How to free memory from overwritten .RDA variable?

General
1

Hey guys,

I have to load some really big dataframes stored as .rda files. In doing so, I did notice the following behavior of my Rstudio Server (same behavior on local Rstudio installation).

Whenever I load multiple variables, the memory seems not to be freed after deleting a variable with rm and calling gc. Example

load("path_to_2gb.rda")
rm(data)
gc()
load("path_to_4gb.rda")

At this point, the session still allocates over 6gb of RAM. My expectation would be that the memory is 4gb, as the first loaded variable is no longer required.

Is this a known behavior or is there another way to free memory? I really need to load all my data sequentially however I have not enough space to have them all stored in RAM.

2

can you post your output if you do the following ?

gc(full = TRUE,verbose = TRUE)
load("path_to_2gb.rda")
gc(full = TRUE,verbose = TRUE)
rm(data)
gc(full = TRUE,verbose = TRUE)
3 
> gc(full = TRUE,verbose = TRUE)
Garbage collection 16 = 12+1+3 (level 2) ... 
43.7 Mbytes of cons cells used (60%)
14.0 Mbytes of vectors used (22%)
          used (Mb) gc trigger (Mb) max used (Mb)
Ncells  817740 43.7    1368111 73.1  1368111 73.1
Vcells 1833753 14.0    8388608 64.0  2206687 16.9
> load("path_to_4gb_file.rda")
> gc(full = TRUE,verbose = TRUE)
Garbage collection 36 = 12+1+23 (level 2) ... 
62.0 Mbytes of cons cells used (56%)
4099.9 Mbytes of vectors used (75%)
            used   (Mb) gc trigger   (Mb)  max used   (Mb)
Ncells   1160025   62.0    2080672  111.2   1368111   73.1
Vcells 537370475 4099.9  712269123 5434.2 537370640 4099.9
> rm(data)
> gc(full = TRUE,verbose = TRUE)
Garbage collection 37 = 12+1+24 (level 2) ... 
43.7 Mbytes of cons cells used (39%)
14.0 Mbytes of vectors used (0%)
          used (Mb) gc trigger   (Mb)  max used   (Mb)
Ncells  817734 43.7    2080672  111.2   1368111   73.1
Vcells 1833743 14.0  569815299 4347.4 537370640 4099.9

After end of code snippet

image
image336×116 4.81 KB

4

Ok, its what I expected. the current used is just 14Mb, but the max used recorded was 4099.9.
I think the situation is something like. R gets memory from the OS for it to allocate to your objects.
R tracks your object and will garbage collect them when you drop them; but it wont release this to your OS; it will retain that memory for other objects (without it needing to go beg your OS for more memory resource).
So this is a double edged sword.
the good ) you can expect to load some other equivalent sized file with R successfully, you really did release that as far as R is concerned ; the used (Mb ) went back down to where it was at the starts 14 to 4099.9 to 14Mb. going back up to 4099.9 should be no issue.
the bad ) its not released as far as the OS is concerned so if your are using other applications outside of R; they won't get that memory that R is hanging onto for you...

I'm not an expert here and would be happy for someone who is to clarify; and also speak up if there are methods to force R to release memory back to the OS without closing or restarting...

5

That would mean that as soon as the R-Session reaches the RAM Limit of the system, all available RAM is reserved for that R session and cannot not be used from other projects until the session is
terminated?

6

This forum has a minimum post length of 20 characters;
to answer your question: yes