Most Treeherder data types (other than pushlog, for which we decided to keep indefinitely) are covered by the cycle_data() task, which removes data older than 4 months: https://github.com/mozilla/treeherder/blob/master/treeherder/model/management/commands/cycle_data.py The initial SETA landing didn't add any handling there, so I don't think any data expiration is taking place. For example if preseed.json is modified, then we only add new rows and don't clean up old ones (either in load_preseed or cycle_data).

FYI the dataset does not grow very much over time. What would it take to deal with this? Also, sometime this year we might tackle the project of making TH manifest based rather than job and might make the current SETA implementation to need a refactoring. Would you still want to invest in this knowing that? Happy it gets taken care of!

emorley, we have discussed how SETA works and we can see that preseed could leave some lingering data but that's about it. We are OK with just deleting anything older than 4 months.