Follow

Reducing 5.x to 6.x Upgrade Times by Increasing Workers for Indexing

Applies To

  • All upgrades from 5.x to 6.x
  • Control Center master hosts with configurable core counts (AWS, VMWare, etc.)

Summary

Resource Manager 6 replaced certain catalogs (and the zencatalogservice daemon) with Apache Solr.  As such, upgrades from 5.x to 6.x require that all cataloged objects be reindexed into the new Solr index as an upgrade step.  

By default, the reindex step is configured to use 8 workers.  It is possible to specify additional workers, but due to resource contention there are diminishing returns after the worker count exceeds the CPU core count.  If your Control Center master already has more than 8 cores, you can use the number of cores for your worker setting.  If not, you may wish to increase your CPU core count prior to upgrading to take advantage of more workers.

Examples

1. If your Control Center master has more than 8 cores, you can:

  a. Perform the upgrade procedure as documented, without altering the reindex worker count.  The reindexing process will run normally.
  b. Increase the reindex worker count as detailed below to reduce the time for the reindex step.  The reindexing process will run in less time due to the increased worker count.

2. If your Control Center master has less than 8 cores, you can:

  a. Perform the upgrade procedure as documented, without altering the reindex worker count.  The reindexing process will run normally, though it will run slower than if you had 8 (or more) cores.
  b. If your Control Center master host is a virtual machine, you may provision additional CPU cores to the VM and then increase the reindex worker count as detailed below to reduce the time for the reindex step.  The reindexing process will run in less time due to the increased worker count and CPU resources.

3. If your Control Center master has 8 cores, you can:

  a. Perform the upgrade procedure as documented, without altering the reindex worker count.  The reindexing process will run normally.
  b. If your Control Center master host is a virtual machine, you may provision additional CPU cores to the VM and then increase the reindex worker count as detailed below to reduce the time for the reindex step.  The reindexing process will run in less time due to the increased worker count and CPU resources.

 

Procedure

  • Increase the core count of your Control Center master, if applicable.
  • On the Control Center master, open the /root/6.1.x/upgrade-resmgr.sh script in your text editor of choice.
  • Find the following line:
UPGRADE_REINDEX='SVC_EXEC NO_COMMIT "Zenoss.resmgr/Zenoss/User Interface/Zope" su - zenoss -c "/opt/zenoss/bin/upgrade_reindex.sh Zenoss.resmgr"'
  • Update the line to read as follows, changing 'cores' to the number of available CPU cores on the host:
UPGRADE_REINDEX='SVC_EXEC NO_COMMIT "Zenoss.resmgr/Zenoss/User Interface/Zope" su - zenoss -c "/opt/zenoss/bin/zencatalog run --createcatalog --forceindex --workers=cores"'
  • Perform the upgrade steps as defined in the Resource Manager Upgrade Guide.
  • On upgrade completion, reduce the core count of your Control Center master, if applicable and desired.

Notes:

For systems with more than 16 cores, set '--workers=16'.  A worker count higher than 16 may lead to resource contention.  Depending on storage throughput, workers may experience read timeouts causing the process to fail and require that the indexing step be restarted.

Was this article helpful?
0 out of 0 found this helpful

Comments

Powered by Zendesk