Copyright © TIBCO Software Inc. All Rights Reserved
Copyright © TIBCO Software Inc. All Rights Reserved


Chapter 9 Process Management : Restarting Failed Processes

Restarting Failed Processes
To make sure that the TIBCO iProcess Engine is always running, the Process Sentinels are constantly monitoring for any processes that fail. Each of the Process Sentinels operating in the node will monitor all processes running on the same server. If a process fails, the Process Sentinels will start a new copy of the process. In many cases, there can be many instances of a process running such as 5 Background processes and the Process Sentinels will make sure that the correct instance of the process is restarted.
The process_attribute table is responsible for the configuration of how processes are restarted. You can specify whether you want the process to be manually restarted, restarted a set number of times, and the minimum time a process must run before it can be restarted. Settings can be applied to the node, a single server in the node, a type of process, or an individual instance of a process.
Restarting Failed Process Sentinels
Process Sentinels consist of two processes: a worker process and a watcher process. On each server in an TIBCO iProcess Engine cluster, these two processes are started. This architecture is designed so that each process can monitor the other one and restart it if it fails. This makes sure that the Process Sentinels are always running. The watcher and worker processes perform different tasks:
Watcher Process
After starting the worker process and establishing a two way communication channel with the worker process, the watcher process monitors the worker process and can restart it upon a failure. If any errors occur, an error message is logged in the SWDIR\logs\sw_error file.
On the DB2 version, the watcher process also runs the iProcess event daemon (see Event Handling).
Worker Process
The worker process performs the process monitoring of the TIBCO iProcess Engine processes and monitors and restarts them if they fail. The worker process also monitors the watcher process to make sure that it is always running and will restart it if it fails. The following is the start-up routine for the worker process:
1.
2.
3.
Connect to the database and read the node_cluster table to determine if the server is configured to be part of the TIBCO iProcess Engine. It can also determine if it needs to operate as a master or slave.
If there is no entry for this server in the node_cluster table, an error message is logged to the operating system event log (syslogd). The worker process will shut down and also shutdown the watcher process.
4.
Read the process_config table to build an in memory process hierarchy model.
5.

Copyright © TIBCO Software Inc. All Rights Reserved
Copyright © TIBCO Software Inc. All Rights Reserved