Performance drops dramatically in controller mode when transientStorePoolEnable=true

[riki-wang]

1. Problem
   Broker在自动故障转移模式下，设置transientStorePoolEnable=true导致benchmark性能测试急剧下降

   • transientStorePoolEnable=true测试性能
     

   • transientStorePoolEnable=false测试性能
     

2. Environment

   • RocketMQ 5.0 1Master-1Slave Dledger Controller Mode
   • broker配置

     brokerClusterName = DefaultCluster
     brokerName = broker-a
     brokerId = -1
     deleteWhen = 04
     fileReservedTime = 12
     enableControllerMode = true
     namesrvAddr = 10.110.150.106:9876;10.110.150.107:9876;10.110.150.108:9876
     controllerAddr = 10.110.150.106:59878;10.110.150.107:59878;10.110.150.108:59878
     allAckInSyncStateSet = true
   
     flushDiskType = ASYNC_FLUSH
   
     transientStorePoolEnable=true
     transientStorePoolSize=3
   
     listenPort = 59011
   
     storePathRootDir=/home/data/rocketmq/broker-a/store
     storePathIndex=/home/data/rocketmq/broker-a/store/index
     storeCheckpoint=/home/data/rocketmq/broker-a/store/checkpoint
     abortFile=/home/data/rocketmq/broker-a/store/abort
     storePathCommitLog=/home/data/rocketmq/broker-a/store/commitlog
     storePathConsumerQueue=/home/data/rocketmq/broker-a/store/consumequeue
     storePathEpochFile=/home/data/rocketmq/broker-a/store/storePathEpochFile
   

[mxsm]

Hi @riki-wang


Little performance impact

[riki-wang]

Hi @riki-wang Little performance impact
首次启动会有这个问题，而且是要在controller模式下，1master-1slave

[mxsm]

Hi @riki-wang Little performance impact
首次启动会有这个问题，而且是要在controller模式下，1master-1slave

我试下首次启动，我上面的模式就是Controller的A-HA的自主切换模式。

[mxsm]

@riki-wang

When first start on controller model, broker has performance impact.
I will try to find it and fix.
thanks for you catch.

[RongtongJin]

Hi @riki-wang, the issue of enabling transientStorePool in controller mode is still fixed in #5722 . After the repair is completed, we can continue to test.

[riki-wang]

Hi @riki-wang, the issue of enabling transientStorePool in controller mode is still fixed in #5722 . After the repair is completed, we can continue to test.

本来是准备给生产环境引入自动故障转移的，就突然测出这个性能bug了

[RongtongJin]

Hi @riki-wang PR #5722 has fixed it. You can try this PR code.

[Oliverwqcwrw]

close as fixed

[riki-wang]

Rocketmq v5.1.0 still has this issue