Hi, Flink serializes all user functions (including source functions) with Java Serialization to ship them to the worker processes. That's also why everything in a user function must be Serializable.
There is not an easy way to synchronize running tasks. Each task has its own function object and these might be distributed across different JVMs. So even a static field won't help here. Best, Fabian 2016-12-22 9:12 GMT+01:00 Saikat Maitra <saikat.mai...@gmail.com>: > Hello, > > I am working on building Apache Ignite connector for Apache flink. I am > currently developing the SourceFunction to consume Cache event from Ignite > cluster. > > Here is the PR https://github.com/apache/ignite/pull/870/files > > I am observing that during unit tests the IgniteSource instances are > different which is created using the IgniteSource constructor and inside > run() method. As a result when igniteSrc.cancel() is called the igniteSrc > instance is not getting stopped. > > I wanted to discuss: > > 1. If Flink create a copy of IgniteSource object when > env.addSource(igniteSrc) > is called? > > A quick work around the problem is to use static boolean stopped variable > which allows single IgniteSource instance but this limits using multiple > IgniteSource with different cache combination. > > > Regards > Saikat >
