Netflix Productionizing Spark On Yarn For ETL At Petabyte Scale

[@ManuPresannakumar]

btw his name is not Aash-win, but ashwin !

[@venar303]

Thanks a lot! I particularly appreciated the explanation of parquet read/write optimizations.

[@oolongtea0922]

Nice talk！

[@anks2186]

For better visibility I suggest use of graphite and graphana sinks.. Rather than relying only on spark UI only. It's awesome. We use it and help a lot to know what's going on.