What are the advantages of siddhi over flink-cep

[shidayang]

I don't konw how to chose siddhi and flink-cep

[haoch]

Some advantages:

• flink siddhi have richer CEP features than flink-cep.
• flink siddhi support flexible SQL language than flink-cep only supports Java DSL.
• flink siddhi support dynamic cep runtime (modify ad-hoc without restart)

[haoch]

@shidayang What's your use case?

[panenming]

dou you have example for (flink siddhi support dynamic cep runtime (modify ad-hoc without restart)) and can one stream apply to more than one dynamic ceps?

[haoch]

@panenming please have a look about https://github.com/haoch/flink-siddhi/blob/master/core/src/test/java/org/apache/flink/streaming/siddhi/SiddhiCEPITCase.java#L441 basically idea is

you could send control event (Add/Deleted/Update) through ControlStream which physically could connect to some Kafka topic or other source and just send the control operation you want and it will apply to the DAG cep runtime dynamically.

Please refer to https://github.com/haoch/flink-siddhi/blob/master/experimental/src/main/scala/com/github/haoch/experimental/CEPPipeline.scala as a complete example

[shidayang]

@haoch thanks，I also have a question. can siddhi resolve unsort event?

[haoch]

@shidayang it doesn't have the concept of watermark in flink for out of order event, but it supports to keep some safe window for resorting events then process.

[GithubZhitao]

Hi @haoch. Although I haven't know all about siddhi, I am really absorbed by it.
I have two questions:

1. "flink siddhi support dynamic cep runtime" : Did you mean that one FLINK Job can run multi cep patterns distributed ?
2. siddhi also has its own distribution plugin(https://github.com/siddhi-io/distribution). After I read the architecture document, I didn't find any description about the "exactly-once" or "watermark " or "event time, process time, session time" concepts. Can I say that FLINK did more than siddhi-distribution did?