot process had to re-create them when starting the first time.

My first tests on the new drive show no problems at all so far.


Wow, Macrium really rocks !

From now on I have my handy toolset :
Now I just have to run the backup / restore above procedure (2x2 hours) from time to time to have a ready-to-use spare drive in case my current one crashes.

And I have 200GB  spare on this drive to store more stuff.
That's pretty cool.

Sunday, June 9, 2013

Windows 7 free backup/restore solutions, still far from the light...

Lately I wrote about my intention to test various free backup/restore solutions for my home computer.
I first shrinked my system partition, then did a backup of it with RedoBackup.
But I hadn't any spare hard drive so I bought a pretty 500 GB 2.5inch SATA on amazon.

I was excited when I received it last friday, and as soon as I had some spare time this week-end I tried to restore the backup image onto it.

Damnation, RedoBackup didn't even try to start the process, complaining about the destination drive being far too small. I couldn't believe it, until I (quickly) found this clear statement in its FAQ :

Q: Can I restore a backup to a smaller drive?

A: No. Doing so likely would require modifying the bootloader, which defeats the simplicity of Redo Backup. Backups must be restored to a drive of equal or greater size.
However, you can always restore to a larger drive, resize the partition, and back up the smaller partition if needed. Partitions can easily be resized using the included partition editor utility (GParted).
As my original drive is 2 TB, there is no surprise my small brand new 500 GB do not fit...

Well, forget about RedoBackup.

I dug a bit further about restoring on a smaller drive, and my first insights were not very promising. It seems that few backup/restore tools can manage such processing, stating that not only the size of the original partition matters, but also the physical location of all its bits on the drive. 

Acronis is claimed to manage this greatly but it costs 50$, so I put it apart for the moment.

Fortunately, Macrium v5 seems to cope with it too, as I read here and there.

Thus I'm gonna delete the image generated by RedoBackup, and build one with Macrium v5 Free.

Stay tuned...

Thursday, May 16, 2013

Running my WebApp locally with HSQLDB while my production target is ORACLE


In the company I work for, the standard JEE stack is quite heavyweight; commonly Websphere Application Server for the webapp container and ORACLE for the database.

I currently develop on a HP Mobile Workstation laptop running an intel i5 chip with 4 GB under Win7, and my common IDE is Eclipse (currently Indigo flavour) with a bunch of plugins, including m2e (as we work intensively with maven).

The company provides a bundle with Eclipse and JBoss, but I had many recurrent troubles with the wtp-jboss plugin which randomly published only parts of the war content, so I keep on using a small tomcat6 install which does perfectly the job.

But for the database, we used to connect to the single ORACLE development platform.
It has the obvious advantage of being the real target in terms of database, but has many drawbacks, including :

  1. All developers share the same instance, so conflicts and overlaps are frequent
  2. Some uneducated developers run data changes straightly in the database with sqldeveloper for example, without writing it first in a script file that would be stored in our SCM (currently SVN); sharing the same database does not help them to realize that such changes are (or should be) short-lived.
  3. Due to the company size, all servers are hosted on a datacenter which is quite far from our office; it's not a big deal for manual testing, but it makes a big difference when we want to run automated integration tests.
  4. When part of the team works overseas, moreover through VPN, SQLNet lags a lot.
I quickly thought that providing each developer a dedicated instance would greatly help, so I first did it by creating as many ORACLE schemas as needed. It answered #1 and partly #2 (each developer having its own schema, it's easier to get that changes are local, and won't be magically propagated to others).

To answer #3 and #4, I thought the best would be to have a local database, on the PC itself.
I first tried Oracle Express, but soon dropped it. Too heavy, not very reliable...

And I was thinking that for the automated functional tests, an in-memory lightweight database would be perfect. Which led me to HSQLDB.

It is a single jar you can draw from maven in a snap, that is very easy to start and stop in a second from a java launcher in eclipse, as well as starting in-memory from java testing code.

Of course, there are several limitations that could prevent from using effectively HSQLDB in place of ORACLE. Using triggers, pl/sql functions and procedures, materialized views, and many others.

Fortunately my current project can be done so far with only plain tables, PK, FK and sequences, so it was worth a try. (well actually the sequence stuff is strangely and actually differently handled by both HSQLDB and ORACLE, which brings some fun... more on that below).

One last constraint, for various reasons we design the physical database with UML Enterprise Architect, using it to generate the ORACLE scripts. Thus this ORACLE scripts are our "reference script set" for building the schema, from which I intended to setup the HSQLDB instance.

Let's extract the gems from the ORACLE coalmine !

Transposing tables and their PKs and FKs

I first tried to generate HSQLDB compatible DDL scripts from those targeting ORACLE, but it was not reliable.

Then I gave a try to ddlutils.

Though it has some some caveats and seems not actively maintained for years, I succeded to make it inspect our Oracle Database to generate an XML file containing an agnostic definition of all the tables and their PKs and FKs.

I then fed ddlutils with this agnostic definition and it perfectly built up the tables in the HSQLDB local database.

Unfortunately, I failed to make ddlutils get both sequences (seems not supported) and data inside the tables (it failed with a complaint about some ORACLE system tables he can't handle, and I didn't succeed to exclude them from its processing).

Transposing sequences

The simpliest to handle the sequences was actually to use the genuine Oracle SQL sequence creation scripts upon HSQLDB.
I just had to remove a few oracle specific sequence parameters (that were currently not necessary for the application).
It gave such sequence creation scripts that work on both ORACLE and HSQLDB :
create sequence MYSEQ increment by 1 start with 1 minvalue 1;

Transposing data

To cope with the data, I switched to dbUnit, and succeeded to make it inspect our Oracle Database to generate an XML file containing an agnostic definition of all the data we needed.

Show me the code, dude

Okay let's stop teasing ;)

First of all, here are the dependencies I put in my pom.xml to draw HSQLDB, ddlutils and dbUnit :




As both ddlutils and dbUnit provide several ant tasks, it came naturally to me to write the oracle to HSQLDB "translator" with ant.

As we work with Maven, I could transmit the proper classpath to ant so declaring the tasks was done without relying on ugly hardcoded paths.

In my pom.xml I added the following profile :

                <property name="classpath" refid="maven.compile.classpath" />
                <ant antfile="${basedir}/shared/${scriptName}.xml" target="${targetName}"/>
So I could easily run any ant script target with a simple maven commandline, like :
mvn.bat validate -Pantrun -DscriptName=YYY -DtargetName=XXX
Note how I passed the compile classpath computed by maven, as a property surprisingly named "classpath"...

Then the ant script was pretty straightforward :

<?xml version="1.0" encoding="UTF-8"?>
<project name="myProject-dao" default="ora2local" basedir=".">

  <property name="local.driver" value="org.hsqldb.jdbc.JDBCDriver" />
  <property name="local.url" value="jdbc:hsqldb:hsql://localhost/myLocalDB" />
  <property name="local.user" value="SA" />
  <property name="local.pass" value="" />

  <property name="ora.driver" value="oracle.jdbc.driver.OracleDriver" />
  <property name="ora.url" value="jdbc:oracle:thin:@...:1521:..." />
  <property name="ora.user" value="..." />
  <property name="ora.pass" value="..." />
  <property name="modelName" value="..." />
  <taskdef name="databaseToDdl" classpath="${classpath}"
           classname="org.apache.ddlutils.task.DatabaseToDdlTask" />
  <taskdef name="ddlToDatabase" classpath="${classpath}"
           classname="org.apache.ddlutils.task.DdlToDatabaseTask" />
  <taskdef name="dbunit" classpath="${classpath}"
           classname="org.dbunit.ant.DbUnitTask" />
  <target name="ora2xmlCommon">
    <echo message="ora.user=${ora.user}" />
    <echo message="===== Dumping Oracle schema for ${ora.user}" />
    <databaseToDdl modelname="${modelName}" schemapattern="${modelName}" verbosity="DEBUG">
      <database driverClassName="${ora.driver}" url="${ora.url}"
                 username="${ora.user}" password="${ora.pass}" />
      <writeschematofile outputfile="src/main/resources/ddlutils/schema.xml" />

    <echo message="===== Dumping Oracle data" />
    <dbunit driver="${ora.driver}" url="${ora.url}"
             userid="${ora.user}" password="${ora.pass}" schema="${modelName}">
      <export dest="src/main/resources/ddlutils/data.xml" />
At this stage, we have 2 XML files, respectively schema.xml for an agnostic representation of the tables, PKs and FKs of the schema, and data.xml for an agnostic representation of their data. Now let's have a look on the other side of the job, injecting all this stuff in our HSQLDB instance.

Let's put the gems on the HSQLDB ring !

First of all, we may already have stuff in our current HSQLDB instance, so let's drop it :
<target name="xml2local">
  <echo message="===== Dropping Local Schema" />
  <sql driver="${local.driver}" url="${local.url}"
       userid="${local.user}" password="${local.pass}"
       onerror="continue" print="true" classpath="${classpath}"
       showheaders="false" showtrailers="false">
Then we build up the schema :
  <echo message="===== Injecting Oracle Schema in Local database" />
  <ddlToDatabase verbosity="ERROR">
    <database driverClassName="${local.driver}" url="${local.url}"
              username="${local.user}" password="${local.pass}" />
    <fileset dir="src/main/resources/ddlutils">
      <include name="schema.xml" />
    <writeschematodatabase alterdatabase="true" failonerror="false" />
Afterwards we manually create the sequences :
  <echo message="===== Injecting Sequences in Local database" />
  <sql driver="${local.driver}" url="${local.url}"
       userid="${local.user}" password="${local.pass}"
       onerror="continue" print="true" classpath="${classpath}"
       showheaders="false" showtrailers="false">
    <transaction src="${sequencesPath}/mySeq.sql" />
And we finish with the data :
  <echo message="===== Injecting data in local database" />
  <dbunit driver="${local.driver}" url="${local.url}"
          userid="${local.user}" password="${local.pass}">
    <operation type="INSERT" src="src/main/resources/ddlutils/data.xml" />
Unfortunately, the last step failed miserably on integrity violation... WHAT THE HECK ??

OMG... currently one of my tables had a FK on itself, implementing a tree of nodes...
Which means that some records in the table depend on the existence of other records of the same table...
Requiring to insert the records in the "appropriate" order...
Which is quite impossible to explain to ddlutil.

If we can't comply the constraints, let's remove them ! (temporarly of course :)
Before inserting tasks, we just disable all the constraints, and afterwards, we enable them back, and that's it.

Moreover it's very straightforward with HSQLDB :

  <sql driver="${local.driver}" url="${local.url}"
       userid="${local.user}" password="${local.pass}"
       onerror="continue" print="true" classpath="${classpath}"
       showheaders="false" showtrailers="false">
  <sql driver="${local.driver}" url="${local.url}"
       userid="${local.user}" password="${local.pass}"
       onerror="continue" print="true" classpath="${classpath}"
       showheaders="false" showtrailers="false">
At this stage, I was able to start my webapp locally, targeting my local HSQLDB, and everything went smooth in my first tests... until I tried to create a new record. BANG, constraint violation. WHAT THE HECK (again) ??

You got an idea ? no ?

Ok, here's a hint :
When data was inserted in ORACLE, the sequences on which some PKs rely got incremented, right ?

You've got it :
The sequences in HSQLDB should be set with the same values as those in ORACLE, but the above process do not handle that.

I could have get the value of each from ORACLE, but as I knew which PK was bound to which sequence, I found better to compute the appropriate value out of the actual dataset, as demonstrates the following snippet (thanks to which helped me a lot) :

  <sql driver="${local.driver}" url="${local.url}"
       userid="${local.user}" password="${local.pass}"
       onerror="continue" print="yes" classpath="${classpath}"
       showheaders="false" showtrailers="false"
    select 'mySeq=' || (max(id_ofMyTable)+1) from myTable;
  <property file="" />
  <delete file="" />
  <echo message="got mySeq=${mySeq}" />
  <sql driver="${local.driver}" url="${local.url}"
       userid="${local.user}" password="${local.pass}"
       onerror="continue" print="yes" classpath="${classpath}"
       showheaders="false" showtrailers="false"
    alter sequence mySeq restart with ${mySeq};
This time we're really done. Have fun !