To ensure that each write is persisted to disk before returning success, performing a sync() for every small concurrent write introduces excessive overhead.
Should we consider adding a batch sync feature? For example, after a write we trigger a sync, but if there are multiple concurrent writes, we merge them according to a certain strategy and perform a single sync before returning.
To ensure that each write is persisted to disk before returning success, performing a sync() for every small concurrent write introduces excessive overhead.
Should we consider adding a batch sync feature? For example, after a write we trigger a sync, but if there are multiple concurrent writes, we merge them according to a certain strategy and perform a single sync before returning.