Watching Aaron Jacobs break down mongolite performance issues... #rstudioglobal
2
15
GIF
@unconj1 does what I proposed at github.com/jeroen/mongolite/… seem like a good solution? You and @opencpu are much more familiar with how data is stored and retrieved in mongoDB than I am
1
From what I figured, the overhead is mostly that all data goes through jsonlite to reshape collections into data frames.
1
Replying to @opencpu @jimhester_ @unconj1
And @knapply_ and I have shown that `RcppSimdJson` can run quite a few performance circles around `jsonlite` (which remains wonderful and tested and complete but also quite a bit slower). cran.r-project.org/package=R…

Jan 22, 2021 · 2:45 PM UTC

1
1
Replying to @eddelbuettel @opencpu @jimhester_
The approach I proposed basically eliminates the need to do expensive JSON operations, but it makes the parser more complex. So maybe RcppSimdJson would indeed be simpler.