This reuses the OpenTelemetry instance created by the Java agent in code. This also switches to the googlecloud OpenTelemetry exporter on GKE, which avoids conversion loss from OpenTelemetry->Managed Prometheus->Cloud Monitoring.
Superfluous instruments were also dropped:
active_non_daemon_thread_count - can be obtained from an existing default JVM instrument
stage_cpu_time_duration_millis - this did not work correctly and is not really useful (it was measuring CPU time across all JVM threads)
Instruments specific to a computation stage - dropped in favor of adding computation type and stage attributes
This reuses the OpenTelemetry instance created by the Java agent in code. This also switches to the
googlecloud
OpenTelemetry exporter on GKE, which avoids conversion loss from OpenTelemetry->Managed Prometheus->Cloud Monitoring.Superfluous instruments were also dropped:
active_non_daemon_thread_count
- can be obtained from an existing default JVM instrumentstage_cpu_time_duration_millis
- this did not work correctly and is not really useful (it was measuring CPU time across all JVM threads)