Memory leak of 30Gb on Linux platform

[juliohm]

MWE:

using NearestNeighbors
using StaticArrays
using Random

x(i) = float.(Tuple(CartesianIndices((100,100,100))[i]))

function mwe()
  Random.seed!(123)

  sources = rand(1:100*100*100, 3000) |> unique

  others = setdiff(1:100*100*100, sources)

  batches = Iterators.partition(others, 10^3)

  kdtree = KDTree([SVector(x(i)...) for i in sources])

  dists = []
  for batch in batches
    xs = [SVector(x(i)...) for i in batch]
    _, ds = knn(kdtree, xs, length(sources), true)
    append!(dists, ds)
  end
end

mwe()

If you run it on Julia >= v1.10 on Linux, it leaks 30Gb of memory. Posterior calls to Base.GC.gc() don't release memory.

[b8a86587] NearestNeighbors v0.4.21
[90137ffa] StaticArrays v1.9.13

Julia Version 1.11.5
Commit 760b2e5b739 (2025-04-14 06:53 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 8 × Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, skylake)
Threads: 8 default, 0 interactive, 4 GC (on 8 virtual cores)
Environment:
  JULIA_EDITOR = code
  JULIA_NUM_THREADS = 8

Moved from KristofferC/NearestNeighbors.jl#208

[d-netto]

Posterior calls to Base.GC.gc() don't release memory.

Just to clarify: if you call GC.gc() after mwe() you still see large memory consumption?

[juliohm]

Yes. That is what I meant. The memory is not released.

[d-netto]

gc_live_bytes seems fine on MacOS (I'm aware that you mentioned you're running on Linux; also, I changed one-line of your MWE to run it a bit faster):

using NearestNeighbors
using StaticArrays
using Random

x(i) = float.(Tuple(CartesianIndices((100,100,100))[i]))

function mwe()
  Random.seed!(123)

  sources = rand(1:100*100*100, 3000) |> unique

  others = setdiff(1:100*100*10, sources)

  batches = Iterators.partition(others, 10^3)

  kdtree = KDTree([SVector(x(i)...) for i in sources])

  dists = []
  for batch in batches
    xs = [SVector(x(i)...) for i in batch]
    _, ds = knn(kdtree, xs, length(sources), true)
    append!(dists, ds)
  end
end

function print_memory_metrics(str::String)
    println(str)
    @show Base.gc_live_bytes()
end

mwe()
print_memory_metrics("Before GC")
GC.gc()
print_memory_metrics("After GC")

Before GC
Base.gc_live_bytes() = 2801882019
After GC
Base.gc_live_bytes() = 21947739

Julia Version 1.10.9
Commit 5595d20a28 (2025-03-10 12:51 UTC)
Build Info:

    Note: This is an unofficial build, please report bugs to the project
    responsible for this build and not to the Julia project unless you can
    reproduce the issue using official builds available at https://julialang.org/downloads

Platform Info:
  OS: macOS (arm64-apple-darwin23.4.0)
  CPU: 12 × Apple M2 Max
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, apple-m1)
Threads: 8 default, 0 interactive, 4 GC (on 8 virtual cores)

[d-netto]

also, I changed one-line of your MWE to run it a bit faster

gc_live_bytes also seems fine in the full version (others = setdiff(1:100*100*100, sources)) running on MacOS and 1.10:

Before GC
Base.gc_live_bytes() = 24212757715
After GC
Base.gc_live_bytes() = 165570362

Note how gc_live_bytes dropped from 24GB to 170MB.

[juliohm]

This issue is specific to Linux. That is the only platform where it happens.

[d-netto]

I wonder if this could be related to #56759.

CC: @vtjnash, @gbaraldi, @oscardssmith.

[oscardssmith]

Does it reproduce on nightly? If so, it's different.

[giordano]

Is this #42566? Rerunning mwe() multiple times doesn't increase memory, and when exiting Julia memory is released. @oscardssmith had a somewhat similar example there, same packages, even same function name. Edit: it was just the same.

[oscardssmith]

@gbaraldi time to switch to Mimalloc?