A Single Open Read Transaction Can Silently Starve SQLite WAL Checkpoints

In SQLite's Write-Ahead Logging (WAL) mode, a single forgotten read transaction left open can prevent checkpoints from completing, causing the WAL file to grow unchecked. This happens because SQLite's WAL mechanism stops checkpointing once it reaches pages beyond an active reader's end mark, a documented but easily overlooked behavior. A controlled drill — pairing one persistent reader connection with thousands of concurrent writes — can reliably reproduce the starvation risk and validate alerting thresholds. Engineers are advised to monitor WAL growth against available disk budget rather than arbitrary file sizes, and to identify and enforce maximum transaction ages to prevent the issue. Standard health checks like SELECT 1 give no indication of checkpoint progress, making dedicated WAL monitoring and a clear operational runbook essential.