November 5, 2011 1 Comment
You know that sometimes you are requested to work on weird situations with crazy errors messages, no clue on what is the actual problem and pretty simple solutions (after you finally discover what is causing the error). It can happen to all of us IT people and this happened to me a few weeks ago.
We were running an .NET application to access an Oracle database. All of a sudden it started to throw the following error message:
System.Data.OracleClient requires Oracle client software version 8.1.7 or greater.
This is a very deceiving error message. It makes you believe you don’t have to Oracle client installed. We are using Oracle client 9i and we detected the same error both on the server and on the developer workstation.
If you search on Google this error message you’ll see a bunch of related articles.
It seems that the main cause of this error message is lack of permissions for the Authenticated Users group on ORACLE_HOME directory. On example of a solution for this issue can be found here. This article describes how to change the permissions on the ORACLE_HOME folder to allow the authenticated users to access it.
The first thing we tried was to grant the correct permissions to the authenticated users group. It didn’t work. We also tried to go to the machines and check for more than one Oracle client installations. We didn’t find it.
Another attempt to debug this issue was to check all the references on the Visual Studio project. I couldn’t find an issue. The reference was pointing to the correct version of the System.Data.OracleClient.dll.
Other step was to make sure the application pool was running in 32 bits since the Oracle client used is a 32 bit version. You just need to go to IIS Manager, select the application pool and edit advanced settings. Change the ‘Enable 32-bit Applications’ property to True.
Then we tried to copy part of the code to a new simple console application and it worked just fine. It was kind of weird. So we thought about checking the Oracle client DLL loading process. Using Processor Monitor to check how the application loaded the Oracle client we could have an idea on how to loading process worked. A good way to do it is using the filter on Processor Monitor to check for the Oracle client library (the oci.dll). The filter used in this case was path ending with “oci.dll”.
First, download and unzip Process Monitor. When you open it, it will start showing all information about all processes running on your computer, so it will lots of entries.
So, you should stop it while you configure the filters. Go to the File menu and uncheck the Capture Events option.
The go to the filter menu and choose Filter or click on the filter button on the toolbar.
Enable the event capture again, run the application and check out where it is looking for the oci.dll. Using this filter we could see that the application was trying to load the oci.dll from a different path than the ORACLE_HOME directory. In our case, you can see by the image, it was looking for it inside the GAC folder where the System.Data.OracleClient is installed and the path defined on the operating system:
- And then a strange folder which doesn’t exist on the server: C:\[some username]\product\instantclient_11_2
This last folder was very weird and I was expecting the ORACLE_HOME folder to be on this list instead. So, the only explanation for that was: the code. Checking the code we found out it was setting the ORACLE_HOME environment variables and that was the cause of the error. Removing those lines of codes solved our issue. It was done by a developer not involved on this debugging process….so it took a while to figure it out 😦
I have a recommendation for this case if you are a developer: DON’T SET ENVIRONMENT VARIABLES INSIDE YOUR CODE!!!! It causes the debug process to be very long and painful if the developer who did it is not involved.
If you really need to do it, think about using configuration files (such as the web.config) or using conditional compilation statements and create specific build configurations so your developers can compile the code to local versions when they need to.
To demonstrate the conditional compilation feature I created a Visual Studio solution. The first task you have to perform is to create a new configuration on the Visual Studio configuration manager (go to Build > Configuration Manager).
Then create a new configuration, give it a name and select which existing configuration should be used as a template for this new one.
Go to the project’s properties and on the Build tab, select the configuration you created. On the Conditional compilation symbols field specify the conditional constants you need for your code separated by space.
So, if you know what is causing the error is pretty simple to solve it, but with non-explanatory error messages you might have a hard time doing it.
This post is dedicated to @tucaz. As they say in Brazil: “It can happen in the best families.” lol… 🙂