Death by Consumption
thedailywtf_com

Tryton Party Module Address Database Diagram

The task was simple: change an AMQ consumer to insert data into a new Oracle database instead of an old MS-SQL database. It sounded like the perfect task for the new intern, Rodger; Rodger was fresh out of a boot camp and ready for the real world, if he could only get a little experience under his belt. The kid was bright as they came, but boot camp only does so much, after all.

But there are always complications. The existing service was installed on the old app servers that weren't setup to work with the new corporate app deployment tool. The fix? To uninstall the service on the old app servers and install it on the new ones. Okay, simple enough, if not well suited to the intern.

Rodger got permissions to set up the service on his local machine so he could test his install scripts, and a senior engineer got an uninstall script working as well, so they could seamlessly switch over to the new machines. They flipped the service; deployment day came, and everything went smoothly. The business kicked off their process, the consumer service picked up their message and inserted data correctly to the new database.

The next week, the business kicked off their process again. After the weekend, the owners of the old database realized that the data was inserted into the old database and not the new database. They promptly asked how this had happened. Rodger and his senior engineer friend checked the queue; it correctly had two consumers set up, pointing at the new database. Just to be sure, they also checked the old servers to make sure the service was correctly uninstalled and removed by tech services. All clear.

Hours later, the senior engineer refreshed the queue monitor and saw the queue now had three consumers despite the new setup having only two servers. But how? They checked all three servers—two new and one old—and found no sign of a rogue process.

By that point, Rodger was online for his shift, so the senior engineer headed over to talk to him. "Say, Rodger, any chance one of your installs duplicated itself or inserted itself twice into the consumer list?"

"No way!" Rodger replied. "Here, look, you can see my script, I'll run it again locally to show you."

Running it locally ... with dawning horror, the senior engineer realized what had happened. Roger had the install script, but not the uninstall—meaning he had a copy still running on his local developer laptop, connected to the production queue, but with the old config for some reason. Every time he turned on his computer, hey presto, the service started up.

The moral of the story: always give the intern the destructive task, not the constructive one. That can't go wrong, right?

[Advertisement] Forget logs. Next time you're struggling to replicate error, crash and performance issues in your apps - Think Raygun! Installs in minutes. Learn more.