2. Monitoring

2. Monitoring

Monitoring is essential for gaining insights into the performance of the load balancing algorithm. In this section, we will discuss the key metrics to monitor and how to interpret them.

2.1 unload bundle count

This metric indicates the total number of bundles that have been unloaded. When the value of this metric increases, it signifies that the load balancing algorithm is actively unloading bundles.

pulsar_lb_unload_bundle_count # before 2.11.0
pulsar_lb_unload_bundles_total # after 2.11.0

The metric name has changed from pulsar_lb_unload_bundle_count to pulsar_lb_unload_bundles_total since version 2.11.0 by PR: Rename Pulsar lb metrics to specify OpenMetrics.

img.png

note: this metric will only be exported by the leader broker, so you can identify the leader broker by checking the pulsar_lb_leader_broker metric.

2.2 bandwidth usage

pulsar_lb_bandwidth_in_usage
pulsar_lb_bandwidth_out_usage

img_1.png

img_2.png

2.3 CPU usage

pulsar_lb_cpu_usage

img_3.png

CPU usage is a critical metric to monitor.

2.4 memory and direct memory usage

pulsar_lb_memory_usage
pulsar_lb_direct_memory_usage

img_4.png

img_5.png

Though the direct memory usage metric and the memory usage metric are not closely related to the actual load, we still need to monitor them to avoid OOM errors.