Attempting to use multiple ODBC connections across parallelized worker threads may fail as in the following example:The problem is the worker processes receive the ODBC connections as closed.
The issue here is that connections are process-specific, so unless the workers are sharing the parent process (as in multicore workers created via forking), the parent's connections can't be shared by the workers. To distribute ODBC computations on non-forked workers, establish the connections on each worker as part of the distributed task.
Example:
loaddata <- function(cn){
result <- sqlQuery(cn,'select * from boston')
return(head(result))
}
library(RODBC)
cn1 <- odbcConnect("RevoTestDB", uid='RevoTester', pwd='RevoTester')
cn2 <- odbcConnect("RevoTestDB", uid='RevoTester', pwd='RevoTester')
cn3 <- odbcConnect("RevoTestDB", uid='RevoTester', pwd='RevoTester')
cn4 <- odbcConnect("RevoTestDB", uid='RevoTester', pwd='RevoTester')
rxSetComputeContext('localpar')
system.time ({
z <- rxExec(loaddata, rxElemArg(list(cn1,cn2,cn3,cn4)),
packagesToLoad='RODBC')
})
Error in do.call(.rxDoParFUN, as.list(args)) :
task 1 failed - "first argument is not an open RODBC channel"The issue here is that connections are process-specific, so unless the workers are sharing the parent process (as in multicore workers created via forking), the parent's connections can't be shared by the workers. To distribute ODBC computations on non-forked workers, establish the connections on each worker as part of the distributed task.
Example:
loaddata <- function(){
library(RODBC)
cn <- odbcConnect("RevoTestDB", uid='RevoTester', pwd='RevoTester')
result <- sqlQuery(cn,'select * from boston')
return(head(result))
}
z <- system.time({z <- rxExec(loaddata,
packagesToLoad='RODBC')})