Yesterday, our engineering team deployed an update to our platform to address several minor bugs. One of the fixes required structural changes to our main database, and an update script had to be executed to implement those changes. Unfortunately, a critical bug in the upgrade script went unnoticed during our internal reviews, resulting in some data loss. While the technical details of the bug are intricate, the outcome was mostly random data loss. This affected a small percentage of our customer base, as some segment conditions were modified for customers who had previously deleted certain custom data, particularly if it was their first custom data entry.
To remedy this, we initiated our standard data restoration process to recover the lost data. Given the complexity of the data changes, it took our team approximately 8 hours to develop functional code for data restoration. The code was then run in the staging environment before being deployed to production.
In order to prevent similar issues in the future, our team is actively developing a new process to more effectively categorize the potential impact of seemingly minor releases. We aim to flag and schedule dangerous releases appropriately.
A communication will be released in-app for all affected campaigns. If one of your campaign is affected, we recommend ignoring the date of the 21st of November by using the timerange filter in your analysis to obtain accurate insights for this particular campaign.
Once again, we apologize for the inconvenience.