OryxProject / oryx

Oryx 2: Lambda architecture on Apache Spark, Apache Kafka for real-time large scale machine learning

Home Page:http://oryx.io

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Serving getYTYSolver bottlenecks in some contexts

srowen opened this issue · comments

Two related issues:

getYTYSolver can be called by two threads at once. It's OK for correctness because one result will just be discarded anyway but it's wasteful ... and if a load of requests hit at once this could result in lots of unneeded load.

Every item update causes a recomputation. This is for the moment going to stay as-is for correctness. However it also causes any subsequent requests to block until a new solver is ready. Instead they could continue with the last one while a new one is computed asynchronously.