Constructing a workload that:

• Inserts 16MB values from 10 threads in parallel
• Configures a memory_page_max of 5MB
• Does not do checkpoints
• Does not use explicit transactions

Can lead to in-memory pages growing to over 1GB. That is bad for a number of reasons, including:

• Reconciliation will need to allocate a large amount of memory when writing the page out.
• Reconciliation will be slow, leading to application stalls.