Skip to main content
Paulo
Paulo
Silver ⭐️⭐️⭐️
March 22, 2019
Solved

Information about query runtime metrics in Impala UI

Hi,

 

I use Impala UI, under /queries tab, in order to get runtime metrics for queries. I'm a little bit confused about the 'duration' and 'waiting time' metrics. What is being computed and tracked in each of them?

 

I have also timed how long it takes to run the query (since statement is submitted until all rows are fetched) from the client side (I use Ibis framework to submit queries to Impala) and I can see a much smaller runtime than the one reported in the UI. Why is that? Is perhaps the metrics reported an aggregated computing time across all impalad nodes?

 

Thanks,

Paulo.

SahilTakiar
SahilTakiarAnswer
Gold ⭐️
March 25, 2019

What version of Impala are you using? I suspect the meaning of "duration" might have changed in IMPALA-1575 / IMPALA-5397.

 

In general, its possible that the your definition of duration is different from Impala's. Depending on the version, Impala might include the time taken until the query has been actually closed (which would include fetching rows and releasing all resources).

 

I *think* the waiting time is the difference between the current time and the time the query was last actively being processed. So this value could be high if the query has been completed, but the client has not closed the query (which is why "waiting time" shows up the section "waiting to be closed").

View in thread

2 replies

SahilTakiar
Gold ⭐️
SahilTakiarAnswer
Gold ⭐️
March 25, 2019

What version of Impala are you using? I suspect the meaning of "duration" might have changed in IMPALA-1575 / IMPALA-5397.

 

In general, its possible that the your definition of duration is different from Impala's. Depending on the version, Impala might include the time taken until the query has been actually closed (which would include fetching rows and releasing all resources).

 

I *think* the waiting time is the difference between the current time and the time the query was last actively being processed. So this value could be high if the query has been completed, but the client has not closed the query (which is why "waiting time" shows up the section "waiting to be closed").

Translate
    Paulo
    PauloAuthor
    Silver ⭐️⭐️⭐️
    March 28, 2019

    Hi,

     

    The Impala version I'm using is 2.11, so I have those changes there.

     

    One thing I noticed is the duration, when query is submitted from impala-shell, seems to match the duration of the query reported after all rows have been fetched, but that does not seem to be the case when we actually time the duration in the client that submitted the query.

     

    As you said, probably Impala is still dealing with further processing/closing/cleanup of the query at the time the client was already able to fetch all the results of the query and print out the elapsed time.

     

    Thanks for your answer,

    Paulo.

    Translate
    Terms & ConditionsAccessibility statement