Hi Ron, The CEP library is built on top of the DataStream / ProcessFunction API and holds all necessary state (the state of the pattern matching state machine) in regular keyed MapState. Hence, CEP does not require a dedicated configuration for checkpoints and savepoints, besides the regular application checkpoint configuration.
That's also why there's no dedicated documentation about this subject. @Dawid or Klou, please correct me if I'm wrong. Best, Fabian 2018-09-17 19:09 GMT+02:00 Ron Crocker <rcroc...@newrelic.com.invalid>: > I’m working with CEP to detect when something stops reporting (which is > very simple), but I need to show the team that the jobs will survive being > shutdown and restarted without either a) declaring that everything stopped > reporting (false positives) or b) missing things that have indeed stopped > reporting (false negatives). > > There seems to be NO documentation regarding CEP and > checkpoints/savepoints. Am I just missing it? Or is it something so simple > that it should be obvious? > > Our graph is fairly straightforward - keyed stream using event time and a > Pattern that is essentially a report followed by a report within a time > window, and we use the timed-out side output as the “events” indicating > “thing stopped reporting”. It seems that we need to checkpoint/savepoint > the pattern state along with the normal things checkpointed (e.g., Kafka > offsets). > > For now, I should be able to sell an assertion from you that it works, but > official documentation would help. > > Ron > — > Ron Crocker > Distinguished Engineer & Architect > ( ( •)) New Relic > rcroc...@newrelic.com > M: +1 630 363 8835 > >
