Replies: 2 comments
-
Any reason why these are disabled? I think their defaults are to be enabled... Is it possible those switches being off prevents the system from splitting load across brokers? See here -> https://pulsar.apache.org/docs/3.2.x/administration-load-balance/#split-namespace-bundles |
Beta Was this translation helpful? Give feedback.
-
Because previously considered turning on loadBalancerAutoBundleSplitEnabled would cause client fluctuations,so the configuration items of 6 brokers are all true,If there is a problem with this configuration, why there is no big difference in the number of bundles on each broker? |
Beta Was this translation helpful? Give feedback.
-
cluster version:3.1.1 with ack-aware placement policy
components: 3broker with 2 region , 3 bookie with 2 region,broker and bookie are mixed deploy.
problem:
One of the brokers has a very high load, which causes its jvm heap memory and gc time to be long, seriously affecting the business.
,but its bundle_msg_in/out/producer_count/topic_count is not the largest,
The following is relevant information,can someone help me?
1: 10.50.100.197 broker has high publish latency
2: 10.50.100.197 broker has high jvm heap use and long gc time
3:broker has almost average bundle msg_rate/topic count
4: broker config
Beta Was this translation helpful? Give feedback.
All reactions