SQL Server 2019, Machine Learning Services, rxExecBy RRS feed

  • Question

  • Hi there,

    I migrated from SQL2017 to 2019, and I am struggling to get R code to run.

    I'm running a function that uses the RevoScaleR library, specifically, rxExecBy that drives parallelism. If the table is just too big, R just hangs. I can run samples (sub 400k rows) and all bits of code if I don't use rxExecBy. I believe it has to do with adding workers - it tends to work when only 1 work is called.

    I suspected the new AppContainer framework, but I can't see anything I missed: checked permissions, worked all the instructions on SQLRUserGroup... I'm stuck. When I cancel the query, i think get "Aborting: mpiexec on Server failed to communicate with smpd manager on Server". ??

    Is there any particular guidance on how to configure rxExecBy for SQL2019 / AppContainers?


    Friday, February 21, 2020 1:09 AM

All replies

  • In your post you say you can run smaller samples if you do not use rxExecBy. Does that mean that you canot use rxExecBy at all, i.e. that you fail as soon as you run rxExecBy?

    Your SQL 2019 instance is that a new SQL 2019 install, or is it an upgrade from a 2017 instance? | @nielsberglund

    Friday, February 21, 2020 2:54 AM
  • I can use rxExecBy up to about 400k rows, at which point it hangs. I checked the Task Manager, and the box is totally idle. When it's sub 400k rows, it looks like only 1 process was used, i.e. it didn't attempt to parallelize, but the query completes successfully.

    Initially I did an upgrade, when it failed, I wiped the machine and did a clean install of 2019. 



    Friday, February 21, 2020 2:59 PM