`obj_size` returns different values before and after accessing an element about lobstr

[hadley]

Hmmm, interesting. I see the opposite:

lobstr::obj_size(baseenv()$.__S3MethodsTable__.)
#> 6,307,928 B
x <- sort(lobstr::obj_sizes(!!!as.list(baseenv()$.__S3MethodsTable__.)))
lobstr::obj_size(baseenv()$.__S3MethodsTable__.)
#> 12,894,608 B

^{Created on 2022-06-17 by the reprex package (v2.0.1)}

I think this environment is the registry of S3 methods for all base functions, but it is surprising that it is so large for you.
I think it's only affected by the packages you load, so you might try creating a reprex by extracting just the library() statements from your shiny app.

The size of the individual methods is also surprising, but I can't reproduce it:

library(DBI)
library(tidyverse)

lobstr::obj_size(baseenv()$.__S3MethodsTable__.$toString.Id)
#> 4,320 B
baseenv()$.__S3MethodsTable__.$toString.Id
#> function (x, ...) 
#> {
#>     paste0("<Id> ", paste0(names(x@name), " = ", x@name, collapse = ", "))
#> }
#> <bytecode: 0x7fbcb887ca90>
#> <environment: namespace:DBI>
lobstr::obj_size(baseenv()$.__S3MethodsTable__.$toString.Id)
#> 4,320 B

^{Created on 2022-06-17 by the reprex package (v2.0.1)}

If possible, I'd also recommend you trying this code in the latest version of R, in case this is a some buglet in base R.

[bersbersbers]

This issue is a few months old, so I don't have anything to reproduce my original results, unfortunately.

What I have found out in the meantime, however, is that the "memory leak" I was after turned out to be a couple of paired saveRDS and readRDS (of the same file, respectively), and these saveRDS operations involved ggplot2 objects, and hence the complete environments. I cannot say for sure whether that actually consumed a lot of memory or just made it appear so, but performance was certainly not good. When I replaced readRDS operations by reading the plots from an environment that I pass around, most memory and runtime issues disappeared. (Any they completely disappeared when I also removed the saveRDS calls, after reading your own tidyverse/ggplot2#3994 (comment)).

One (very much uninformed) hypothesis that I have is that apparent memory consumption goes down when inspection of the members of some __S3MethodsTable__ reveals that many duplicates of each other that use the same memory, but again, that is a very wild guess.

[bersbersbers]

Bottom line: my original issue is solved and I can hardly contribute to reproducing it. If you are not worried by the behavior that I observed (or the opposite one you are observing now), feel free to close this.

[hadley]

I'll probably continue do a little light exploration for my own edification, but my experience from working on this sort of problem in the past is that fully understanding the phenomenon is likely to be a lot of work for little gain.