Tmptrash was the safety plan.
GetGlue's web servers were accumulating logs, and files older than fourteen days needed to move into Amazon S3. I wrote a script, put it on a schedule, and gave every server a matching offsite directory.
The easy version would upload a file and delete the local copy. I did not trust the easy version yet.
For the first few days, anything that appeared to upload successfully went into a local directory called tmptrash. That gave me a place to recover it if the S3 copy was missing, incomplete, or sitting somewhere unexpected.
I planned to purge those directories by hand only after the process had run long enough to prove itself. I was also testing ways to inspect the remote object and make sure the upload was actually there before the local deletion happened.
The rollout started on two web servers. I would check the result with the person who knew the environment, then enable the rest. Database dumps were next.
None of this was technically exotic. That was the point. The scripts looked like the backup structure the team already used. Somebody else could read them, understand the filenames, and see what the automated job was doing without learning an entirely new system.
I had been at AdaptiveBlue for about a week. The job was a small consulting role layered around Peek, but the operating habits were the same ones I had built at Beatport: automate the repetitive part, keep the first rollout narrow, verify before deleting, and leave the system understandable for the next person.
The name tmptrash was not elegant. It was accurate.
The files were probably safe to delete. Probably was not enough. They could sit in the temporary pile until the copy proved itself.