Before, the finally block would always call _insert_done(), which writes out the vdb_* and kv_store_* files ... even if there was nothing to insert (because all docs had already been inserted). This was causing the speed of skippable inserts to become very slow as the graph grew.
LarFii
commented
1 week ago
Before, the
finallyblock would always call_insert_done(), which writes out thevdb_*andkv_store_*files ... even if there was nothing to insert (because all docs had already been inserted). This was causing the speed of skippable inserts to become very slow as the graph grew.