@RunLLM namespace unload or rebalance is not balancing brokers .

If namespace unload or rebalance is not balancing brokers, it is often due to load balancer misconfiguration, insufficient bundle splitting, or the chosen load shedding strategy. Ensure that the load balancer is enabled, the shedding thresholds are set appropriately, and that bundles are split enough to allow redistribution. Also, check that the load shedding strategy (e.g., ThresholdShedder or UniformLoadShedder) is suitable for your workload and that new brokers are eligible to receive bundles. Manual unloads may not be effective if the system is not configured to redistribute load properly or if bundle distribution is uneven at the start. For more, see Pulsar load balancing concepts, administration guide, and troubleshooting UniformLoadShedder. Would you like more detail on diagnosing or tuning your configuration? Sources: - Pulsar load balancing concepts - Pulsar administration load balance - UniformLoadShedder troubleshooting 👍 0 button 👎 0 button 💡 Hint: Mention @runllm in the thread for followups.

@RunLLM Threshold load shedder is not distributing load among 4 broker instead only 2 are loaded .

If ThresholdShedder is only loading 2 out of 4 brokers, it's likely due to the average-based threshold logic: bundles are only unloaded if a broker's usage exceeds the cluster average plus the configured threshold. If two brokers are heavily loaded and two are idle, the average may not be high enough to trigger unloading, so idle brokers remain unused. To address this, enable the lower boundary shedding feature (

lowerBoundarySheddingEnabled=true

), which allows the system to offload from the most loaded broker when another broker is far below the average, thus utilizing idle brokers. See ThresholdShedder documentation and related improvement. Would you like step-by-step guidance on enabling this feature or tuning your configuration? Sources: - ThresholdShedder documentation - ThresholdShedder lower boundary improvement 👍 0 button 👎 0 button 💡 Hint: Mention @runllm in the thread for followups.

@RunLLM

org.apache.pulsar.broker.loadbalance.impl.ThresholdShedder is correct shedder?

Yes,

org.apache.pulsar.broker.loadbalance.impl.ThresholdShedder

is the default and recommended load shedding strategy in Pulsar since version 2.10.0. It unloads bundles from brokers whose usage exceeds the cluster average plus a configurable threshold, helping to evenly distribute load across brokers. See official docs. Would you like more detail on when to use ThresholdShedder or alternatives? Sources: - ThresholdShedder documentation 👍 0 button 👎 0 button 💡 Hint: Mention @runllm in the thread for followups.

@RunLLM

org.apache.pulsar.broker.loadbalance.impl.ThresholdShedder keeping 2 broker idle out of 4 where bundle is 16

This happens because ThresholdShedder only unloads bundles if a broker's usage is above the cluster average plus the threshold. If two brokers are busy and two are idle, the average may not be high enough to trigger unloading, so the idle brokers remain unused. To fix this, enable

lowerBoundarySheddingEnabled=true

in your configuration, which allows the system to utilize idle brokers by offloading from the most loaded broker when another is far below average. See ThresholdShedder docs and lower boundary improvement. Would you like step-by-step instructions to enable this feature? Sources: - ThresholdShedder documentation - ThresholdShedder lower boundary improvement 👍 0 button 👎 0 button 💡 Hint: Mention @runllm in the thread for followups.