Calling an R model via a web service - "Error in raw connection", "all connections are in use"

Question

Hi there,
I have built an R model in Azure ML using the "Create R model" module and have trained it, turned it into a scoring experiment and a web service. I am now sending new data to it to be scored from an Excel workbook, and getting it to return
the prediction to the same workbook. Sometimes (but not always) when I try to run it from the workbook I get the following error:
{"error:{"code":"LibraryExecutionError","message":"Module execution encountered an internal library error.","details":[{"code":"FailedToEvaluateRScript","target":"Score Model (RPackage)","message":"The following error occurred during the evaluation of R script: R_tryEval: return error: Error in rawConnection (raw(0),\"wb\"):all connections are in use \n"}]}}
Usually, if you just press "OK", it works, but then sometimes it comes up again (seems quite random). I have tried a bit of troubleshooting and know that when you open lots of connections (e.g. with connections()) in R this error can occur, but
I'm not sure why in this context the error is coming up. Also, before this error I was getting the error regarding "cannot change value of locked binding for 'dataset'" described here:
https://social.msdn.microsoft.com/Forums/azure/en-US/e20b8cbf-2e89-4898-b602-d2138bf8901c/cannot-change-value-of-locked-binding-for-dataset?forum=MachineLearning
and so I made the suggested fix, adding
unlockBinding("dataset", environment())
to my Create R Model's Score script. I don't know whether this has anything to do with it or not?
Could anyone offer advice of how to get round this?
Many thanks.

Answer 1

Hi,
I've solved the problem. For the record I added:
closeAllConnections()
at the end of the R scoring script (and at the end of another R-script I was using higher up the pipeline) and that seemed to do the trick.
Still, I'm intrigued as to why this was necessary.