Avoid repeated term serialization
jacoscaz opened this issue · comments
We can improve write performance levels by avoiding duplicate and unnecessary term serializations. Terms can simply be serialized once and their serialized versions can then be concatenated around based on each index's permutation of terms. Whether the ROI of doing this is high enough to justify the effort is TBD.
In-depth profiling on a private instance working with roughly 250k quads indicates that this might potentially accelerate ingestion by 10 - 15%.
closed via 5a39a65