ESP Best Practices: Shutting down ESP - Sequential Shutdown process

Symptoms

This article will compare and contrast performing a simple shutdown vs. a sequential shutdown in Fast ESP.

Resolution

Using "nctrl stop" to stop an ESP installation is known as a simple shutdown.  That can often work well to shut down single node environments or even multinode environments that are small in number.  However, in the case of larger multinode environments, Microsoft Technical Support recommends using a sequential shutdown process as a best practice.

The act of performing a sequential shutdown suspends or stops components individually and in a specific sequence.  It also allows all components to keep the same view of the content after a restart. This ensures consistent behavior and data before and after the shutdown. Conversely, a simple shutdown with "nctrl stop" will stop a node completely without first suspending or stopping individual components, and does not ensure consistent behavior or data after a restart of the system.

The steps for a sequential shutdown are outlined below:

1. Stop all feeding via connectors or other data sources
2. Run the command:  nctrl stop contentdistributor
3. Run the command:  nctrl stop procserver_1
4. Run the command:  nctrl stop qrserver
5. Run the command:  indexeradmin -a suspendindexing
6. Verify that the fixmlindex process for each partition is aborted, waiting up to 5 minutes using ps or rc

Additional details:
After suspendindexing, fixmlindex processes should stop what they are doing and become idle, but they will still be running. This can be verified by:
Running the following on a Unix platform: rc -r indexing | grep "state of partition”
Running the following on a Windows platform: rc -r indexing | findstr /C:"state of partition"

The above commands will provide an output of all idle partitions, which is the intent of the suspendindexing step prior to running "nctrl stop indexer". If, and only if, they are not idle after 5 minutes, then it may be necessary to kill the one that is not idle (that is, the one that had some problem stopping its work, and whose ID is already known). Since the rc command returns the partition number, this can be used to identify the process number to kill.  %FASTSEARCH%\var\run\producer_X.pid (Linux:$FASTSEARCH/var/run/producer_X ) will also contain that process number.

7. Run the command:  nctrl stop indexer
8. Run the command:  nctrl stop search-1
9. Run the command:  nctrl stop

To start ESP, always start the admin node first and then the additional node or nodes.

When ESP is stopped on each node, it is recommended that the running processes are checked to make sure there are no hung or lingering processes from ESP that may still be running. This can be accomplished by executing the ps command in Linux, or Task Manager/Process Explorer/Process Monitor in Windows, once ESP has stopped. With those tools, one can verify that no ESP processes are still running. It is recommended to always check for processes that may not have terminated gracefully after shutting down ESP. If there are any ESP related processes still present after ESP is stopped, they should be killed before starting ESP.
Rekvizīti

Raksta ID: 2529030. Pēdējo reizi pārskatīts: 2011. gada 31. marts. Pārskatījums: 1

Atsauksmes