Deployment policies THREAD_LOCAL & CONTAINER_LOCAL are strict, if it cannot be satisfied application won't be launched. Otherwise, it is a best effort to satisfy user requirement. Apex also has anti-affinity feature.
How are you planning to use Apex? On Wed, Jun 29, 2016 at 3:27 PM Junguk Cho <[email protected]> wrote: > Hi, Sandesh. > > Thank you for reply. > > One following question from second one, as far as I know we can give hints > to deployment policies, but that is just hint and APEX decides the > deployments. > Is it right? > If so, does it check current cluster states and deploy topologies? > > Thanks, > Junguk > > > 2016-06-29 16:55 GMT-04:00 Sandesh Hegde <[email protected]>: > >> 1. Not Netty but https://github.com/DataTorrent/Netlet >> 2. Operators can be deployed in the same container or in different >> containers and the containers can be in the same machine or in different >> machines. >> >> Within Container: >> Between the operators, it is either is a queue or function call. >> (CONTAINER_LOCAL, THREAD_LOCAL) >> Every container has a BufferServer, to communicate with downstream >> operators. Communication between BufferServer and operators in >> that container is local loop back.. >> >> Different Containers in the same machine -> Local loop back >> Different Containers in different machines -> All bets are off. >> >> 3. Kryo - by default. Users can use different codec for different streams. >> (take a look at StreamCodec) >> >> On Wed, Jun 29, 2016 at 12:55 PM Junguk Cho <[email protected]> wrote: >> >>> Hi, >>> >>> I have some questions for Apex architecture. >>> Does it use netty for transport layer? >>> In addition, if two operators are in different processes in one machine, >>> does they use local TCP connection? >>> >>> Also does it use kryo serialziation for object between operators? >>> >>> >>> Thanks in advance. >>> Junguk >>> >> >
