Zhi Lin
Zhi Lin
## Why are these changes needed? It is crucial for raydp to implement fault tolerance. It solves the following issue: if an object returned by an actor needs to be...
Right now, the native spark way to configure the logging level does not work. We need to fix it. Update README if necessary.
This PR implements fault tolerance of ray dataset converted from spark dataframe using raydp. This requires spark session to be holded by an ray actor. This PR also introduces such...
Now that we have RayDPSparkMaster as an actor, we no longer need an explicit object holder. We can move its functions to spark master actor.
Raydp should be able to recover/re-calculate data lost in ray as Spark does.
⚠️This PR might have breaking change. Previously, we use Py4j to start `RayAppMaster`, and it will connect to the ray cluster. In other words, it is also a driver. Therefore,...
This PR creates a new branch to be compatible with Ray nightly, 2.0.0-dev. The ray jar version is still 1.1.0-SNAPSHOT because they didn't update the version for long, but it's...