You got the point about firewall rules and the complexity of setting individual websocket connections for each interpreter.
And yes, the browser will probably die way before you push more than 10Mb of data through the stream. Web front-end is not Big Data :) On Wed, Jan 27, 2016 at 10:46 PM, moon soo Lee <m...@apache.org> wrote: > Hi, > > I'm not sure it's technically possible to make front-end connect to > individual interpreter process against cross-domain restrictions of web > browser. > > Also, i think managing firewalls, proxy a single port for a communication > between web-browser and Zeppelin is much simpler than each interpreter > process has own port that web-browser need to connect through firewall / > proxy. > > Regarding overhead of Zeppelin server, i guess normally, before Zeppelin > server get congested, web-browser will freeze or crash from streamed data. > > Thanks, > moon > > On Thu, Jan 28, 2016 at 2:50 AM DuyHai Doan <doanduy...@gmail.com> wrote: > > > Hello > > > > I did not have a chance to look at the code of the PR > > https://github.com/apache/incubator-zeppelin/pull/611 at the time it was > > developed so I ask the question here. > > > > Looking at the source code, it seems that the remote interpreter streams > > data back to ZeppelinServer, which then forward the data to the front-end > > web page. Is it correct ? > > > > More importantly, the question I'm asking is why don't we stream data > > directly from the remote interpreter process to the front-end itself > > instead of going through the main Zeppelin server process ? The Zeppelin > > server become a SPOF and if there are a lot of data to be stream, ONE > > interpreter can overload the server quickly and impact other users > > > > Of course, this will require setting up websocket infrastructure for > each > > remote interpreter. > > > > What you you think guys ? > > >
