Data Leak in DataLoader ?

[camilodlt]

The MLUtils.DataLoader does not free up memory even if the object is garbage collected ? :

➜  tmp julia --project -t 16
               _
   _       _ _(_)_     |  Documentation: https://docs.julialang.org/
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.11.6 (2025-07-09)
 _/ |\__'_|_|_|\__'_|  |  Official https://julialang.org/ release
|__/                   |

julia> using MLUtils

julia> function random_data(n=1000)
           return [rand(Float32, 1000,1000, 10) for i in 1:n]
       end
random_data (generic function with 2 methods)

julia> function test(i)
           data = random_data(i*10)
           dl = DataLoader(data; batchsize = 100)
           nothing
       end
test (generic function with 1 method)

julia> for i in 1:100
           @info "RAM left : $(Sys.free_memory()/(2^20))"
           test(i*10)
           GC.gc(true)
           GC.gc()
       end
[ Info: RAM left : 108832.73046875
[ Info: RAM left : 105014.9375
[ Info: RAM left : 100209.421875
[ Info: RAM left : 96386.6953125
[ Info: RAM left : 92556.125
[ Info: RAM left : 88755.32421875
[ Info: RAM left : 84891.40625
[ Info: RAM left : 82037.58984375
[ Info: RAM left : 77238.71875
[ Info: RAM left : 74415.28515625
[ Info: RAM left : 69477.1484375
[ Info: RAM left : 65713.12109375
[ Info: RAM left : 62047.74609375
[ Info: RAM left : 58244.33203125
[ Info: RAM left : 53576.16015625
[ Info: RAM left : 50848.9375

Running the GC after the loop can free all the ram.
Running the loop without the dataloader frees at every iteration

[CarloLucibello]

cannot reproduce. Also, it is unlikely this is an issue with the dataloader, since it is not even iterated over in this example. The dataloader's constructor just creates plain julia object, so if there is any issue here is with how julia gc behaves.