Hello,
I pushed and deployed several patches that: - update metrics in a single transaction - register builds in a single transaction - use a single write database worker, queuing queries and submitting them by batches (in a single transaction). - optimize some SQLite parameters (decrease WAL size, use more RAM caching) this should improve the situation, even if I still observe some inconsistent execution duration. I still have a few improvement ideas such as pre-allocating a large database file to decrease fragmentation and running periodic vacuums. Thanks, Mathieu