Information Technology Reference
In-Depth Information
target code or command, results are notified back to the CM, which routes mes-
sages to the corresponding Task Handler instance using the Task Service.
To finally get the results of the parallel (distributed) execution, application
main
waits for all Task instances to finish. Again, Task Handler provides proper
methods for this purpose.
current master
remote nodes
:Application
:Task Handler
:Task Service
:Worker 1
:Worker N
create_task()
Ta s k
execute(Task)
Execute: Task
Instances: N
Run: Task
Instances: 1
Run: Task
Instances: 1
target
target
result
result
wait(Task)
End: Task
Instances: 1
End: Task
Instances: N
Done: Task
Done: Task
results
Fig. 4.
Distributed task sequence diagram
The target code needs to be available in every remote node. This requirement
can be accomplished in two ways:
1. Using external remote node configuration software to maintain desired target
code updated across the network, then the Task just has to execute a shell
command.
2. Using the shared storage system to upload desired code, then the Task is
configured to download and execute it.
A combination of those options can be used to run complex tasks -commonly
found in legacy applications- that may require to run multiple commands, dif-
ficult parameter settings and processing partial results.
