Open deleonenriqueta opened 3 months ago
Another customer noticed some increase in overhead. Given the versions the customer was using, we advised them to disable some classes from being weaved and it decreased the overhead.
The configuration for the exclusions is:
class_transformer:
excludes: ^reactor/core/publisher/MonoSubscribeOn, ^reactor/core/publisher/MonoSubscribeOn$SubscribeOnSubscriber, ^reactor/core/publisher/FluxMapFuseable, ^reactor/core/publisher/FluxMapFuseable$MapFuseableSubscriber
The $ character in the "excludes" stanza needs to be escaped, so use the following to exclude those classes:
class_transformer:
excludes: ^reactor/core/publisher/MonoSubscribeOn, ^reactor/core/publisher/MonoSubscribeOn\$SubscribeOnSubscriber, ^reactor/core/publisher/FluxMapFuseable, ^reactor/core/publisher/FluxMapFuseable\$MapFuseableSubscriber
Description
When using a Reactor Stack and processing a high volume of requests, we've observed an increase in application overhead (CPU usage and latency).
Acceptance Criteria
Research spike. Investigate whether anything can feasibly be done to optimize performance and reduce this overhead, specifically focusing on token linking.
Additional context
Review the GTSE that references this behavior to gather more information.