Hi Yun,
Thanks for your reply. You mention that
" ‘snapshotState’ and ‘initializeState’ interfaces are used mainly to
snapshot and initialize for operator state"
but..."mainly" is not "exclusively" right? So, I guess my question tries to
figure out whether doing something like this is valid/makes sense?
```
override def snapshotState(context: FunctionSnapshotContext): Unit = {
if (models.nonEmpty) {
modelsBytes.clear()
for ((k, model) <- models) {
modelsBytes.put(k, model.toBytes(v))
}
}
}
```
Indeed, the above code seems to work well ... so it seems like a bug that
`clear` works sometimes but sometimes not as I noted in my reply to Congxian
and others have noted in this extended question posted in stackoverflow:
https://stackoverflow.com/questions/59123188/state-handling-on-keyedcoprocessfunction-serving-ml-models
--
Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/
