<dependency>
<groupId>somegroup</groupId>
<artifactId>someartifact</artifactId>
<version>0.1</version>
<scope>provided</scope>
</dependency>

But, when I look at the target container, I see no such library.

Upon further investigation, I realize that this scope was being used to exclude the library from the deployed artifact.  It might have been a loose dependency on another library.  Instead of using and exclusions list, to tell *me* that they just didn’t want it in the artifact, they used the provided scope to tell me that it’s already in the container.

Dear developers using maven, please choose to exclude unwanted dependencies rather than giving them a “provided” scope and sending those that follow you down this same rabbit hole.

I would prefer to see this:

<dependency>
<groupId>somegroup<groupId>
<artifactId>someartifact.thatdependson.someartifact</artifactId>
<exclusions>
<exclusion>
<groupId>somegroup</groupId>
<artifactId>someartifact</artifactId>
</exclusion>
</exclusions>
</dependency>

Thanks!

VERR_SUPLIB_WORLD_WRITABLE

I’m using OSX, I did a fair bit of googling, the solution that worked for me:

chmod o-w /Applications

Apparently, if there are multiple, different java mail versions, the system just falls down and throws the NoSuchProviderException.

My application usually includes the latest version of the javax.mail library. So I simply renamed the java-mail.jar to java-mail.jar.orig in the tomcat “common” library.  Then I restated the tomcat server.

Changing the extension makes it not get loaded automatically when the system restarts.  I like to use the .orig so I know what changes I made in case I want to revert them.

• The error from the logs pointed me in the right direction of what functionality was causing the issue.

I didn’t know exactly what was causing the issue, but I knew where to start looking.  I pulled out my favorite tool for testing web applications and web services, JMeter.  I wrote a test that would fully stress the parts of the system that might be causing the issue.  I didn’t write the code that was causing the problem.  This was actually the first time I was in this part of the system.  A good first pass at a JMeter test does CRUD+L (Create, Read, Update, Delete, List) for a user. Once I had a simple test that would perform some easy “smoke” tests, I was ready to start stressing the system.

• I’m now familiar with this part of the system and have a JMeter test for coverage of the functionality.

My test worked just fine at this point.  Still no clues.  Next step is to increase the load.  I modified my test to use a different, random user on each login attempt.  Then I increased the number of threads.  Immediately, I started to see the errors.  For most of my tests of this system, I’m able to send through 100 threads (users) without running into other bottlenecks on my local (dev) environment.  With this test, at this time, I could only get 3 successful threads.  This is unacceptable.  With just 4 users going through my test, I was creating “blocking sessions” according to the Oracle Enterprise Manager (OEM) reporting tool.

• Now I have a test that is replicating the problems (locally) that we are seeing in production

Our system is using JPA with Hibernate and annotations for most of our ORM.  When I started to dig into code, I realized that on “inserts” we are doing 3 different commits during the transactions.  In the methods, I saw multiple “commit” and “flush” calls.  Instead of letting JPA/Hibernate handle the generation of IDs for (grand)child objects, there was some additional commits that were done with native queries in order to get IDs and follow the particular pattern that was chosen.  I wasn’t happy with the pattern, but that was irrelevant to the problem at hand.  But I couldn’t understand why we needed to do all of this extra work that I haven’t done in years.

• The multiple, native commits, within the transaction were a red flag.  Clearly, the low (3 user) load threshold was caused by the poor use of the transaction.  The inserts, updates, and deletes were all doing this multiple commits and multiple flush to finish each action.

I really wanted to get away from this multiple step process.  My plan was to rewrite the POJOs with readable annotations.  This way, the rest of the team could see what was actually happening without having to dig like I did.  Quickly I realized that I couldn’t make many changes because the pattern that was chosen was used throughout the entire system.  So, I got too it.  I used annotations for all the objects that were affected.  It took a while, to replace the hand written stuff and at the same time keep the pattern in place.  Again, it was extra work, but not relevant to the problem.

• I reduced the number of required DAO objects in the system.  The entire group of objects is wired with annotations.  Its readable, cleaner, and still works with the rest of the system.

I figured everything would magically be fixed.  It wasn’t.  I used my JMeter tests and determined that the queries that were being ran against the database looked good.  Everything was in order and exactly what I expected.  But my changes had only increased the number of successful users from 3 to 5.  It also improved the performance slightly.  At the same 3 users going through the system, my changes made this section about 20% faster.  But what now?

• I pulled in the database admin.  The DBA told me that the database was doing cascading deletes on some of the tables.

We went around in circles for at least two days getting to a happy place.  Since the cascading deletes were in place, my “delete” method was changed from using the annotations to using a native query.  Luckily, this native query was just 1 call on 1 table and the oracle database did the rest.  I had to believe that letting Oracle do the deletes was a better option that letting Hibernate make all the calls.

So in the end, it turns out that the problem causing my ORA-00060, was that the code was trying to do the same deletes that the Oracle database was also trying to do.  This was causing the blocking sessions that were affecting the rest of the system.  Before you let Hibernate/JPA handle your cascade deletes, make sure that you don’t already have cascades setup at your database tier.

• The test that I wrote the first week, now handles the 100 users that I expected the system to handle.  Problem solved.

This problem wasn’t really as big as I let it become.  If I had brought in the DBA earlier, and was aware of the cascade deletes at the database tier earlier, I would have fixed the problem in the original DAO objects.  This would have fixed our problem sooner.  On the other hand, the code for this group of functionality is now much easier to read, work with and understand.  The next developer to work on this stuff should get to success much faster.

I’m actually happy that I had to get all the way to the database tier to fix this “database” problem.  I was able to learn one of the patterns used in a large part of our system.  I feel confident that the next person to look at this code will have a much more pleasant experience.  I also have a nice test that provides coverage and can be used for stress testing this part of the system.

• gd_intermediate_bundle.crt
• mydomain.crt
• mydomain.key

I modified the file /etc/httpd/conf.d/ssl.conf according to the examples.

But when I restarted the httpd server, I kept getting an error

/path/to/mydomain.crt does not exist or is empty

I’m pretty sure the path stuff can can be switched up, just be considerate of permissions.  I moved my files into the directories that were shown in the examples.  Even then I got the same errors.

After plenty of searching I figured out that it was all because of SELinux.  I did not want to disable SELinux to get this to work so here are the steps I took:

1. chcon --reference=/etc/pki/tls/private/localhost.key /etc/pki/tls/private/mydomain.key
2. chcon --reference=/etc/pki/tls/certs/localhost.crt /etc/pki/tls/certs/mydomain.crt
3. touch /etc/pki/tls/certs/server-chain.crt
4. chcon --reference=/etc/pki/tls/certs/server-chain.crt /etc/pki/tls/certs/gd_intermediate_bundle.crt
5. Restart the server in your favorite way. Ding!

I don’t really understand exactly what happened but here is my best guess.  SELinux is looking for specific filenames.  I changed the references for SELinux to know that they are legit.  The files that were referenced in the ssl.conf file as examples must be what SELinux is looking for by default.  The server-chain.crt file referenced isn’t created by the system, so I had to “touch” it.

I’m thinking that if I had just changed the filenames to match those in the ssl.conf that I would have had equal success.

I’m not a big fan of my column names repeating my tables names:

For instance an Address table with column names like address_id, address_street1, state_id.

I understand that when your doing SQL joins this allows you to avoid naming collisions.  But in my world, I’m never using the table names or column names except for in my annotations.

Following this convention actually requires me to add @Column(name = “address_street1″) to my object, instead of simply @Column

With my preferred approach, my columns match up nicely with my domain objects/data models.  I don’t want to have to add the “name” part to my @Column annotation.

Can we all agree that from now on, we write our model objects and database tables so that we don’t need the (name = “address_street1″) part?  Please?

I downloaded the Ubuntu software component from the website.

Connected my netbook to the Evo via the USB connection and mounted the EVO as a drive.  I then installed the .deb that I just downloaded from the “downloads” directory on the Evo.  I just double clicked and then was asked for admin login to finish the install.  I’m sure a command line ‘sudo deb something’ would have worked just as well.

The application asked me to change a USB setting and then took me right to the setting.  I made the change.

Then I clicked finish.

On the netbook, I had to run two commands to get it going, but the application told me exactly what to run.

Ding!  I’m connected.

It works!  No problems what so ever.  I will install the same thing on my wife’s Evo and netbook.  I will just give her a couple of icons in her menu to use instead of having her use the terminal like the software suggested.

• Downloaded sources from here.
• sudo aptitude install build-essential libssl-dev libreadline5 libreadline5-dev zlib1g zlib1g-dev sqlite3 libsqlite3-dev
• ‘unzip’ sources into a directory and ‘cd’ into that directory
• ./configure
• make
• make test (to make sure everything is working)
• sudo make install

Because Ruby 1.9.2 has Gem 1.3.7 included, I can now install rails

• sudo gem install rails
• rails new path/to/source
• cd path/to/source
• bundle install (to create sqlite3 gem)
• rails server

Bling!

http://localhost:3000

I read the manual and saw that it was a notification of hot key mode. After hitting the ESC key everything was normal.

After working through this issue I couldn’t be happier. The 4 port KVM switch fits perfectly under my mac mini. The fact that it handles audio makes it that much better.