Cascading

Let’s understand better how to read the timeline view. To do that, let’s go deeper into understanding the different states for the Cascading application.

As the Cascading framework composes the Flows, they enter the "Pending" state. Since Flows within a Casacade can have dependencies, the Flows stay in the "Pending" state until all the dependencies predicating its execution have successfully completed.

Similary, as the query planner maps the assembly of steps into Mappers and Reducers (or other execution primitives associated with the underlying fabric), they now enter the "Submitted" state waiting the work unit to be executed, at which time they enter the "Running" state

Measuring how much time the application, flow, or a Cascading step spends in a particular state can give valuable insights about how to improve the performance of your application.

For example, a large time interval between the submitted and running time can allude to concurrency issues in running MapReduce jobs, and suggesting that increasing the number of mapper or reducer slots can significantly improve the application performance.

Similarly, if the application spends a large proportion of its total run time in pending state, it may be cause to review your application architecture and see how some dependencies between the Flows be eliminated.

You can visualize this information related to latencies for each state from two places. Hover your mouse over the timelime bars, and you can see the state associated with each color band. For a more detailed and an analytical view, select "Add Columns", and select the following counters: "Pending Time", "Start Time", "Submit Time", "Run Time", "Finished Time"

Now with these detailed insights, you are ready to take your application tuning to the next level!