EPL, like languages such as Java or C#, relies on garbage collection. Intermittently, the correlator analyses the events that have been allocated, including dictionaries, sequences, closures and streaming networks, and allows memory used by events that can no longer be referenced to be re-used. Thus, the actual memory usage of the correlator might be temporarily above the size of all live objects. While running EPL, the correlator might wait until a listener or onload() action completes before performing garbage collection. Therefore, any garbage generated within a single action or listener invocation might not be disposed of before the action/ listener has completed. It is thus advisable to limit individual actions/listeners to performing small pieces of work. This also aids in reducing system latency.

The cost of garbage collection increases as the number of events a monitor instance creates and references increases. If latency is a concern, it is recommended to keep this number low, dividing the working set by spawning new monitor instances if possible and appropriate. Reducing the number of event creations, including string operations that result in a new string being created, also helps to reduce the cost of garbage collection. The exact cost of garbage collection could change in future releases as product improvements are made.