We experienced persistent performance issues with our main MySQL database in late October/early November and late January/early February, resulting in the In-App Messenger endpoint being down for an hour each time. The problems were caused by a combination of factors including lock contention, long-running transactions, and rollbacks due to custom data processing issues. To address these issues, we made changes such as disabling transparent huge pages, tweaking MySQL tunables, and implementing caching across the application using Shopify's IdentityCache. We also built an SQL query fingerprinter to help identify new queries and trends in our database. Despite some initial challenges and red herrings, reducing load on the database and improving Redis performance helped us regain stability.