Sure Jongyoul, go ahead.
Jongyoul Lee <jongy...@gmail.com>于2017年5月13日周六 上午10:19写道： > I have tried to remove it for a while but someone claims he has used it. > Make it again and try to review it! It can also remove some logic in > interpreterFactory, too. May I handle it? > > On Thu, May 11, 2017 at 7:53 AM, Jeff Zhang <zjf...@gmail.com> wrote: > > > ClassloaderInterpreter could also be removed. > > > > > > Jeff Zhang <zjf...@gmail.com>于2017年5月10日周三 下午3:39写道： > > > > > > > > Hi moon > > > > > > Considering we use remote=true for interpreters now, and seems no user > > > complain about that. So I think we could remove it completely. Agree > with > > > you that It could make code simpler, e.g. We don't need to check to use > > > RemoteAngularObjectRegistry or AngularObjectRegistry. > > > > > > > > > https://github.com/apache/zeppelin/blob/master/zeppelin- > > server/src/main/java/org/apache/zeppelin/socket/NotebookServer.java#L1458 > > > > > > > > > > > > Jeff Zhang <zjf...@gmail.com>于2017年5月8日周一 下午6:15写道： > > > > > >> Although it can bring some benefits, but there are 2 disadvantages for > > >> remote=false as I can think of > > >> > > >> 1. The interpreter log would mix with zeppelin server log. This may > > cause > > >> diagnosing difficult, markdown might be OK, but I am afraid it would > be > > a > > >> problem for jdbc interpreter. > > >> 2. Extra memory consumption. This might be a problem for jdbc > > interpreter > > >> when it gets large amount of data from database. > > >> > > >> > > >> > > >> moon soo Lee <m...@apache.org>于2017年5月8日周一 下午5:27写道： > > >> > > >>> The option is legacy and being used in few unit tests as far as i > > >>> remember. > > >>> > > >>> I think we can either try completely remove this code (to keep code > > base > > >>> simple) or we can try re-introduce this feature (to optimize resource > > >>> usage, less restriction on data sharing between interpreters). > > >>> > > >>> I think interpreters like markdown, jdbc will have no big problem > with > > >>> remote=false while Spark interpreter require separate JVM and > > remote=true > > >>> is required. > > >>> > > >>> remote=false will provide some chances to save some system resource > > >>> because > > >>> it does not run separate JVM process for interpreter. > > >>> > > >>> Also communication between interpreter (through ResourcePool) can be > > much > > >>> faster and have less restricted on object serialization. > > >>> > > >>> So remote=false option definitely provides some advantages. > > >>> However, keeping code base for remote=false and remote=true > introduces > > >>> more > > >>> work. > > >>> > > >>> What do you think? > > >>> > > >>> Thanks, > > >>> moon > > >>> > > >>> On Mon, May 8, 2017 at 8:13 PM Jeff Zhang <zjf...@gmail.com> wrote: > > >>> > > >>> > I see all the interpreter setting set remote of interpreter option > as > > >>> true, > > >>> > and didn't find any false case in interpreter-setting.json, does > > anyone > > >>> > know what is the use case of remote of interpreter option is false. > > Is > > >>> this > > >>> > a legacy mode that should be removed ? > > >>> > > > >>> > > > >>> > > > >>> https://github.com/apache/zeppelin/blob/master/zeppelin- > > zengine/src/main/java/org/apache/zeppelin/interpreter/ > > InterpreterSettingManager.java#L183 > > >>> > > > >>> > > >> > > > > > > -- > 이종열, Jongyoul Lee, 李宗烈 > http://madeng.net >
