I'm still trying to figure this one out but I'd like to know why a Data Loader extraction completely fails when invalid custom fields are added to the 'sfdc.extractionSOQL' parameter of the process-conf file. My trace output shows no errors nor provides a reason why the extraction process fails. I just wind up with a file containing zero bytes.
To answer your question:
Why does a Data Loader extraction fail when invalid custom fields are specified?
Because it is supposed to fail. And the failure message that you will receive whenever you specify an incorrect field will be as below:
No such field [field name] on entity [object name]
My trace output shows no errors nor provides a reason why the extraction process fails.
These errors are generated at the application level (in this case its the data loader). In event of any failures, the error messages are displayed on the console from where you are running the process. However, there is no option though to capture such failures by specifying any specific file in your local folder. But, you can see any such failures in a file called sdl.log which is generated by data loader in the temp directory. Refer to this link for details. Excerpt from the link below:
The log file, sdl.log, contains a detailed chronological list of Data Loader log entries......The log file can be opened with commonly available text editor programs, such as Microsoft Notepad.
If you are using Data Loader for Windows, view the log file by entering %TEMP%\sdl.log in either the Run dialog or the Windows Explorer address bar.
If you are using Data Loader for Mac OSX, view the log file by opening terminal and entering open $TMPDIR/sdl.log.