HDFS Collection Agent Transaction Behavior
The transaction behavior for the HDFS collection agent is presented here. For more information about general transaction behavior please refer to the section Transactions Workflow Monitor.
Input/Output Data
The agent emits commands that change the state of the file currently processed.
Command | Description |
---|---|
Begin Batch | Emitted before the first part of each collected file is fed into a workflow. |
End Batch | Emitted after the last part of each collected file has been fed into the system. |
The agent acquires commands from other agents and based on them generates a state change of the file currently processed.
Command | Description |
---|---|
Cancel Batch | If a Cancel Batch message is received, the agent sends the batch to ECS. Note If the Cancel Batch behavior defined on workflow level is configured to abort the workflow, the agent will never receive the last Cancel Batch message. In this situation ECS will not be involved, and the file will not be moved, but left at its current place. |
Hint End Batch | If a Hint End Batch message is received, the collector splits the batch at the end of the current block processed (32 kB), If the block end occurs within a UDR, the batch will be split at the end of the preceding UDR. After a batch split, the collector emits an End Batch message, followed by a Begin Batch message (provided that there is data in the subsequent block). |