Prometheus向量运算报错：求各durable_name序列差值咨询

需求 #

针对每个durable_name，计算指标nss_chan_last_seq的max值与nss_chan_subs_last_sent的min值的差值。

原查询与报错 #

原last_seq查询 #

max(nss_chan_last_seq{channel="some-channel", instance=~"some-cluster-cl.*"}) by (channel)

原last_sent查询 #

min(nss_chan_subs_last_sent{channel="some-channel", durable_name=~"durable-name-s.*", durable_name!~".*test"}) by (channel, durable_name)

原运算语句 #

max(nss_chan_last_seq{channel="some-channel", instance=~"some-cluster-cl.*"}) by (channel) - on (channel) min(nss_chan_subs_last_sent{channel="some-channel", durable_name=~"durable-name-s-.*", durable_name!~".*test"}) by (channel, durable_name) 

报错信息 #

execution: found duplicate series for the match group {channel="some-channel"} on the right hand-side of the operation: [{channel="some-channel", durable_name="durable-name-s16"}, {channel="some-channel", durable_name="durable-name-s14"}];many-to-many matching not allowed: matching labels must be unique on one side

解决方案 #

报错原因是PromQL默认禁止多对多匹配，此处左边单个channel对应一个聚合结果，右边单个channel对应多个durable_name的聚合结果，属于一对多匹配场景，需要用group_left显式声明匹配规则，并保留右侧独有的durable_name标签。

正确查询语句：

max(nss_chan_last_seq{channel="some-channel", instance=~"some-cluster-cl.*"}) by (channel)
- on (channel) group_left(durable_name)
min(nss_chan_subs_last_sent{channel="some-channel", durable_name=~"durable-name-s-.*", durable_name!~".*test"}) by (channel, durable_name)

说明 #

• group_left(durable_name)：指定以左侧的channel聚合结果为基准，匹配右侧所有同channel的durable_name聚合结果，同时保留右侧的durable_name标签。
• 最终结果会包含channel和durable_name两个标签，每个durable_name对应一条差值数据，完全满足需求。

内容的提问来源于stack exchange，提问作者Desa