Skip to content

Files

Latest commit

author
honma
Mar 20, 2015
af42057 · Mar 20, 2015

History

History
37 lines (21 loc) · 1.79 KB

Design.md

File metadata and controls

37 lines (21 loc) · 1.79 KB

Job Engine Design

Purpose: The Job Engine is a coordinator which manage the tasks' life cycle and CRUD of the tasks. To be clearified, the Job Engine does not run the task itself, instead the task runs on hadoop(or whatever it depends).

Executable

Class Diagram

Executable is a top-level interface for all kinds of jobs or tasks.

AbstractExecutable is a abstract implementation of Executable, it provides:

  1. some getter and setter method
  2. default implementation of Executable.execute()
  3. life cycle method of an Executable and their default implementation

DefaultChainedExecutable is an implementation of AbstractExecutable which contains a group of Executable

ExecutableManager

ExecutableManager provide the CRUD function for an Executable

ExecutableDao

ExecutableDao provide the access of the persistent object for Executable

There are two persistent object for one Executable

  1. ExecutablePO is to store the runnning parameters for the Executable, and once the Executable is submitted, ExecutablePO is unmodifiable.
  2. ExecutableOutputPO is to store the running result for the Executable, for instance the current state, error log.

DefaultScheduler

DefaultScheduler is a coordinator for Executables.

There is a daemon thread call JobFetcher running periodically. It is responsible for scheduling the Executables

Note: there should always be only one instance running in the cluster. And it is configured using "kylin.server.mode" in the "kylin.properties", there are two modes "all" & "query", "all" means it will defaultly start the scheduler. So if there are multiple kylin instances, make sure there is only one instance whose "kylin.server.mode" is set to "all".