Hi, Le Xu, If the job is a streaming job, all tasks should be scheduled before any data can flow through the pipeline, and tasks will run in parallel. I think the Execution Mode[1] and FLIP-134[2] will help you to understand more details.
Best, Hang [1] https://nightlies.apache.org/flink/flink-docs-release-1.16/docs/dev/datastream/execution_mode/#execution-behavior [2] https://cwiki.apache.org/confluence/display/FLINK/FLIP-134%3A+Batch+execution+for+the+DataStream+API Le Xu <sharonx...@gmail.com> 于2022年12月14日周三 02:56写道： > Hello! > > I have a quick question about slot-sharing behavior specified at this link > ( > https://nightlies.apache.org/flink/flink-docs-master/docs/concepts/flink-architecture/). > My understanding would be each task slot in Flink cluster represents a > sequentially running operator (and therefore can be seen as a thread). But > it seems like each operator in Flink is modeled as a thread and many > threads can be pinned to the same task slot. For a task slot that contains > both task A and task B (thread A and thread B), could these two tasks run > in parallel and effectively parallelize tasks assigned to a task slot? > > Thanks! > > Le >
