spring mvc - Spark MLlib and Rest -


i working proof of concept of having spark mllib training , prediction serving exposed multiple tenants form of rest interface. did poc , running seems bit wasteful has create numerous spark contexts , jvms execute in wondering if there way around or cleaner solution having in mind spark's context per jvm restrictions.

there 2 parts it:

  1. trigger training of specified jar per tenant specific restrictions each tenant executor size etc. (this pretty out of box spark job server, sadly doesnt yet seem support oauth), there way it. part don't think it's possible share context between tenants because should able train in parallel , far know mllib context 2 training requests sequentially.

  2. this trickier , can't seem find way that, once model has been trained need load in kind of rest service , expose it. means allocating spark context per tenant, hence full jvm per tenant serving predictions, quite wasteful.

any feedback on how can possibly improved or re-architected it's bit less resource hungry, maybe there spark features i'm not aware of facilitate that. thanks!


Comments

Popular posts from this blog

get url and add instance to a model with prefilled foreign key :django admin -

css - Make div keyboard-scrollable in jQuery Mobile? -

ruby on rails - Seeing duplicate requests handled with Unicorn -