Jmeter压测Milvus单条数据插入性能时,发现QPS稳定在162左右 #23296
Replies: 6 comments 27 replies
-
|
Milvus的insert是按如下流程执行:
由于在第2步,insert请求返回结果给客户端,所以客户端的插入耗时只包含从1到2的耗时。 |
Beta Was this translation helpful? Give feedback.
-
|
有尝试过增加proxy的数目或者提高客户端的并发么? |
Beta Was this translation helpful? Give feedback.
-
|
另外,你不应该每次写入的时候都调用hasColleciton,这个会造成额外的RPC开销,影响性能 |
Beta Was this translation helpful? Give feedback.
-
|
我增加了线程数,看到压测的时候写入耗时800ms~1s左右了,想问下写入pulsar是同步写入还是异步写入。增加pulsar配置或者扩容pod数目会有提升吗?如果切换到kafka会不会好一点? @xiaofan-luan |
Beta Was this translation helpful? Give feedback.
-
|
我用python写一个简易的压测,代码如下。
我在本地用docker-compose起了一个milvus cluster,直接用官网上的docker-compose.yml |
Beta Was this translation helpful? Give feedback.
-
|
看起来,写入是有一些瓶颈,接下来我们会花时间优化写入的瓶颈 |
Beta Was this translation helpful? Give feedback.



Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Beta Was this translation helpful? Give feedback.
All reactions