Architecture: How to use Spark ML predictions as HTTP service -
i have spark streaming application trains model , periodically stores model hfs. in http based web service, post values , retrieve prediction it. service should reload model on demand (e.g. via request).
i implemented web server spark , spray, works proof-of-concept. i'm not sure if design solution. providing web server external services if runs on cluster? how can define on node service available? i'm not sure if right idea use prediction models in way. maybe best-practice integrate spark in standalone application , access model on shared filesystem (e.g. hfs), lack cluster support, wont't it?
summary: best-practice design build prediction web service apache spark.
Comments
Post a Comment