Thanks Congxian. From what I've read, it seems that using the keyed state in
`snapshotState` is incorrect...what confuses me is that if I do something
like this
```
override def snapshotState(context: FunctionSnapshotContext): Unit = {
if (models.nonEmpty) {
modelsBytes.clear() // This raises an exception when there is no
active key set
for ((k, model) <- models) {
modelsBytes.put(k, model.toBytes(v))
}
}
}
```
Then, when there is data (`models` is populated within `processElement1`),
the `clear` and subsequent calls to `put` work just fine. This seems like a
bug to me, as others have pointed out in this somehow extended question
posted in stackoverflow:
https://stackoverflow.com/questions/59123188/state-handling-on-keyedcoprocessfunction-serving-ml-models
Do you think the fact that `clear` works within `snapshotState` under
certain circumstances is indeed a bug?
--
Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/
