如何排查大查询负载不均衡问题 #26283
Unanswered
ponponon
asked this question in
Q&A and General discussion
如何排查大查询负载不均衡问题
#26283
Replies: 1 comment 1 reply
-
|
1).取决于你有多少数据,如果数据量足够在7台机器上分,我建议你不要用replica功能。每个查询只能在一个replica使用,并发不够高的情况下可能并不能把所有replica使用起来。 |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
你好,我现在的 milvus 集群,每个节点的机器配置是 128 核 512GB 的 RAM
一共有 7 个机器,每个机器上安装了一个 querynode
每个机器各一个querynode,一共 7 个 querynode
副本设置为 7 个副本、3 个分片
然后现在搜索的时候遇到一个问题
就是搜索负载不均衡,有的机器满载,有的机器半载,有的机器零星有几个任务
这类问题如何排查呢?
母本大概是 1200w 的向量
样本的话,每次提交查询的样本中包含的向量数在 100-10000 都有,一般都是一次提交上千个向量
然后提交次数不多,可能几分钟一次
查询的不会插入新的向量
针对查询负载不均衡的问题如何排查呢?
Beta Was this translation helpful? Give feedback.
All reactions