[
https://issues.apache.org/jira/browse/FLINK-25424?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Zhipeng Zhang updated FLINK-25424:
----------------------------------
Description:
Flink DataStream runtime does not support checkpointing for operators that
implements `InputSelectable` interface.
It throws an UnsupportedOperationExcetpion when checking the StreamJobGraph —
"Checkpointing is currently not supported for operators that implement
InputSelectable".
The machine learning use case is as follows:
We have a two-input operator A. The first input of A is the model parameters,
the second input of A is the training data.
In this case, we have to read the model parameters first, then we read the
training data. That is we need to finish consuming the first input before we
consume the second.
was:
Flink DataStream runtime does not support checkpointing for operators that
implements `InputSelectable` interface.
It throws an UnsupportedOperationExcetpion when checking the StreamJobGraph —
"Checkpointing is currently not supported for operators that implement
InputSelectable".
The use case is as follows:
We have a two-input operator A. The first input of A is the model parameters,
the second input of A is the training data.
In this case, we have to read the model parameters first, then we read the
training data. That is we need to finish consuming the first input before we
consume the second.
> Checkpointing is currently not supported for operators that implement
> InputSelectable
> -------------------------------------------------------------------------------------
>
> Key: FLINK-25424
> URL: https://issues.apache.org/jira/browse/FLINK-25424
> Project: Flink
> Issue Type: New Feature
> Components: Runtime / Checkpointing
> Reporter: Zhipeng Zhang
> Priority: Major
>
> Flink DataStream runtime does not support checkpointing for operators that
> implements `InputSelectable` interface.
> It throws an UnsupportedOperationExcetpion when checking the StreamJobGraph —
> "Checkpointing is currently not supported for operators that implement
> InputSelectable".
>
> The machine learning use case is as follows:
> We have a two-input operator A. The first input of A is the model parameters,
> the second input of A is the training data.
> In this case, we have to read the model parameters first, then we read the
> training data. That is we need to finish consuming the first input before we
> consume the second.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
