hivemind
hivemind copied to clipboard
partial flushing queued objects
To reduce the amount of repeated work (and to even out the work done per block) we can update certain levels of the cached_post buffer over a n-block period. This is already done for accounts (p=8), and the basic mechanism is implemented in CachedPost. It's currently disabled because (1) it needs more testing, and (2) one of the main uses is payout sweeping, which requires an in-memory buffer to be effective. The same pattern can be used for sweeping accounts.