Skip to:

  1. Skip to Jira Navigation
  2. Skip to Side Navigation
  3. Skip to Main Content

Pollerd take a long time to start on systems with large inventories

Description

On a system with close to 100k nodes, pollerd was found to take over an hour to start:

We should be able to optimize the startup routines so that poller services takes a reasonable amount of time to start, ideally < 10 minutes in this particular case.
We should optimize the code to

Acceptance / Success Criteria

None

Lucidchart Diagrams

Activity

Show:

Benjamin Reed July 12, 2022 at 2:06 PM

Merged to release-30.x

Jesse White July 9, 2022 at 10:44 PM

Jesse White July 8, 2022 at 8:02 PM

Doing some initial analysis, the following inefficiencies were found:
1) Package IP list is always rebuilt on a miss
https://github.com/OpenNMS/opennms/blob/opennms-30.0.0-1/opennms-config/src/main/java/org/opennms/netmgt/config/PollerConfig.java#L318
2) Doing .contains() on a large list to check for presence of an item
https://github.com/OpenNMS/opennms/blob/opennms-30.0.0-1/opennms-config/src/main/java/org/opennms/netmgt/config/PollerConfigManager.java#L527
Data structure could be changed to a Set instead
3) Cache is always flushed before lookups
https://github.com/OpenNMS/opennms/blob/opennms-30.0.0-1/opennms-config/src/main/java/org/opennms/netmgt/config/PollerConfigManager.java#L489
4) Cache is always rebuilt for each service schedule
https://github.com/OpenNMS/opennms/blob/opennms-30.0.0-1/opennms-services/src/main/java/org/opennms/netmgt/poller/Poller.java#L559

Fixed

Details

Assignee

Reporter

Components

Fix versions

Affects versions

Priority

PagerDuty

Created July 8, 2022 at 7:56 PM
Updated July 12, 2022 at 2:06 PM
Resolved July 12, 2022 at 2:06 PM