[ 
https://issues.apache.org/jira/browse/FLINK-4840?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15584570#comment-15584570
 ] 

zhuhaifeng commented on FLINK-4840:
-----------------------------------

The TPS is numRecordsInPerSecond, is it will be add into with MetricView once 
FLINK-3950 are merged ? 
I think we can infer the capacity of records processed per second from the 
processing time metric for an operator. But, we can't infer the processing time 
from recordsPerSecond when operator process is not reach to capacity. so I 
think the lantency metrics is necessary and can be add into 
OperatorIOMetricGroup.

> Introduce an OperatorSystemMetricGroup
> --------------------------------------
>
>                 Key: FLINK-4840
>                 URL: https://issues.apache.org/jira/browse/FLINK-4840
>             Project: Flink
>          Issue Type: Improvement
>          Components: Metrics
>            Reporter: zhuhaifeng
>            Assignee: zhuhaifeng
>            Priority: Minor
>             Fix For: 1.2.0
>
>
> There will introduces the OperatorSystemMetricGroup, which encapsulates the 
> insantiation of TPS meter, lantency/proc_time_cost histograms. 
> Operator related System metrics are not instantiated directly by the specific 
> Operator, but instead within the OperatorSystemMetricGroup contained in the 
> respective OperatorMetricGroup. They are then later accessed by relevant 
> components(maybe different places), instead of instantiated them identically 
> with static name constants. Other system scope metrics (maybe 
> delay/queue_in/queue_out) can add into the OperatorSystemMetricGroup some 
> later.
> TPS: collect records per second(StreamSource), processed elements per 
> second(other operator)
> lantency/proc_time_cost : collect an record time cost(StreamSource), 
> processed an element time cost (other operator)



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to