This section provides a reference list of performance gains of partial updates.
Partial updates do not require periodically running baselines for performance improvements. Update operations done through multiple partial updates do not require running a periodic baseline update due to performance concerns or memory-use constraints.
After you run the first few partial updates, MDEX Engine query performance decreases slightly. After this initial performance decrease, query performance stabilizes.
Startup performance after partial updates decreases slightly and stabilizes afterwards. Overall, startup time is roughly proportional only to the total size of the MDEX Engine index, regardless of how many updates played a role in its state.
Partial updates with high turnover and high frequency perform fast. High turnover means that a large portion of the data is being updated or deleted. Any mix of add, delete, and update operations on a large number of records is handled gracefully during partial updates.
In addition, you can combine record updates into larger batches. Running such large-batch partial updates results in better overall throughput for the MDEX Engine. (The overall downtime for running one specific partial update with high data turnover may be longer, but in total, the time it takes to run one large-batch partial update is shorter compared with running many smaller scale partial updates in previous releases.)
The MDEX Engine is stable in the face of hardware crashes. A power failure of the MDEX Engine server does not affect the state of indexed data. It leaves indexed data on disk in a consistent state no matter at which point in time a crash occurs. If a crash occurs during a partial update, the files from the
dgraph_input/updatesdirectory are not deleted. After a restart, the MDEX Engine checks thedgraph_input/updatesdirectory for the presence of any files that were not applied and applies them.
