WT is exposing a cursor configuration such that an application can communicate it's intending to read the data only once. Given that knowledge, WT will change its caching behavior of data paged in via that cursor. The expected use-case is for this behavior is when MongoDB knows a client is performing a table scan for initial sync. It may also be useful for oplog reads when the reader is behind and has to start reading a large amount of data.
It's not expected that the storage glue layer will have enough information to know when opening a cursor whether it should opt-in to this behavior. This ticket is to expose a public API in the storage layer to allow higher levels declare their "read once" cases.
- depends on
-
WT-4090 Low priority reads
- Closed
- is depended on by
-
SERVER-36068 Expose a user-accessible cursor option to avoid caching data from reads
- Closed