Scheduled data collections from Performance Analytics can cause nodes to run out of memory when there are a large number of records in the indicator source. When a node runs out of memory (or is low on memory) the following may occur:
- slow performance
- logged out
- instance unavailability
Steps to Reproduce
There are several ways to reduce memory usage during data collection:
- Configure your indicator sources so that a smaller number of records are included. For example, add additional filter conditions to the indicator source.
- Create additional indicator sources instead of including more records in a single indicator source.
- Set the property com.snc.pa.dc.max_row_count_indicator_source to a lower value. This limits the maximum number of records included in any indicator source. By default, this property allows 50,000 records per indicator source.
- Set the property com.snc.pa.dc.max_records to a lower value to limit the maximum number of records collected for snapshots. By default, this property allows 5000 records to be collected. You can also clear the Collect records field for indicators to disable snapshots for those indicators.
- Reduce the number of breakdowns for collected indicators. You can exclude breakdowns in the data collection job or clear the Collect breakdown matrix field for indicators to disable 2nd-level breakdowns for those indicators.
Some changes were made in Jakarta (not direct fixes for this problem) that have improved the memory footprint of the Performance Analytics data collection jobs by 20%. The development team is working on some architectural changes that are aimed at resolving this issue, so for now this problem is closed. Please contact Customer Support if you need help with the workaround.
Related Problem: PRB964769