Allow Invalid Fee Payers: A key step towards enabling asynchronous execution
Auto Program Resizing: Refunds surplus rent during program upgrades
Standardizing Shreds: Protocol now enforces 32-data, 32-coding FEC sets
Inactive Stake Cleanup: Excludes inactive stake from reward calculations
Rent Reduction: Final three planned stages of the 90% reduction in state bonds
Reading from a key-value store sounds straightforward: provide a key and get a value. However, the previous article in this series explores why this thinking is wrong. LSM trees make the trade-off that writes are cheap because they defer the work of organizing data.
Reads have to pay for it.
A key’s current value might live in the memtable, a file at level 0, or any of the sorted levels below. The read path’s job is to find the newest version of this value without checking everywhere.
This article covers the machinery that keeps that search fast, and the accounting that determines what it costs.