Ah, I think I'd missed this message.  Thank you.  In case you didn't see, we've logged

DRILL-8119 - MongoDB query regression: mem leak and data type error <https://issues.apache.org/jira/browse/DRILL-8119>

On 2022/01/28 19:28, Daniel Clark wrote:
I've attached the json version of the profile. Is that what you are requesting? The latest run ended in another error.

On Fri, Jan 28, 2022 at 9:47 AM luoc <[email protected]> wrote:


    Hi Daniel,
      The 1.20 is actually included a feature to improve performance
    using the new push down framework.
      For describe the specific things, could you please post the
    profile file of query?

    > On Jan 28, 2022, at 22:32, Daniel Clark <[email protected]> wrote:
    > 
    > Hello,
    >
    > While evaluating 1.20.0-SNAPSHOT release performance, I ran a
    mongo query that runs in 15 minutes in the 1.19 release (below).
    >
    > SELECT `Elements_Efforts`.`EffortTypeName` AS `EffortTypeName`,
    >   `Elements`.`ElementSubTypeName` AS `ElementSubTypeName`,
    >   `Elements`.`ElementTypeName` AS `ElementTypeName`,
    >   `Elements`.`PlanID` AS `PlanID`
    > FROM `mongo.grounds`.`Elements` `Elements`
    >   INNER JOIN `mongo.grounds`.`Elements_Efforts`
    `Elements_Efforts` ON (`Elements`.`_id` = `Elements_Efforts`.`_id`)
    > WHERE (`Elements`.`PlanID` = '1623263140')
    > GROUP BY `Elements_Efforts`.`EffortTypeName`,
    >   `Elements`.`ElementSubTypeName`,
    >   `Elements`.`ElementTypeName`,
    >   `Elements`.`PlanID`
    >
    > The query runs for 34 minutes before returning this error; "Sort
    exceeded memory limit of 104857600 bytes, but did not opt in to
    external sorting. Aborting operation. Pass allowDiskUse:true to
    opt in.' on server localhost:27017." Any ideas? I realize that
    it's a mongodb error, but the mongo database doesn't raise this
    error with the 1.19 release. I was expecting improved performance
    with the mongo storage plugin in the upcoming 1.20 release.
    Nothing in my environment has changed. I've attached the full
    stacktrace.
    >
    > <stacktrace.txt>

Reply via email to